The existing institutional and data infrastructure in Belgium and Greece provides a strong basis for developing tailored AI-powered solutions that can better support MIS beneficiaries on their journey towards employment. This chapter proposes a concept for a comprehensive job matching and career recommendations platform targeting the needs of this specific target group and beyond in the two countries. The underlying functionalities of such a platform are structured around five main pillars: 1) improving the quality and personalisation of job matching; 2) enhancing vacancy management; 3) identifying jobseekers’ skills, needs and barriers; 4) getting access to labour market insights; and 5) supporting career development and activation pathways. The chapter also proposes strategic options for deploying such solutions in each context and discusses governance considerations and potential users.
AI and Digitalisation for Employment Support in Belgium and Greece
Abstract
Despite the fact that digital tools to support labour market integration exist in both Belgium and Greece, neither currently has solutions specifically tailored to the needs of MIS beneficiaries and other disadvantaged jobseekers. This gap is particularly relevant for job matching and career recommendations purposes, as these individuals often require more tailored support to identify suitable opportunities, understand how their skills relate to labour market demand, navigate job postings that are not always written with their profiles in mind and move into better and more sustainable opportunities. At the same time, employers may struggle to reach or identify candidates from vulnerable groups even when suitable applicants exist, leading to missed matches on both sides.
This chapter proposes concepts for digital solutions enhanced by AI, to support job matching and recommend career pathways for this specific target group and beyond in the two countries. Rather than developing multiple standalone tools or parallel initiatives, the proposed approach recommends designing and implementing a single digital platform in which the core functionalities are integrated. This will allow users to access all relevant services through a consistent interface, improving usability, efficiency, and data coherence. This approach also avoids the fragmentation often created by a patchwork of digital unintegrated initiatives, ensuring interoperability, scalability, and sustainability. Nevertheless, if developing such an ideal platform is not feasible, the chapter proposes a second-best option of developing a career recommendation tool as a standalone instrument.1 The present concepts outline a set of potential functionalities, structured around five main pillars. While these functionalities intend to have a wider application, which can be relevant regardless of governance models and country-specific institutions, features and considerations that are specific to each country are also being highlighted.
The chapter is structured as follows: Section 5.2 outlines the key objectives and benefits that such tools can generate. The next section discusses how these tools can be integrated to enhance the customer journey for MIS beneficiaries in the two countries. Sections 5.4, 5.5 and 5.6 describe the proposed concepts and user groups, and outline all the various underlying functionalities, strategic options, and governance considerations for each country.2 The further development of career recommendation functionalities as a distinct tool is also presented as an alternative policy option, should deeper integration with the existing job matching tool and other relevant digital services not be technically or operationally feasible (Section 5.7).
The overarching objective of a comprehensive job matching platform is to enhance labour market mediation by matching jobseekers’ profiles with relevant job vacancies, assisting both jobseekers and employers.3 Digital career recommendation tools can complement a job matching platform by providing guidance on skills development, career progression opportunities, and longer-term employment prospects. Such solutions can be particularly valuable for MIS beneficiaries and other vulnerable groups, such as long-term unemployed individuals or people facing multiple barriers to employment who often experience repeated unemployment spells and are more likely to cycle between short-term, low-quality jobs with limited prospects for advancement.
The introduction of such solutions can potentially generate substantial benefits not only for the end users themselves but also for the institutions involved in supporting them, such as PES and social services. In this respect, the experiences of PES across EU and OECD countries that already employ digital and AI‑powered tools for labour market mediation (e.g. France, Austria, Finland, Korea, the Netherlands, Sweden, etc.) can provide useful insights that can inform the development of similar solutions in Greece and further improvement of existing tools like in the case of the VDAB in Belgium (Brioscú et al., 2024[1]; Dromundo Mokrani, Lauringson and Xenogiani, 2024[2]).4 The most important benefits include:
Greater efficiency and more effective service delivery. AI can speed up job searches, automate parts of the process such as skills comparisons and career pathway suggestions, and reduce administrative burdens. In turn, this can lower operational costs, enable more efficient allocation of staff and resources, and allow counsellors to focus more on personalised support and complex cases.
More accurate and personalised matching. AI can analyse jobseekers’ skills, preferences and circumstances alongside employers’ needs, enabling more accurate matches and more tailored recommendations. This can help contribute to improved employment outcomes and more effective activation, while it can help guide individuals towards sustainable career pathways, training opportunities, and occupations with stronger prospects. These benefits can be particularly important for MIS beneficiaries and other disadvantaged groups, for whom more personalised guidance can support sustainable labour market integration.
Improved user experience and accessibility. By offering timely, relevant, and user-centred recommendations, AI can support users in making informed choices, increase visibility of opportunities and reduce the burden of navigating fragmented sources of information. In turn, this can make employment and social services more accessible, attractive, and better meet the expectations and up-to-date needs of the users.
Better ALMP targeting. By identifying skills gaps, employment barriers, and labour market opportunities more accurately, PES can direct jobseekers toward training programmes and activation measures that are more likely to improve employability and employment outcomes, thereby increasing the effectiveness of public spending.
Stronger monitoring and policy insights: The data generated by these tools can help PES monitor jobseeker trajectories, identify common barriers to employment, and detect emerging skills shortages. The information generated can then support evidence‑based policy design and service improvements.
Figure 5.1 presents an indicative results chain for a comprehensive AI-powered job matching and career recommendations platform. It outlines the main inputs, activities, outputs, outcomes, and long-term impacts that the tool could generate, particularly in relation to improving labour market integration for MIS beneficiaries and other vulnerable groups. It also proposes some example indicators that could be used for monitoring and evaluation purposes.
Indicative results chain for an AI-powered job matching and career recommendations platform, with a focus on MIS beneficiaries and other vulnerable groups. It includes inputs, activities, outputs, outcomes, long-term impact, as well as a set of indicative efficiency indicators appropriate for monitoring and evaluation
Note: The elements depicted in the results chain figure are not exhaustive and do not capture every operational detail, function or contextual factor that may influence the platform’s development and implementation.
Source: Authors’ elaboration.
Given the level of digital development in Greece and Belgium and the clearly identified needs, introducing matching and career recommendation platforms to support MIS beneficiaries and other jobseekers with vulnerabilities is relevant and feasible in both contexts. The platform doesn’t aim to replace existing services or alter the eligibility criteria for these services. Rather, its core objective is to streamline and enhance the effectiveness of each step in the process by providing data-driven insights and personalised guidance, and particularly in the Belgian context to expand access to opportunities beyond regional boundaries.
The present section discusses how the introduction of a comprehensive job matching and career recommendation platform (conceived as a single digital solution that integrates key functionalities) can be embedded within and facilitate the standard journey of MIS beneficiaries. It is structured into two sections; the first one focusses on Greece and the second one on Belgium.
5.3.1. In Greece, the job matching and career recommendation tool can be embedded within the journey of MIS beneficiaries towards self-sufficiency
Figure 5.2. depicts the standard journey of MIS beneficiaries in Greece from a high-level perspective. Based on its multiple functions, the job matching and career progression platform can be used in several points within this journey to facilitate the social inclusion and activation of MIS beneficiaries and other jobseekers (yellow boxes). It can also support other key end-users such as job counsellors, social workers and employers.
Source: Authors’ elaboration.
The tool can support the early detection of potential barriers and function as a centralised hub for accessing social services
After eligibility to enter the MIS scheme is verified and under the second pillar of the MIS scheme in Greece, beneficiaries are entitled to social inclusion services (for a description of services, see Chapter 2).
At this stage, the tool can help identify potential barriers such as housing instability, caregiving responsibilities, mental health challenges, or lack of childcare. Based on this information, it can guide MIS beneficiaries themselves as well as social workers at the Community Centres to direct individuals towards appropriate services, community support, or specialised assistance to address barriers that affect their ability to work. By directing users to the right support early on, the system helps ensure that underlying issues are addressed before individuals engage fully in employment-related activities. The platform could also detect early signals of vulnerability or complex needs by combining information on beneficiaries’ social circumstances and service use. This would allow social workers to prioritise individuals who may require more intensive support and ensure that appropriate interventions are provided in a timely manner.
In addition to identifying barriers, the platform can act as a centralised hub for social services available within the beneficiary’s region (but also beyond), allowing information on available services to be gathered in one place and made readily accessible to all relevant actors. This can improve visibility of available support, reduce awareness gaps and facilitate faster referrals. The platform could therefore be used both by beneficiaries themselves in case they are able to use the tool independently, and by social workers and job counsellors who support them in navigating available services and progressing towards greater self-sufficiency. The tool could also support case management by allowing social workers to record identified barriers, track referrals to services, and monitor whether beneficiaries have accessed the support recommended to them. This can help ensure that individuals do not fall through the cracks and that follow-up actions are indeed taken when needed.
Such functionalities can prove particularly valuable for supporting the mobile units of the Community Centres that serve residents in remote mountainous or island municipalities, where access to up-to-date information on available services may otherwise be limited. In these cases, the platform could provide a shared digital interface through which staff can quickly identify relevant support options and co‑ordinate assistance even when services are geographically dispersed.
The tool can support MIS beneficiaries throughout their employment journey, extending beyond initial job placement
Naturally, the matching and career recommendation tool plays a central role in the activation process. Following registration with DYPA, the tool can be used throughout the journey towards employment. In an ideal world, the profiling tool deployed currently by DYPA could be either part of this platform or be interoperable with it. This way the information about jobseekers’ work history, education and skills would feed the platform and therefore the results that it will provide would be more tailored to the individual’s capabilities and circumstances. The diversity in the profiles of MIS beneficiaries and the complexity of their needs make the integration of such a platform at this stage particularly useful, as it helps ensure that social and employment support are addressed in a co‑ordinated and complementary manner.
When jobseekers meet with DYPA’s job counsellors to develop the Digital Individual Action Plan, the tool can provide additional insights that support decision making and inform referrals. Based on the user’s profile and labour market conditions the tool can help match the jobseeker with available vacancies. It can also suggest possible occupations that align with the person’s skills, identify alternative career paths, and highlight sectors where demand for labour is increasing. It can identify skill gaps and recommend training, upskilling and reskilling programmes and other ALMPs like employment incentives, vocational training programmes, work experience programmes and public works that may help individuals move towards more stable or better-quality employment and also services such as career guidance and targeted seminars. The system can also provide information about labour market trends, such as occupations that are difficult to fill, emerging roles, or sectors experiencing growth. These insights can help both counsellors and jobseekers focus their efforts on opportunities where the likelihood of employment is higher. When relevant, it can also direct individuals back to social or health services if additional support is needed. As above, the tool can be either used independently by the jobseeker or in co‑operation with the job counsellor.
One challenge that arises in the activation phase of the customer journey concerns the fragmentation of employment support services. Currently, in addition to DYPA, some municipalities operate integrated labour offices that provide job search assistance to their citizens. While these initiatives can increase the availability of support at the local level, they may also lead to duplication of efforts or differences in the type and quality of support available depending on the municipality. At the same time, Community Centres, municipal labour offices and KPAs often rely on separate information systems with only partial interoperability. As a result, relevant information about MIS beneficiaries, such as their employment history, skills profile, participation in programmes or previous referrals, may not always be consistently shared across institutions. This fragmentation can affect the labour market prospects of beneficiaries and complicate service delivery.
A matching and career recommendation platform could help address this challenge by acting as a shared digital infrastructure used by all relevant actors involved in the activation process. Through a common interface, Community Centres, municipal labour offices and KPAs could access and update key information on beneficiaries’ profiles, action plans, and referrals, thereby improving co‑ordination among the different agencies. For example, when a MIS beneficiary is referred to a job vacancy, a training programme or a social service, the information could be registered in the system and become visible to other authorised actors ideally in real time. In practice, this would allow counsellors and social workers to see which interventions have already been proposed or implemented, reducing the risk that individuals are repeatedly referred to similar programmes or asked to provide the same information multiple times.
In addition, the platform could support the development of shared or co‑ordinated action plans between institutions involved in the activation process. Social workers at Community Centres could record information on social barriers affecting employability, while PES counsellors could update information related to employment support measures, job search activities or participation in training programmes. By bringing together these different elements in one place, the platform could provide a more comprehensive picture of the beneficiary’s situation and progress towards employment.
From the perspective of jobseekers, such integration would simplify the system and reduce the administrative burden associated with navigating multiple institutions. MIS beneficiaries would not need to repeatedly provide the same information to different institutions, and the support they receive would be more consistent regardless of where they initially engage with the system.
Another option is rethinking and clarifying even further the different agencies’ roles and responsibilities. The introduction of such a platform could serve as an opportunity for this. While municipalities and Community Centres focus on addressing social barriers and providing complementary support services, KPAs are the sole actors charged with the primary responsibility for labour market activation and employment-related services, such as job matching, training referrals and employer engagement. The platform would therefore act as a co‑ordination tool, allowing each actor to contribute within their mandate while ensuring that the overall support pathway for beneficiaries remains coherent.
Another important consideration concerns the moment or timing at which such a tool should be used within the activation process. Rather than being limited to a single stage, the matching and career recommendation platform could support beneficiaries throughout their entire employment journey. Its use could begin at the moment of registration with the PES, where it would assist in profiling jobseekers and identifying potential employment opportunities, training needs, and possible barriers to employment. During counselling sessions, the platform could support job counsellors in developing action plans by providing information on suitable occupations, skills requirements, training options, and labour market trends. The tool could continue to play a role even after referrals to jobs, training programmes, or other services have been made. Beneficiaries could use it to obtain additional information about recommended opportunities, explore alternative career pathways, identify relevant education or training programmes, or search for better employment opportunities as their situation evolves. In this way, the platform would not only support entry into employment but could also assist individuals in improving their labour market situation over time.
Finally, the platform would not be limited to jobseekers alone. As mentioned earlier, the platform could be used in different ways depending on the needs and capacities of users. Some beneficiaries may use the tool independently to explore opportunities, obtain guidance, or monitor their progress. Others may rely on support from job counsellors or social workers, who could use the platform during counselling sessions to guide discussions and identify appropriate next steps. The tool could therefore function both as a self-service digital resource and as a decision-support instrument for the staff working with MIS beneficiaries.
Such a platform should continue to add significant value beyond initial job placement by supporting individuals throughout their employment journey. It should remain accessible to individuals even after they have entered employment, allowing them to explore opportunities for career progression, identify training that could improve their skills, or consider alternative employment pathways. This continuous support is particularly relevant for MIS beneficiaries, many of whom face multiple and evolving barriers to employment. Their pathways into the labour market are often gradual and may involve periods of training, temporary employment, transitions between different types of work and even inactivity. The tool can provide timely guidance when circumstances change and direct them to (additional) training, re‑matching, more suitable roles, or guidance towards better career pathways. In doing so, the platform shifts the focus from short-term placement to long-term labour market integration, ensuring that individuals not only access employment, but also progress towards more stable, rewarding, and sustainable careers and therefore greater economic independence.
5.3.2. In Belgium, the job matching and career recommendation tool can lead to more tailored and effective pathways to employment
Similarly to the case of Greece, Figure 5.3. depicts how the standard journey of MIS beneficiaries in Belgium can be facilitated due to the integration of the job matching and career progression platform (yellow boxes).
Source: Authors’ elaboration.
The tool can help identify barriers upfront and gather available services in one hub
Once enrolled in the minimum income scheme, beneficiaries are entitled to a wide range of social services (for a description of services, see Chapter 2). At this stage, the matching and career recommendation platform can help identify potential barriers of the beneficiaries such as housing instability, caregiving responsibilities, mental health challenges, or lack of childcare. Based on this information, it can help social workers direct individuals towards appropriate services ensuring that underlying issues are addressed before or in parallel with employment-related activities. In addition to identifying barriers, the platform can act as a centralised hub for social services that are available within the beneficiary’s region and beyond. The information about the available services would be gathered in this hub, making it easily accessible to all relevant stakeholders. Both the identification of barriers and the map of available services would be particularly valuable, as information on beneficiaries’ needs and circumstances (typically collected by the CPAS/OCMW) is often rich, yet not systematically captured or connected across municipalities or with employment services.
The platform could be used both by beneficiaries themselves in case they are able to use the tool independently, and by social workers. The tool could also support case management by allowing social workers to record identified barriers, track referrals to services, and monitor whether beneficiaries have accessed the support recommended to them. This can help ensure that individuals do not fall through the cracks and that follow-up actions are taken as faster as possible and as needed.
The tool can support MIS beneficiaries from job search through post-placement by surfacing opportunities beyond local PES jurisdictions
In Belgium, ALMPs are delivered by the competent regional employment service (Le Forem, Actiris, VDAB, ADG). In addition to social services and ALMPs, MIS beneficiaries in Belgium could be guided towards other education and training services such as language courses, vocational training, or second-chance education.
The matching and career recommendation tool should be used throughout the MIS beneficiaries’ journey towards employment. In the first place, based on the user’s profile and labour market conditions the tool can help match the jobseeker with available vacancies. To some extent, this is already possible especially in the PES with more mature advance digital systems which include job matching tools. The key difference here is that the matching and career recommendation tool envisioned under this project should be capable of suggesting and displaying opportunities from all different regions and not only those within the individual’s PES jurisdiction.
It can also suggest possible occupations that align with a person’s skills, identify alternative career paths, and highlight sectors where demand for labour is increasing. Following the identification of gaps and taking into account the individual’s barriers and constrains, the tool can recommend training, upskilling and reskilling programmes that are directly aligned with labour market demand. Instead of assigning generic training, the tool can recommend targeted pathways based on real job opportunities, showing how specific training leads to concrete employment options. It can also suggest other ALMPs like employment incentives and work experience programmes that may help individuals move towards more stable or better-quality employment and also services such as career guidance and targeted seminars. The system can also provide information about labour market trends, such as occupations that are difficult to fill, emerging roles, or sectors experiencing growth. These insights would help counsellors, social workers and jobseekers focus their efforts on opportunities where the likelihood of employment is higher. When relevant, it can also direct individuals back to social or health services as deemed necessary.
In the case of Belgium, the tool can also include the schemes available under Article 60 and Article 61. This is particularly important as currently this information is not consistently kept and interoperable across municipalities. In this very specific case, the platform can ensure that this pathway is used in a strategic rather than systematic manner. Before entry, the tool could help assess whether the individual could already transition directly into the labour market. During the pathway, it can track skills development and continuously recommend next steps, ensuring that this experience accumulates into sustainable employment. As the Article 60/61 contract approaches to an end, the platform becomes especially valuable by identifying suitable job opportunities or ALMPs and facilitating transition, thereby reducing the risk of returning to unemployment.
The platform can continue to add value after initial job entry by tracking retention, progression, and job quality. It can detect early risks such as for example those related to instability and suggest corrective actions including additional training or better job opportunities. This ensures that the system focusses not only on placement, but on sustainable employment and career development.
This section proposes a concept for new digital solutions enriched by AI elements to enhance the capacity of Greece and Belgium to support MIS beneficiaries and other vulnerable jobseekers in their journey towards employment and career progression. The proposed concept outlines a set of potential functionalities that those solutions could ultimately embrace, along with the key steps that the two countries would need to take to implement these features effectively. The technical and functional specifications of such solutions fall beyond the scope of this project and this specific report.
While the tool’s long-term vision is to support a broad range of users, the development process can begin small. This includes a limited set of core functionalities that appear most important based on the data analysis and the country context. These initial features would form the foundation of the tool and could be gradually expanded over time, as additional resources become available and as lessons emerge from implementation and user experience. This phased approach would allow the tool to start delivering value early, while enabling continuous improvement and enrichment.
This section is organised as follows: It first introduces the core pillars around which the platform can be developed as well as the relevant functionalities under each pillar and how these can operate in an interconnected manner. Next, the section discusses how the new AI matching solution should interact with and support a range of users. Section 5.5 outlines strategic options for deploying the AI-powered matching platform in each country, highlighting the key elements that differentiate these options. Section 5.6 explores the governance structure of the platform in each country – who could own, manage, administer, and access the system –, as well as how these roles could be distributed across the relevant institutions. The primary or ideal scenario relies on the assumption that career recommendations functionalities are developed as part of a comprehensive matching platform. However, if for any reason the ultimate decision is not to develop such a comprehensive platform, an alternative option would be to develop the career progression tool as a standalone instrument. Section 5.7 examines this alternative scenario and discusses specific considerations for each country.
5.4.1. An AI-based matching tool for MIS beneficiaries can be developed around five core pillars with each pillar representing a bundle of related functionalities
The proposed concept for the new AI-based job matching tool outlines functionalities that span the entire job matching process and could be structured around five pillars: improving the quality and personalisation of matching, enhancing vacancy management, identifying jobseekers’ skills, needs and barriers, getting access to labour market insights and career pathways and progression options (Figure 5.4).
The proposed concept for a comprehensive AI matching platform outlines different functionalities that could be structured around five pillars
Note: Career progression related functionalities including suggesting pathways that extend beyond immediate job placement can be developed either as an additional pillar of the platform or as a separate, complementary AI solution. In the former case, the platform would serve as a comprehensive, AI-powered system for job matching and career progression.
Source: Authors’ elaboration.
The functionalities discussed in this section are intentionally described in a general way, accompanied by practical examples. Such capabilities can be implemented in both Greece and Belgium, regardless of their respective differences in institutional structures or data systems. Examples are also provided to illustrate how key findings from the data analysis can be translated into concrete platform functionalities in the two countries.5
Pillar 1: AI algorithms can improve the matching quality between MIS beneficiaries and employers
Improving matching quality is at the core of a job matching platform because it orchestrates the fundamental task of connecting jobseekers with suitable employment opportunities and employers finding candidates who meet their needs. Its success – whether translated into improved labour market outcomes, reduced frictions or enhanced access to opportunities – can generate substantial benefits for the users, especially for vulnerable groups who often experience weak connections to the labour market or face additional barriers in navigating job search processes.
The matching process is complex as it relies on a wide set of underlying and often interconnected functionalities that prepare, structure, and mediate the exchange between the different stakeholders interacting with the platform. It involves collecting, organising and combining information from multiple sources and interpreting the results with the aim to create meaningful links between the different actors. Furthermore, the matching process is a dynamic process. In other words, it is a process that evolves over time as user profiles are updated, vacancies change, or labour market conditions shift. The underlying components of matching create the foundation upon which AI-powered techniques can be introduced to enhance precision, inclusiveness, and transform the overall user experience.
AI can streamline user registration, onboarding, and profile creation
A key feature of AI-powered matching solutions lies in their ability to create dynamic and adaptive user profiles. AI can streamline registration and onboarding by automatically extracting skills from CVs and certificates, pre‑filling fields, and guiding users with prompts that help them describe their experience. To do so it can pull information from existing systems such as MIS files, PES profiles, employment and vacancy records and previous interactions with social and employment services (see also the pillar on skills and needs identification). A platform’s matching quality is only as strong as the profile data it collects.
Profiles can be adapted for different user types – jobseekers, employers, job counsellors, social workers etc. – ensuring that each of them sees the information most relevant to their role and receives relevant and accessible instructions. For MIS beneficiaries and vulnerable groups specifically, the system can incorporate critical constraints such as health conditions,6 care responsibilities, mobility limitations, or the need for flexible arrangements. These factors can be integrated directly into the matching algorithm to generate realistic and accessible recommendations, avoiding making job suggestions that are either unsuitable or incompatible with the users’ individual circumstances. This information provides the basis for more accurate and realistic matching, reflecting the specific needs and characteristics of MIS beneficiaries in Greece and Belgium.
AI can make search more intuitive by understanding users and suggesting relevant options beyond simple keyword matches
Matching requires tools that allow jobseekers and employers to manually explore options in addition to receiving automated suggestions. Traditional search tools rely on user-entered keywords and static filters, often producing inconsistent or incomplete results. AI improves both search and matching by understanding the meaning behind queries and incorporating users’ preferences, constraints and circumstances directly into the recommendation process.
Rather than relying solely on exact keyword matches, AI can recognise synonymous job titles, related occupations and transferable skills, which is particularly useful for jobseekers unfamiliar with industry and occupational terminology. This can encourage jobseekers to consider occupations that offer a good fit with their skills but may not have been identified through conventional search methods. Such functionality is particularly relevant for MIS beneficiaries, who may otherwise focus on a narrow set of familiar occupations and become confined to unstable or low-quality employment. By identifying adjacent occupations and less obvious opportunities, AI can help widen career horizons and improve employment prospects. Finland’s Job Market Finland platform, for example, combines structured criteria matching with natural language processing to generate compatibility scores that capture both direct and indirect matches (TEM, 2023[3]). This helps jobseekers to better articulate their strengths and may be particularly helpful for those with more complex barriers. It also enables employers to list vacancies using the same skills taxonomy as for jobseekers, creating a more harmonious link between applicants and vacancies (Box 5.1).
Finland has undertaken a comprehensive reform of its digital employment services through Job Market Finland and Expert’s Job Market Finland, which together form the core customer interface of the new customer information system for Finnish PES. The platform acts as a single meeting place for jobseekers, employers, service providers and PES counsellors. It integrates vacancies, training opportunities, e‑services and labour-market information. It is provided free of charge as a publicly funded service, with a parallel private workspace (Expert’s Job Market) enabling PES counsellors to use the same data and matching logic in case management and employer engagement.
At the technical level, Job Market Finland relies on competence‑based matching powered by natural language processing (NLP) rather than solely on job titles or qualifications. Jobseekers and employers describe skills and requirements in free text, which are then processed using a combination of structured data (including ESCO occupational and skills classifications) and NLP techniques such as TF-IDF and FastText. These methods extract relevant skill signals, cluster related concepts and calculate suitability scores that rank job vacancies for jobseekers and candidate profiles for employers. The same matching logic is embedded both in the public user interface and within the PES customer information system, ensuring consistency between self-service digital matching and counsellor-supported placements. The application gives service providers better information about how they can further develop their services and find new business opportunities.
The implementation of Job Market Finland illustrates several lessons relevant to other PES. Replacing a long-standing legacy system required phased deployment, large‑scale data migration and a modular cloud-based architecture built on microservices, combined with agile delivery frameworks. From a policy perspective, the Finnish approach highlights the benefits of skills-based matching for improving transparency and inclusiveness, particularly for jobseekers whose competencies are not easily captured by formal qualifications. At the same time, it underscores the need for robust data governance, continuous model monitoring and close integration between digital tools and frontline services to ensure that AI-supported matching enhances, rather than substitutes for, professional judgement and personalised support.
Source: Vainio (2025[4]), Presentation during International Workshop 2: Digital and AI-powered job matching and career recommendation tools conducted on 5 November 2025, as part of this project, https://www.oecd.org/content/dam/oecd/en/about/programmes/dg-reform/using-ai-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries/Job-Market-Finland-and-Experts-Job-Market-Finland.pdf.
At the same time, it can account for constraints and personal circumstances that affect employability, such as mobility limitations, care responsibilities, preferred working arrangements, or availability. This allows search filters and recommendations to adapt dynamically to user preferences such as commuting distance, working hours or flexibility needs. For example, given that most MIS beneficiaries in Greece are women, the platform could prioritise jobs with shorter commuting distances, part-time or flexible working hours. For users facing mobility constraints, it could highlight remote work options, while for those lacking specific skills it could identify relevant training opportunities. Employers can benefit as well, as AI helps them discover suitable candidates even when experience is described in non-standard ways.
AI can facilitate communication and keep users engaged
The matching process also includes mechanisms that enable direct interaction between jobseekers and employers, such as application submissions, messaging, interview scheduling, and progress tracking. These mechanisms help ensure that matches translate into real engagement and not simply a list of potential opportunities. Communication between jobseekers and employers is critical for turning matches into actual employment. AI can facilitate this process by automating reminders, generating suggested follow-up actions and notifying users of new opportunities. Employers can receive support in grouping applicants, drafting initial messages and scheduling interviews.
Conversational AI, such as chatbots or digital assistants, can also offer tailored search tips and complementary guidance to users through applications or platform navigation, reducing dropout and increasing engagement. These tools can also address practical questions in real time, such as “How do I write a cover letter?” or “Do I need a certification for this job?”, helping users overcome uncertainty and take the next step. These automated interactions are especially valuable for vulnerable users who struggle with digital tasks, need additional guidance or be less likely to initiate communication independently. A relevant example is Berufsinfomat, a generative AI chatbot launched by the Austrian PES in 2024. The tool answers questions on occupations, training, and salaries in plain language, and it is accessible in multiple languages. While not a full matching engine, it provides jobseekers with rapid, tailored guidance based on their specific queries (Box 5.2).7
Austria’s Public Employment Service (AMS) has developed Berufsinfomat, an AI-powered digital “career concierge” designed to provide accessible, personalised labour market and occupational information to a wide audience, including vulnerable jobseekers. The tool allows users to ask free‑text questions (for example, about career pathways, training requirements or emerging occupations) without registration or the provision of personal data, and offers multilingual support across around 90 languages. This low-barrier design aims to improve access for migrants, young people and individuals with limited familiarity with formal labour market institutions, while complying with accessibility standards (WCAG 2.2) and privacy-by-design principles.
Technically, Berufsinfomat is built as a retrieval-augmented generation (RAG) system that combines large language models with a curated and verified content database owned and maintained by AMS. User queries are embedded and matched against structured and unstructured labour market content using vector databases and ranking models (including BERT-based and GPT-based components), before responses are generated via an Azure‑hosted large language model. This architecture is intended to reduce hallucinations and ensure that answers are grounded in validated public information, with explicit source referencing and ongoing content governance distinguishing the tool from general-purpose AI chatbots.
The implementation of Berufsinfomat highlighted key challenges relevant to other public employment services. Early public scrutiny focussed on risks of gender bias in occupational suggestions, technical robustness and perceived costs. AMS responded by introducing multi-layer bias detection and mitigation (covering inputs, content and outputs), regular audits, manual review of sensitive content, and stronger governance over training data and prompts. The experience illustrates the trade‑offs faced by public authorities between innovation, transparency, fairness and operational complexity, and underscores the importance of continuous monitoring, clear accountability and institutional capacity when deploying generative AI tools in labour market policy contexts.
Source: Krishnan Somos (2025[5]) International Workshop 2: Digital and AI-powered job matching and career recommendation tools, https://www.oecd.org/content/dam/oecd/en/about/programmes/dg-reform/using-ai-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries/Concept-Note-and-Agenda-International-workshop-Digital-and-AI-powered-job-matching.pdf
AI can streamline screening and shortlisting candidates
Screening and shortlisting candidates are among the most time‑consuming and resource‑intensive stages of the recruitment process. Employers often receive large volumes of applications, many of which do not meet the basic requirements of the role. As a result, valuable time is spent manually reviewing CVs, identifying suitable profiles, and narrowing down candidate lists. These challenges are even more acute for small employers who may not have dedicated human resource departments or experts or sectors with high turnover, where recruitment capacity is limited. PES usually mediate by screening and shortlisting candidates on behalf of employers, as part of their engagement and support to them, a process that also requires resources and time.
AI can significantly streamline this process by automatically analysing CVs, application forms, and candidate profiles to filter applicants based on predefined criteria, core skills, or required qualifications. It can also identify transferable skills that may not be immediately evident, enabling employers to consider a broader pool of suitable candidates. Advanced models can rank applicants by degree of fit, highlight strengths and potential gaps, and generate concise summaries that help recruiters focus their attention on the most promising profiles.
Earlier matching technologies often required exact word-for-word matches between terms in a vacancy and those in a CV. Today’s AI functionalities can interpret meaning rather than relying on identical wording. For example, when a term like “accountant” appears, AI can automatically recognise related job titles and synonyms – such as “account specialist” – and link them to the candidate’s profile, expanding possibilities for more accurate and inclusive matching (Broecke, 2023[6]).
AI can also support screening by engaging directly with candidates through automated conversations or assessments. These tools can analyse a candidate’s responses and behaviours to help determine who is best suited to move forward in the recruitment process. Some systems use interactive tasks or gamified assessments to identify patterns linked to cognitive, social, or emotional traits associated with successful workers. In addition, AI can assist with background checks by reviewing publicly available online information and flagging content that may be relevant to employers. Such practices raise important ethical and privacy considerations that shouldn’t be ignored or perceived as purely beneficial. Nevertheless, they reflect the expanding range of AI applications designed to streamline early stages of recruitment and may provide richer insights into candidate suitability.
Screening should ideally occur within the first few weeks from the registration with the PES, which is particularly important for vulnerable groups, as early identification allows counsellors to prioritise individuals facing the greatest labour market difficulties and ensure they receive timely and tailored support.
ΑI continuously learns from outcomes to improve matching accuracy and detect inequities
Finally, matching requires ongoing refinement based on user behaviour, hiring outcomes, employer feedback, and labour market trends. This ensures that the system remains relevant, accurate, and responsive to changes in supply and demand.
Traditional systems rely on manual updates or periodic reviews to adjust matching rules. Machine‑learning models continuously learn from user behaviour and outcomes and update recommendations based on which matches lead to applications, interviews or hires. For example, click data from the jobseeker’s job browsing patterns can be used to recommend similar vacancies to those viewed by the jobseeker, thus accounting for the preferences of the jobseeker (e.g. as implemented in the job matching tool in the Korean PES). AI can identify patterns in employer demand, jobseeker engagement and vacancy fulfilment, ensuring the system evolves with labour market changes. It can also flag when certain groups – such as older workers, migrants or MIS beneficiaries – are under-represented in successful matches, triggering fairness checks and adjustment of algorithms.
Pillar 2: AI can help improve the quality and consistency of vacancy information
Vacancy management is a key component of the matching process. It refers to the tasks and processes that can improve the quality, structure, organisation and usability of vacancy postings with the ultimate objective of enabling more effective matching. AI-powered vacancy management can generate a dual benefit by reducing information gaps for both jobseekers and employers. On the one hand, for MIS beneficiaries and other vulnerable individuals, AI can expand access to a broader range of vacancies and highlight opportunities that match their profiles. It can also provide clearer information about each position, such as required skills or conditions and make it available in a timely and accessible manner. On the other hand, AI can support employers in improving the description of their vacancy postings. Postings of better quality are more visible and clearer to jobseekers, helping employers connect with suitable candidates whose profiles they might otherwise have overlooked.
Vacancies should be sourced from multiple systems, consolidated into one consistent and standard format
To expand opportunities, especially for vulnerable groups, the matching platform must ensure high vacancy coverage. Given that vacancy information is usually fragmented, achieving such coverage generally entails integrating data from several systems to minimise gaps and improve usefulness for both jobseekers and employers. Rather than depending only on vacancies posted directly on the platform, AI can gather and analyse job opportunities from a wide range of sources, including employer submissions, external job boards, open labour market data, and professional networks such as LinkedIn. This can significantly broaden the range of available vacancies, giving jobseekers – especially the vulnerable ones – access to opportunities they might not otherwise encounter.
However, relying on multiple systems introduces several challenges: inconsistent formats, outdated vacancies that are still shown as active, duplicate listings submitted multiple times or through different channels, and even inaccurate or fraudulent posts aimed at misleading job seekers. These issues can have an adverse effect in peoples’ trust and confidence in the platform and make it harder for users to find relevant and suitable opportunities. AI can play a key role in mitigating these challenges by automatically consolidating and standardising data across them. This way, the information is presented in a unified and consistent format, identifying duplicates and merging them, detecting and flagging outdated or potentially suspicious listings (e.g. as done through the vacancy-cleaning functions of a tool used by the Swedish PES), thereby ensuring that the final vacancy database is not only comprehensive, but also accurate and trustworthy.
AI can help employers develop clear, complete, and more effective postings
From the employer’s side, drafting a job description is typically the first step in the matching process. It plays a crucial role in determining the information available to potential candidates and in influencing their first decision to apply or not. Nevertheless, vacancy descriptions are not always complete, clear or effective enough to achieve their intended goal, in other words, attracting the right talent.
There are multiple reasons why this may occur: employers may assume certain skills or requirements are obvious, overlook important details, prepare job descriptions quickly with limited attention to detail, or simply lack the time or expertise to draft clear and complete postings. This can lead to vague or misleading descriptions, making it difficult for candidates to understand what is actually required. Consequently, the effectiveness of the matching process can be hindered, even when the matching tool that performs the matching is highly accurate. The consequences can be significantly exacerbated for MIS beneficiaries and other vulnerable individuals, who often rely on basic keyword searches or filters and, due to their limited skills or experience, depend heavily on clear and precise job descriptions. As a result, when descriptions are vague, they are more likely to misunderstand the requirements and self-exclude from applying.
AI-based functionalities are able to refine and improve job advertisements across multiple dimensions, ideally before these are made available to the public. In the first place, AI can identify whether job descriptions are written in overly complex or confusing language, whether their structure is unclear or poorly organised, and whether key skills, qualifications and responsibilities typically associated with a certain job title are omitted. It can also flag other important missing elements, such as essential information about the firm and the specifics of the role in question. Having identified any potential inconsistencies or gaps, the next process where AI can help is suggesting targeted improvements. Such improvements may include clearer and simpler wording, consistent and uniform language, language and key words that can be attractive to candidates, a better organised structure, addition of missing elements as well as consistent employer branding across multiple job postings. In addition, AI functionalities can analyse the content of CVs, job vacancies and occupational taxonomies, and propose a standardised job title that better fits the description as well as the addition of missing or more relevant skills and responsibilities (e.g. VDAB in Flanders, Canada, Germany, Norway). These suggestions can then be reviewed by employers or recruiters, who can decide whether to accept them.
In addition, employers describe roles and skills in very different ways, resulting in job postings that are unstructured, inconsistent, and difficult to compare. This can have various implications. On the one hand, jobseekers cannot easily understand whether they are a good fit while on the other, it is difficult for job counsellors and social workers to match candidates to suitable opportunities. As a result, mismatches occur, important skills may be overlooked, individuals may miss opportunities that actually fit their profile, and employers may fail to connect to the right talent. AI addresses this problem by extracting skills from job descriptions, mapping them to standard classification frameworks like ESCO, and automatically categorising vacancies by sector, occupation and other characteristics such as experience levels (e.g. this is done in matching tools in Finland, Flanders, Wallonia, Luxembourg). This improves searchability and enables more accurate matching between job requirements with candidate profiles.
AI functionalities can detect bias in job descriptions and make them more inclusive
Job postings may include biases or elements that undermine fair access for all potential candidates. These biases can be related to gender, age, disability, ethnicity, or socio-economic background and could be reflected not only in the language used but also in the requirements of the description itself. For instance, reproducing gender stereotypes, asking for younger candidates or physical abilities that are not truly needed or even degrees that are not necessary for the job. For instance, women may perceive a role as less appealing when job advertisements use masculine‑coded terms such as leader or dominant (Broecke, 2023[6]).
AI can address these issues in several ways. Firstly, AI models trained on large datasets can detect biased or exclusionary phrases or words in job descriptions and suggest replacing them with neutral and inclusive language (e.g. the tool used by the French PES automatically flags wording that may discourage women, older workers or other groups helping employers adjust their vacancies). AI can also compare the job to thousands of similar roles and flag whether there are unnecessarily excessive requirements that may limit access and reduce candidate diversity. In this case too, AI is not only able to flag the issues but rewrite the text based on diversity, equity and inclusion standards. Inclusivity and fair access can also be enhanced by using AI to identify overly complex phrasing and then simplify the language so more candidates can understand it and hence apply, as well as to encourage richer vacancy information (with elements such as hours, flexibility, transport options, childcare) so that job listings better reflect and accommodate the diverse needs of users. The aforementioned functionalities can be particularly beneficial for encouraging applications from vulnerable groups, as they help ensure that postings are more inclusive, accessible, and less likely to discourage – unintentionally or intentionally – candidates who already face greater labour market barriers.
AI algorithms can help employers to adjust jobs to accommodate people with weaker labour market prospects
The aforementioned functionalities focussed on helping employers develop more inclusive job postings, while also identifying potential bias that undermine fair access for all potential candidates. The functionality discussed here goes a step further as it can provide employers with support on how to adjust existing roles to accommodate MIS beneficiaries (and individuals facing labour market barriers more broadly). For even more successful results, this can be done in co‑operation with employer counsellors, as they are the ones who provide support to employers from the PES end.
While AI cannot redesign jobs on its own, it can analyse job descriptions as these are provided in vacancies posted by employers and break them down into specific tasks and skill requirements, using recognised classification systems such as ESCO. Based on this analysis, the system can identify tasks that could potentially be reorganised or even simplified to create roles that better match the skills and capacities of individuals with limited experience or qualifications. For example, in sectors such as logistics or retail, certain tasks, such as preparation of orders or organisation of storage could be grouped into entry-level roles that provide a feasible entry point into employment for these target groups.
While information on workplace adjustments, flexible job design, mentoring possibilities, or the use of available support services such as job coaching or job shadowing, or training programmes offered within the workplace are not typically included in job postings, they could be systematically incorporated into job descriptions in co‑ordination with employers. This would make opportunities more accessible and transparent for individuals with vulnerabilities and additional support needs.
Pillar 3: AI functionalities can help identify the needs of MIS beneficiaries as well as their skills and make them more visible to employers
AI functionalities embedded in a matching platform can help identify the needs and skills of MIS beneficiaries and other vulnerable jobseekers more accurately, thereby building more structured and complete profiles. These insights can be then used to connect individuals with realistic employment opportunities based on available vacancies, as well as to suggest training or other measures should specific gaps be identified. AI can also flag individuals who require additional support, providing staff supporting end-users such as job counsellors and social workers with information that can help them prioritise and tailor their assistance accordingly.
AI can help identify the skills and needs of jobseekers and therefore support personalised matching
Identifying the skills and needs of jobseekers is a fundamental step in effective matching. Yet many individuals struggle to articulate their skills and abilities, recognise skills they may possess but cannot easily track, or understand how their skills and prior experiences can relate to labour market requirements. This challenge is often more pronounced among vulnerable jobseekers, who may have limited work histories, fragmented careers, or limited skills.
AI can help overcome these barriers by analysing information provided by jobseekers. Such information may range from past jobs and education to informal experience, or even short descriptions of tasks performed. It can then map this information to recognised skills frameworks and occupational taxonomies or job dictionaries. Through this process, AI can identify relevant skills that jobseekers may not know they possess or may have but do not explicitly report and highlight potential gaps. These functionalities can combine information from various administrative and non-administrative sources, such as jobseeker data and employment records, preferred occupations and platform behaviour to provide a more complete and accurate picture of each person’s abilities and circumstances. For example, in Greece, MIS beneficiaries often end up in roles different from their stated preferences, which could be partly explained by gaps in how their skills are captured. AI can help identify transferable skills and suggest occupations that users may not have considered but are compatible with their experience.
AI-based skills assessments can also be integrated into a matching platform to evaluate a wide range of competencies – occupational, digital, soft, basic, or language skills – or even to uncover informal or transferable abilities that may not be captured in a traditional CV. These insights are particularly valuable for individuals with limited formal work experience or formal certificates. In addition, AI can help identify barriers that affect employability by analysing indicators such as prolonged unemployment, limited work histories, or low engagement with employment services. This enables the system to flag jobseekers who may require more intensive support or specialised services.
Together, these functionalities contribute to the development of a consistent and comprehensive skills profile that can in turn generate more opportunities for improved matching and personalised guidance. Having this understanding of jobseekers’ profiles, AI algorithms can then prioritise vacancies that meet individual constraints or preferences. For example, vacancies that are close to the jobseeker’s location, offering flexible hours, or aligned with care responsibilities or mobility limitations. Users can also rely on filtering options should they want to narrow down the opportunities suggested by the tool. For instance, contract type, work modality (remote, hybrid, in-person), wage levels, or access to childcare support.
AI can also compare each individual’s strengths and skills with labour market requirements and highlight potential gaps. Based on this analysis, it can provide personalised guidance, accompanied by clear explanations and the reasoning behind the relevant recommendations. For vulnerable jobseekers, such as those with limited work experience, low qualifications, weak digital skills, or a history of long-term unemployment, AI-generated recommendations can sometimes seem opaque or difficult to trust or follow. Providing a clear, simple rationale behind each suggestion is therefore particularly important. Transparent explanations help jobseekers understand why a certain vacancy or occupation is relevant to and suitable for them, which increases confidence in the system and encourages follow-through.
AI can support jobseekers in expanding their job search strategies or encouraging them to consider a broader set of occupations through a range of practical steps often relying on behavioural insights.8 For example, it can prompt a user to expand their search radius by 10%, as such a strategy could increase the number of suitable vacancies by 20%. AI can also identify near-match occupations, showing individuals roles that align with their existing skills – as second or even third best fits – and what additional competencies they may need to develop to qualify for them.
The microdata analysis shows that vulnerable jobseekers are often concentrated in part-time, temporary, or unstable forms of employment. In such cases, AI could indeed recommend complementary job opportunities that match their available hours, skills, and schedule constraints, alongside full-time alternatives. By analysing work patterns, location, and required skills, AI can identify suitable secondary roles that help individuals top up their income or transition gradually towards more stable jobs. This type of tailored recommendations can be especially valuable for vulnerable individuals who may be employed – yet in fragmented or low-hours jobs – and who need flexible options to improve their overall labour market situation. Finally, to support engagement, AI can further break down recommended actions into small, manageable steps, helping jobseekers more easily act on personalised advice and progress toward suitable employment opportunities.
AI can assist jobseekers in articulating their skills and experience in a clear and compelling way
Even after identifying their skills and needs, many jobseekers may struggle to present themselves effectively to employers. This is particularly true for vulnerable jobseekers, who may be unfamiliar with key job search processes such as drafting a CV, a cover letter and preparing for an interview or just feel uncomfortable in doing so.
Similar to its role in improving job descriptions, AI can draw on large databases of CVs and job vacancies to identify relevant skills or experiences a candidate may have missed and suggest improvements to their CV or cover letter (e.g. these functionalities exist in the matching tools of VDAB in Flanders as well as the French PES). An AI algorithm can also recommend keywords that may increase the likelihood of being shortlisted for an interview. It can also help candidates present their skills more effectively to employers during an interview, for example via the help of tutorials or written guidance.
Especially for the most vulnerable jobseekers, AI can greatly simplify the application process. Instead of submitting a fully developed CV, applicants can draft and provide only a few bullet points describing their key competences and basic information. AI can then transform these inputs into a complete, well-structured CV or automatically populate the employer’s application form when required. As a consequence, jobseekers save significant time and effort on the one hand while on the other their profiles gain more visibility and are less likely to be overlooked by employers – an advantage of particular importance for the most disadvantaged candidates. For individuals in vulnerable situations who may lack the skills, confidence, or resources required to engage with AI-assisted application systems, this functionality can be complemented by additional support measures such as support from a job counsellor, guided interfaces or digital skills training (for a detailed discussion on this topic see also Chapter 6 – Section 6.3.2.).
AI can promptly identify individuals who require additional support and guide them towards the right human assistance
Even when algorithms operate highly accurately and provide high-quality matching suggestions, some individuals – particularly those facing complex barriers – may still struggle to navigate the job search process on their own. Vulnerable jobseekers may experience low confidence, limited digital or labour market literacy, or practical obstacles that prevent them from acting on recommendations. As a result, they may need an additional push or tailored human support.
AI can help address this challenge by identifying when users are struggling or disengaging and prompting them to take appropriate action. For example, if a jobseeker repeatedly receives suitable matches but does not apply to these positions, or if they abandon applications halfway, the system can also provide concrete, actionable recommendations that can help them move forward. These can include direct prompts that encourage users to take the next step, such as “Go ahead and apply, you meet most of the requirements”, or motivational messages like “Congratulations, you’re only one step away from your dream job!”. In the case of Greece, many GMI beneficiaries were already registered with DYPA before applying for entering the scheme. In practical terms this implies that being “registered” does not necessarily lead to activation or strong labour market outcomes. The machine platform should therefore use behavioural nudges such as reminders, simple tasks, or step-by-step guidance, to help jobseekers stay engaged and act upon the recommendations offered. In addition, the low transition rates into employment indicate the need for intervention as early as possible. This suggests that the platform must identify individuals at higher risk of long-term unemployment early on, so that job counsellors intervene sooner and offer more intensive support when needed.
Importantly, when behavioural patterns indicate that a jobseeker may benefit from more personalised, human assistance, the system can automatically prompt them to connect with job counsellors or social workers. For instance, if an individual has not logged in the platform for several weeks, shows signs of distress in chatbot interactions, or repeatedly applies to unsuitable roles, the platform can recommend scheduling an appointment with an employment counsellor or social worker. In this way, AI not only improves matching per se, but also acts as an early warning system (EWS), helping ensure that individuals who need extra support are quickly identified and directed to appropriate human assistance.9 First, early detection can help prevent vulnerable jobseekers remaining in limbo for long while. Second, it can help relevant institutions take action before problems escalate. In Belgium, since more than half of new entrants remain on the scheme after one year, the platform should integrate early-warning signals that identify users at risk of long-term dependency, who may require more intensive activation or in-person support.
Pillar 4: AI can help end users make more informed decisions
End users of matching platforms – whether jobseekers, employers, or counsellors and social workers – often struggle to translate labour market information into decisions that are meaningful for their specific needs. AI can bridge this gap by transforming complex labour market data into digestible information or even personalised recommendations, so platform users make well-informed, forward-looking decisions.
AI can turn complex labour market information into actionable insights
AI can analyse large volumes of data, both historical and real-time, to identify patterns in labour market trends, industry demands, salary expectations, and high-growth sectors. It can also predict future labour market needs and trends, offering valuable insights to the stakeholders interacting with the matching platform.
Jobseekers, employers, and employment services often struggle to interpret labour market information (LMI), which is typically fragmented, overly technical, or quickly outdated. As a result, many end-users make decisions based on incomplete or inaccurate perceptions of labour market conditions.
AI can address this challenge by analysing large volumes of both historical and real-time data to detect patterns, trends, and shifts in labour demand. By processing job postings, economic indicators, company growth signals, and sectoral developments, AI can generate clear and timely labour market insights that help users understand where opportunities are growing or declining. This allows jobseekers to make more informed decisions about where to apply, which skills to develop, and how to adapt their job search strategies in line with market realities. For jobseekers, this may include guidance on sectors with rising demand, suggestions for skills to develop, or alerts about emerging roles aligned with their profile. For employers, AI can highlight labour market tightness, the availability of talent for specific roles, or competitive hiring trends. PES can use these insights to shape counselling, targeting, and programme design.
In Greece, the gap between the occupations MIS beneficiaries want and the jobs they actually obtain suggests that many users may not have accurate information about labour market demand or their own skill profiles. The platform should be therefore able to provide information regarding in-demand occupations, showcase to users which jobs match their skills and aspirations more closely, and highlight realistic job pathways. Ideally, the presentation of such information should be done in a clear and easy to understand way. Large discrepancies between preferred occupations and actual job outcomes may signal unrealistic expectations or barriers not captured by the system’s algorithms. The platform should be able to flag such cases to job counsellors so that more personalised guidance can be provided. In the case of Belgium, the significant differences between regions underline the need for region-specific labour market intelligence. AI tools can help by adjusting recommendations to reflect regional demand, available sectors, and common hiring practices.
AI can improve employers’ understanding of who they reach and where to concentrate their efforts
Employers often lack visibility into whether their job advertisements are reaching a broad and diverse pool of candidates, which can unintentionally limit access for underrepresented groups. Without such information, employers cannot easily tell which recruitment channels work well or where outreach gaps persist.
AI-powered analytics are able to assess who is engaging with job postings, analyse the relevant data and therefore identify groups that are underrepresented in applicant pools. They can also detect where outreach is falling short. In other words, AI can provide employers with insights into which audiences they are reaching effectively, and which ones require additional effort. In practical terms, this can be done by comparing these patterns to broader labour market benchmarks at the sectoral, national and even regional or local levels. This helps employers refine their recruitment strategies, increase their reach, improve diversity in applicant pools, and engage more effectively with vulnerable or overlooked groups.
AI algorithms can generate insights that support policy design, particularly training and ALMPs
As users interact with the platform in multiple ways and for various reasons (such as exploring long-term career options, searching for training opportunities, or following recommended pathways), the system collects valuable information on skills, career interests, barriers etc. At the same time, labour market data also feed in the platform and therefore data on skills demanded and emerging labour market patterns are also being accumulated.
AI algorithms can analyse this information in an aggregated and anonymised form to identify broader trends. For example, the system could highlight which training programmes are most frequently recommended or searched for, which career transitions are most common among MIS beneficiaries, and which skills appear most often in demand across different sectors. It could also detect emerging occupations or shifts in labour market demand by combining user behaviour with vacancy postings.
These insights can be particularly valuable for counsellors when recommending career pathways and ALMPs. For policymakers, the aggregated results can help inform the design of programmes whether training or other ALMPs, identify emerging skills needs, and support more evidence‑based policy decisions.
Pillar 5: AI algorithms can facilitate career management related decisions for both MIS beneficiaries and employers
By integrating career recommendations into the platform, the system would not only match jobseekers to current vacancies but also guide them towards occupations with stronger employment prospects, including those that may require additional skills or experience. These functionalities are particularly relevant for vulnerable groups, such as MIS beneficiaries or long-term unemployed individuals, who may face multiple barriers to labour market integration and may benefit from clearer guidance on realistic and sustainable career trajectories.
AI can link jobseekers to support services, helping address barriers that may need to be resolved before or alongside employment
A key AI functionality of the career recommendation solution should be its ability to automatically suggest and direct jobseekers to services beyond employment. This is particularly relevant for MIS beneficiaries and other vulnerable groups who often face complex and multiple barriers that may hinder their employment prospects. Such challenges can include housing insecurity, psychological distress, caregiving responsibilities, or lack of childcare. Addressing these issues is often a prerequisite for securing or sustaining employment. The platform could therefore guide users towards relevant social and psychological support services, community-based assistance, childcare or eldercare facilities, and even trusted external websites offering practical information.
More broadly, digital career recommendation tools should be embedded within a wider continuum of employment and social support services. The outcome of a career recommendation is often not a job itself, but a pathway involving training, certification, social support, and ultimately employment. If these complementary interventions are not available, recommendations may be difficult to translate into practice. Countries with more advanced approaches increasingly integrate digital career guidance with active labour market programmes and social services. For example, Korea’s one‑stop Employment Welfare Centres combine digital tools with personalised support for housing, financial assistance, and mental health needs, recognising that vulnerable jobseekers often face multiple barriers that need to be addressed simultaneously (Box 5.3).
Korea has developed an integrated network of one‑stop shops that unify employment, welfare and re‑employment services to better support jobseekers facing complex barriers to labour market access. Originating in the late 1990s with Employment Stability Centres and evolving through successive reforms into Employment Welfare+ Centres, these facilities now operate nationwide and aim to provide seamless linkage between income support, personalised counselling, job training and referral services. Core services include targeted counselling for specific groups (such as youth, women with career interruptions and low-income jobseekers) and co‑ordinated support across agencies to address obstacles like skills gaps, financial instability or care responsibilities.
Beyond traditional face‑to-face support, Korean authorities have begun to introduce digital and AI-enhanced components to strengthen job matching and service delivery within the one‑stop ecosystem. Recent initiatives include AI-based online and mobile consultation systems that augment in-person counselling with data-driven diagnostics of jobseekers’ skills, preferences and local labour market information, supporting more tailored employment pathways. These digital tools are designed to complement case management, improve the responsiveness of services to individual needs and expand the reach of employment support for those who might ot0herwise be disengaged from formal systems.
Implementation of the one‑stop model has highlighted important operational and governance considerations for multi-service delivery. Sustained inter-agency co‑ordination, adequate staffing and counsellor capacity are needed to deliver comprehensive support that spans income support, training and placement assistance. As services become more digitalised, ensuring equitable access remains a strategic priority. This is particularly so for individuals with limited digital skills or connectivity. Korea’s integrated service delivery, supported by evolving digital tools, aims to enhance inclusivity in labour market support while also underscoring the need for ongoing monitoring, capacity building and alignment of digital and in-person offerings.
Source: Yeo (2025[7]), Presentation during International Workshop 1: Identifying employment obstacles and opportunities for vulnerable groups conducted on 15 May 2025 as part of this project, https://www.oecd.org/content/dam/oecd/en/about/programmes/dg-reform/using-ai-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries/Integrating-the-most-vulnerable-jobseekers-The-examples-of-Korea-one-stop-shops.pdf.
Taking into consideration, the results of the data analysis in Chapter 4, this functionality can be particularly useful for the subset of MIS beneficiaries who have been detached from the labour market for an extended duration, as well as those who are unable to work due to various circumstances. Within this framework, this functionality would be particularly valuable if Greece and Belgium developed a centralised map of social services, allowing the platform to automatically suggest nearby or relevant support options based on each user’s profile following the example of the Lithuanian PES. The information generated by the platform could also support case managers, including job counsellors and social workers, in making informed decisions about referrals and support measures while retaining responsibility for final decisions.
The findings from the data analysis further reinforce the importance of this functionality. While MIS beneficiaries participate in similar types of ALMPs as the general jobseeker population, the most vulnerable groups exhibit particularly low participation in ALMPs overall. This may reflect limitations in the targeting and suitability of existing measures. In such cases, the platform could recommend alternative or complementary forms of support, including social, health, or counselling services, either alongside or prior to labour market interventions.
AI can provide recommendations on pathways beyond immediate placement to support career growth and sustainable labour market integration
The matching related functionalities of the job matching platform allow the matching of jobseekers with entry level opportunities, according to their profiles and based on labour demand. This functionality goes further beyond immediate placement to support individuals if a feasible match is not possible. Instead of suggesting the first available vacancy, AI algorithms can – by analysing patterns in historical data on the one hand and a person’s skills, work history and interests on the other – identify postings or occupations (even when such postings are not yet available) that have more frequently led to longer-term employment opportunities. The system then can suggest jobs that could serve as entry points to a longer-term career or better jobs. This would allow individuals to move towards more stable and better-quality employment over time. Longer-term or better jobs can be defined using observable labour market outcomes such as wage progression, employment stability, contract duration and jobs requiring higher skills. The definition may vary depending on the country context and policy priorities and can be refined over time.
By analysing labour market data, including at a regional/local/sectoral level, this component of the platform could also identify occupations that are growing in specific regions, sectors, or local labour markets. Within this framework, information from administrative sources like employment data and non-traditional sources such as job vacancy data can be useful inputs for AI algorithms to be able to detect patterns that indicate which sectors and most importantly occupations are experiencing increasing demand (or on the contrary are experiencing slowdowns and therefore could be avoided). These can be complemented, wherever and whenever feasible, with information derived from selected sectoral reports and business studies to enrich the analysis. The results can be presented to users as potential career opportunities. As part of this functionality, predictive models such as early warning systems could also be deployed to detect individuals at risk of job loss and therefore cycling back into unemployment.
The current support systems in both countries prioritise immediate labour market entry over sustainable integration. The data analysis in Chapter 4 indicates that a significant share of MIS beneficiaries in Greece is employed while simultaneously enrolled in the scheme and receiving the relevant benefit and services. This implies that the income is not sufficient to support their independent living. Another important finding is that MIS beneficiaries are frequently employed in precarious jobs with limited prospects for career progression. Employment spells also tend to be short, with many individuals returning to unemployment after relatively brief periods (e.g. around eight months), pointing to a pattern of recurrent labour market instability. Furthermore, transitions into employment do not consistently lead to improved outcomes, as only a small share of beneficiaries (approximately 13%) move into higher-quality jobs, while the majority enter positions of similar or lower quality. Similarly, the data analysis for Belgium shows that one in two MIS recipients remain on the benefit one year after signing up, and one in three, three years after. In addition, MIS beneficiaries – especially those with Article 60 experience – are often channelled into short-term, low-quality jobs (such as in the service sector as well as administrative and support services including temporary agency work and cleaning) with limited upward mobility, resulting in repeated transitions between the MIS scheme, unemployment benefits and employment.
In this context, AI-driven pathway recommendations can really help those individuals in accessing and transitioning into more profitable, longer-term and sustainable employment opportunities, until they exit cycles of precarious employment and benefit dependency. In addition, it can contribute to reducing the duration between successive employment spells, especially if combined with strong and proactive outreach efforts. The tool could also use outcome‑based data to highlight career pathways and job transitions associated with more positive and sustainable outcomes, while signalling those linked to recurrent instability, and based on these insights guide individuals towards potentially better choices. In the case of Belgium, higher persistence on the MIS scheme and lower labour market integration rates among MIS beneficiaries in certain regions (i.e. Brussels and Wallonia) are observed. This is even more prominent in the case of unemployment benefit recipients who previously participated in the Article 60 programme, who are found to have substantially lower employment rates than those without prior programme participation. While this partly reflects negative selection, it also highlights that initial activation measures alone are not sufficient, reinforces the need for more targeted support and underscores the need to support progression into stable, higher-quality jobs. AI-driven pathway recommendations are particularly relevant in this regard, as they can incorporate regional labour market data to propose personalised and context-specific career pathways, improving the chances of sustainable integration.
This type of functionality could be particularly useful for vulnerable jobseekers, including MIS beneficiaries. Many people in this group have fragmented work histories, limited or even no formal qualifications or unclear career direction. As a result, they may repeatedly move between short-term or low-quality jobs. In addition, these individuals may not be familiar with local labour market dynamics or emerging employment opportunities. AI-supported functionalities can showcase alternative occupations that require similar competencies and highlight realistic progression routes. This helps jobseekers understand whether they can arrogate positions that can lead to more sustainable employment and therefore the steps they should follow to get into these roles.
For job counsellors and social workers, such AI functionalities can help them propose more targeted career pathways, rather than focussing only on immediate job placement. At the same time, they can direct individuals towards occupations with stronger employment prospects and better long-term opportunities. For beneficiaries experiencing recurrent unemployment spells, the system could automatically trigger alerts to counsellors so they can intervene providing more intensive support as needed (in conjunction with the functionality for referrals to human-led services discussed in Chapter 7). As with all AI related recommendations, such recommendations too are only indicative and can be taken into consideration by counsellors as appropriate and relevant.
A recurring issue for counsellors is that jobseekers may focus narrowly on familiar occupations. This issue may be particularly acute for MIS beneficiaries, who often have more basic skills and may lack confidence to expand their job search horizons. This can trap individuals in a cycle of unstable jobs. AI-enabled career orientation tools can broaden these horizons by suggesting alternative occupations that build on existing skills. These tools build further on those focussed more narrowly on job matching, by providing jobseekers (and counsellors) suggestions as to how existing competencies may be added to or developed further, to unlock new career pathways, in the same or different professions. These recommendations may also incorporate training or placement suggestions as part of the pathways towards a new career.
Arbetsförmedlingen in Sweden provides a relevant example. Its new AI system was explicitly designed to recommend alternative occupations and career pathways beyond the individual’s original search, using machine learning and graph analytics to highlight related fields. Occupations are recommended which build upon an individuals’ existing competencies (Brioscú et al., 2024[1]). This guidance helps to provide inspiration that can help jobseekers adapt to emerging opportunities in the labour market and to continue development of skills to unlock different career paths (see Box 5.4 for a discussion about digitalisation and AI use within the Swedish PES). For MIS beneficiaries, who may not have formal qualifications, this approach makes transferable skills more visible and supports more inclusive matching. There are plans to extend the tool to provide more specialised job recommendations to jobseekers with a disability (Brioscú et al., 2024[1]), as well as improve the tool design with newer AI technologies.
Using both jobseeker profile information from administrative data and vacancy information, the AI Job Matching tool suggests job opportunities to users by extracting key competency and occupational keywords. The PES website provides access to additional tools that are freely available to anyone including both registered and non-registered jobseekers and workers which can provide information about each occupation, recommend career pathways including jobs beyond the profession they initially selected, education and training opportunities, study and career counsellors in their municipality to book individual counselling sessions – features that are being continuously evaluated and refined using customer feedback and analysis of user behaviour.
Job vacancies, posted by employers as well as imported from external websites, are collected in Arbetsförmedlingen’s Job Bank, Platsbanken. Employers can search for candidates, access recruitment guidance and resources, and apply for schemes such as wage subsidies, new start jobs, training, and internships. Since 2023, Arbetsförmedlingen employment officers use an AI application to review job advertisements by detecting fraudulent or discriminatory content. In 2024, a large‑scale fraud detection system was introduced, combining multiple advanced techniques, including machine learning, deep learning, and network analysis across extensive datasets. Arbetsförmedlingen also deploys AI to generate labour market insights by analysing job postings to identify skill needs, education requirements, as well as task composition across occupations. These insights support monitoring labour demand and feed into the job-matching and career recommendations tools that Arbetsförmedlingen has at its disposal (Brioscú et al., 2024[1]).
Source: Study visit to Arbetsförmedlingen, the Swedish PES conducted in February 2026 as part of this project.
Following the identification of gaps in skills or experience, AI can direct individuals to skills development programmes and ALMPs
AI functionalities can also recommend training opportunities and ALMPs that help jobseekers overcome the gaps and bridge them with employment. After analysing a person’s profile, such as their skills, qualifications, work experience and career interests, the system can map this information against labour market requirements derived from job vacancy data and occupational classifications (e.g. ESCO). By comparing the individual’s existing skills with those required for target occupations, the system can identify specific skill gaps. Based on these gaps, the platform can then suggest relevant programmes or employment support measures that would improve the chances of finding or even retaining work. These programmes can include a range of options, such as training programmes, short courses, apprenticeships, certification schemes or other ALMPs. Personalised nudges and prompts can be used to further encourage taking up these options.
Naturally, training is the most straightforward outcome of such a functionality, as it helps address skills gaps and can lead to employment, which is the ultimate objective of a comprehensive matching platform. Recommendations on training, upskilling or reskilling programmes can span across various topics including on improving technical/vocational, basic, soft, language, computer skills. It can also highlight training programmes that have led to positive employment outcomes for jobseekers with similar characteristics in the past. To make these recommendations stronger and more convincing for users, the platform could provide some indicative information on how completing a programme may be associated with improved employment prospects or with access to a wider range of job opportunities. Providing such practical explanations can help users better understand the rationale behind the suggestions, build trust in the tool, and encourage them to act on the recommended steps.
Potential integration with individual learning accounts (if and when such systems might be available) could further strengthen this functionality.10 The platform could then inform users about the training opportunities they are eligible for, the funds available to them, and how these can be used to finance recommended courses. This would make it easier for individuals to translate career guidance into concrete action by directly linking suggested training options with the financial resources available to support participation.
Other recommendations may include short courses or certifications that can help users move from entry-level positions to better-quality jobs. The platform could also suggest seminars, mentoring programmes or workshops on topics such as job search strategies or CV preparation where these are identified as needed. Beyond informal training programmes, the platform could also recommend formal education pathways where these are relevant for an individual’s career progression. While training courses often help address immediate skills gaps, some career transitions require more structured educational pathways that lead to recognised qualifications. If interconnected with national education data systems, the platform could identify and suggest relevant educational programmes at different levels. These may include university degrees, formal VET programmes or post-secondary professional programmes. The system could also recommend second-chance education opportunities, such as programmes that allow adults to complete secondary education or obtain equivalent qualifications, which can be particularly important for individuals with low educational attainment. For example, if a user expresses interest in moving from their current care assistant role to a licensed nurse, the platform could highlight the relevant formal education pathways required for this transition, including university degrees in healthcare or vocational nursing programmes. To support even more informed decisions, the platform could outline key information about these programmes, such as entry requirements, duration, location, and potential employment outcomes/testimonials from graduates. Where available, it could also indicate whether financial support, scholarships, or learning accounts could help cover participation costs.
In addition to training and education opportunities, the platform could also recommend other types of ALMPs offered by the PES. These may include subsidised employment schemes, work experience placements, public works programmes, initiatives supporting the creation of a new business or other less common ALMPs. Such measures can provide immediate opportunities to gain work experience, maintain labour market attachment, or generate income while individuals continue developing their skills. These recommendations could be based on a combination of user data, such as skills, work history, identified barriers and aspirations as well as administrative data on previous participation in ALMPs and employment outcomes of individuals with similar characteristics. They could also draw on information about local labour market conditions. Programme availability should be also taken into consideration.
Recommendations should be tailored to each individual. In some cases, although training may be beneficial in the longer term, a jobseeker’s immediate priority may be to secure a source of income or re‑enter the labour market after a period of inactivity. In these situations, programmes such as subsidised employment or work experience placements may represent a more appropriate first step. The platform could take these inputs into consideration and suggest the types of programmes that best match a person’s needs at a given moment, helping counsellors and jobseekers identify pathways that combine both short-term stability and longer-term career development.
This functionality could be particularly advantageous for vulnerable jobseekers, including MIS beneficiaries who are often found to have been out of the labour market for extended periods. The tool can support their reintegration by connecting them with relevant training and ALMPs, designed to improve employability. Many individuals in this group may not know which training programmes or ALMPs exist, or which ones are most relevant for their situation. AI can simplify this process by presenting a limited set of clear and realistic options, helping users focus on opportunities that are most likely to lead to employment. Furthermore, this type of guidance can make career planning more understandable and manageable. Instead of searching through many programmes, they receive targeted suggestions that match their profile and local labour market needs. This can increase motivation and help them take practical steps toward improving their employability. This functionality can prove beneficial for counsellors and case workers, too, as they can propose more evidence‑based pathways and save time when assessing possible interventions. Nevertheless, these results should be seen as indicative as the final decision for a referral should come from a human.
As mentioned above, in the case of Belgium, the data analysis reveals weaker employment outcomes among former Article 60 participants once they transition into the unemployment system. This highlights – in addition to improving job quality and aligning placements with local demand – combining Article 60 with training as well as post-placement support. AI can support this process by identifying skills gaps and directing individuals towards complementary measures such as relevant training programmes and other ALMPs, helping to move beyond Article 60 as a standalone measure and towards more durable labour market integration.
AI can go beyond identifying skills to making them visible, interpretable, and usable in the labour market
Another potential use of AI within a career recommendation tool is to support skill recognition, going beyond simple skills identification. In other words, the objective is not only to detect skills but also to make them visible and usable in the labour market. Many jobseekers, particularly those from vulnerable backgrounds, face difficulties in demonstrating their abilities to employers. Hiring processes often rely heavily on formal qualifications, certificates, or clearly documented work experience. However, many individuals possess valuable skills acquired through informal work, volunteering, household responsibilities, or other life experiences that are rarely recorded in formal ways and therefore are very difficult to be proven.
AI algorithms can help address this gap by analysing information provided by users, such as descriptions of tasks performed in previous jobs, informal work experience, or everyday activities (usually performed in the form of skills assessments), and translating this information into formal job and skills requirements. In doing so, the system can map these experiences to established skills frameworks, such as the European Skills, Competences, Qualifications and Occupations (ESCO) classification. This process allows informal and previously unrecognised competences to be expressed in a structured way that employers can easily understand and value, even though they cannot really assess the exact tasks reported nor how well those tasks were performed. This is particularly valuable for the jobseekers themselves as they can relate their own strengths to the skills demanded in postings as well as job counsellors who can support them better using these inputs.
For example, a jobseeker who has spent several years caring for their elderly parents may not consider this experience relevant to employment. However, an AI-enabled system could analyse the tasks performed such as providing daily assistance, managing medication and nutrition planning, managing household expenses, scheduling appointments and co‑ordinating with healthcare providers, and translate them into recognised skills such as care co‑ordination, basic health support, time management, and communication. These skills could then be mapped to occupations in sectors such as social care, healthcare assistance, or community support services. By making such skills visible and comparable to formal qualifications, the tool can help broaden the range of opportunities available to individuals whose experience is not captured through traditional credentials even across different labour markets.
Training and certification schemes can be integrated within the environment of the job matching and career recommendation platform
Many jobseekers, particularly those further from the labour market, face skills gaps that limit their access to good employment opportunities. At the same time, they may struggle to identify relevant training programmes or obtain certifications that would help them demonstrate their competences to employers, especially without support from a job counsellor or social worker.
Career recommendations solutions can integrate active learning modules and certification opportunities directly within their digital environment. By integrating training courses, certification schemes, and learning resources directly into the platform, users can easily identify programmes that address their specific skills gaps and support their career goals. The platform could redirect users to external training providers offering relevant courses. Training providers could in turn use the platform to upload and showcase the programmes they offer, making them more visible to jobseekers and counsellors. The platform could also notify users about upcoming training sessions, newly available courses, or local training opportunities relevant to their profile. In this way, the platform becomes not only a tool for finding jobs but also a mechanism for continuous learning and skills development.
Basic reminders such as notifications before the start of a course or alerts for upcoming sessions can be delivered through standard automated systems. AI can enhance this functionality by enabling personalised prompts. For example, AI models can analyse the user’s profile as well as their engagement patterns to identify individuals at risk of dropping out and tailor the timing, frequency, and content of messages accordingly. Such features can help encourage participation and engagement in recommended training programmes and reduce dropout rates.
Over time, and if such tools prove effective in improving labour market outcomes, employers may also have an interest in supporting their further development. For instance, large employers or groups of employers facing similar skills shortages could contribute to the design or co-financing of training programmes hosted on the platform, ensuring that the courses offered respond to real labour market needs while helping build a poll of qualified candidates.
AI can significantly enhance career planning and task management
Another potential use of AI within the career development component of the job matching platform is to support individuals in organising and managing their career progression more effectively. Many jobseekers, and especially those facing multiple barriers, often struggle not only with identifying suitable opportunities but also with structuring the steps required to move towards better and sustained employment. Career development can involve numerous actions some of those taking place in parallel, such as improving a CV, drafting cover letters, completing training courses, exploring alternative occupations, preparing for interviews, expanding job search strategies and others. Without clear guidance, these tasks may appear overwhelming, leading individuals to disengage from the process.
AI-enabled career development solutions can help address this challenge by translating career recommendations into clear and manageable actions. To do so, the system would need access to data such as the user’s education, skills, work history, employment preferences, participation in previous ALMPs and trainings, and any identified barriers to employment, combined with labour market information on vacancies, skills requirements, typical occupational transitions, available ALMPs and training opportunities. Based on these inputs, AI can identify gaps between a person’s current profile and a target occupation and generate a personalised structured career plan that breaks the pathway towards career advancement into intermediate smaller steps. Such steps may include for example, intermediate jobs, training options, and skill-development actions. This plan can help users prioritise tasks, track their progress, and mark completed actions and can be validated and reviewed by the job counsellor from time to time. This functionality would rely on a combination of AI approaches, including recommendation systems, skills-matching models and predictive models trained on historical career trajectories and employment outcomes.
For example, if a user is interested in transitioning from a low-paid service job to a role as a healthcare assistant, the platform could automatically generate a series of recommended actions. These might include completing a short certification in basic healthcare support, updating the user’s CV to highlight relevant transferable skills, attending a local training programme, and applying for specific entry-level positions in the healthcare sector. As the user progresses, the system can update recommendations, remind them of upcoming tasks, and suggest additional steps if needed.
Such functionalities can be particularly beneficial for vulnerable jobseekers who may face difficulties in navigating complex administrative processes, identifying relevant opportunities, or maintaining motivation during the job search process. By breaking down career progression into smaller, achievable steps, the tool can help individuals stay engaged and maintain a sense of progress. Prompts that recognise the completion of these smaller steps are a nice way to sustain motivation, encourage users’ engagement with the tool and keep them remaining focussed on their longer-term career objectives. Behavioural techniques could be used in this respect. For example, phrases such as “Great job completing your CV, that’s an important step. Keep going!”
At the same time, these tools can also benefit individuals who are already employed but wish to improve their labour market situation and conditions. Workers in unstable, low-quality and precautious jobs may use this functionality to organise the steps needed to transition into more stable employment, which is particularly important given the limited amount of time they might have at their disposal. In this way, AI‑supported career planning tools can contribute not only to labour market entry but also to longer-term career mobility and progression.
In the case of both Greece and Belgium where many MIS beneficiaries seem to move repeatedly between the scheme, unemployment and low-quality jobs, there is a clear need for more continuous and structured support. Career pathway recommendation systems that map step-by-step trajectories, combined with adaptive guidance that updates recommendations as users transition between different states and statuses, can help organise next steps in a structured way and ensure greater continuity in support.
AI can help identify hard-to-fill vacancies and predict emerging opportunities and employers’ hiring patterns
Jobseekers and also counsellors often lack clear visibility into which vacancies are genuinely difficult to fill, future labour market needs and emerging opportunities or which employers are likely to hire in the near future. Without such insights and given that labour market changes occur in a rapidly and sometimes unexpected fashion, jobseekers’ efforts may be misdirected or may focus on highly competitive roles with low chances of success. At the same time, employment services may not prioritise support where it is most needed.
AI can help by analysing labour market data to detect vacancies that remain open for long periods, require scarce skills, or repeatedly appear over time, all of which are strong indicators of hard-to-fill roles (like the tool deployed by the French PES that can identify such positions and subsequently help employers make them more appealing to jobseekers). For example, AI algorithms in this domain can i) predict the time it will take to fill a vacancy, and ii) calculate the attractiveness of a vacancy. This allows employer counsellors at the PES to reach out to employers to resolve any issues and to propose solutions to enhance the prospects of a vacancy (Brioscú et al., 2024[1]).
AI can also use predictive analytics to anticipate future labour market needs by analysing long-term patterns and real-time signals. By analysing large volumes of labour market data, such as job postings, company growth indicators, industry reports, and economic trends, machine‑learning models can highlight emerging occupations as well as rising job titles (e.g. the PES in Slovenia supplies information on past vacancy trends and current regional demand for specific occupations, providing a useful reference point for both jobseekers and policymakers). They can also track changes in demand for specific skills, alerting users to upcoming growth areas before they peak. This early visibility enables jobseekers to prepare in advance and supports PES in designing more relevant and effective programmes. For employers, these insights can strengthen workforce planning by signalling where skill shortages or recruitment challenges are likely to arise. AI can also identify employers or even whole specific sectors most likely to recruit new staff, particularly MIS beneficiaries and individuals from vulnerable categories – even before job vacancies are officially posted – by examining historical recruitment patterns, expansion signals, and company-level characteristics and trends (e.g. France, the Netherlands). These insights can enable jobseekers and counsellors to target employers with higher hiring likelihood. They can also help employers understand where they may face recruitment challenges and employer counsellors where to focus their outreach and support efforts.
5.4.2. The platform’s functionalities should be able to work together through key workflows, transforming inputs into outputs
In addition to the platform’s core functionalities (the so-called five pillars outlined above in Figure 5.4), it is essential to understand how these interact with each other through key flows. Figure 5.5 presents a simplified conceptual overview, illustrating the main inputs required for the operation of the matching and career recommendations platform, as well as the flows through which these inputs are transformed into key result outputs relying on the underlying algorithms.
At the core of the platform is a centralised matching algorithm that processes validated inputs from both the supply side (i.e. jobseekers and workers) and the demand side (i.e. employers) to generate immediate job‑candidate optimal matches. Individual-level data such as personal information, demographic characteristics, skills, experience and others once verified, are fed into the system. At the same time, validated employer inputs also enter the system – these include business related information and, most importantly, vacancy data. Vacancy data may come from multiple sources, including existing platforms, as well as direct inputs on this platform. Over time, the platform can be progressively integrated with external databases such as LinkedIn to expand and enrich available vacancy data.
Based on this information, the platform generates tailored matching outputs in real time. For jobseekers, it identifies opportunities aligned with their profiles, preferences and aspirations. For employers, it recommends candidates who meet the specific requirements of the vacancies that they post. The matching results can inform all other pillars of the platform: skills identification (e.g. by revealing recurring skill gaps), vacancy management (e.g. by highlighting difficult-to-fill roles), and career recommendations (e.g. when an immediate match is not possible). The matching results can also feed in the original inputs through feedback loops for continuous improvement.
Individual-level information feeds directly into the skills identification pillar of functionalities. The underlying assessments provide evidence of each individual’s strengths and weaknesses. The results flow into both matching and career recommendations. The rationale behind these flows is that more standardised and validated skill profiles, increase the accuracy and quality of immediate matching and support the generation of tailored career pathways. These results are then fed back into the user profile, enriching and updating it over time through a feedback loop.
The pillar that includes the vacancy management related functionalities processes information from the business side, hereby improving job postings. This structured vacancy information feeds directly into matching, ensuring that job requirements can be accurately compared with candidate profiles. It also supports career recommendations, allowing the system to suggest opportunities and career pathways aligned with real job requirements.
Simplified indicative conceptual overview outlining the key functions of the platform and their interconnections
The figure illustrates the five main functional pillars of the platform and the key flows through which inputs from individuals and employers are transformed into matching and career recommendation outputs (green, blue and orange arrows). Feedback loops across components ensure continuous refinement and improvement of the platform (grey arrows). Labour market intelligence operates as a cross-cutting layer that connects to all components, providing market insights while continuously learning from system outputs through bidirectional feedback flows. However, for clarity, not all interactions with this pillar are shown.
Source: Authors’ elaboration.
The career recommendations pillar of functionalities can receive inputs from the labour demand and supply side as well as from all pillars (skills identification, vacancy management and matching). By analysing all the different inputs, it suggests realistic career pathways, training opportunities, and next steps. As users act upon these recommendations, their profiles evolve, feeding back into the system and further improving future matches and recommendations.
The final pillar of functionalities, the so-called labour market intelligence, serves as the cornerstone of this matching and career recommendations platform. It leverages both real-time and historical data, including job postings, user profiles, matching and other outcomes, as well as broader labour market data including at the national, regional and sectoral level. By processing this information, it generates insights on labour demand, emerging sectors and occupations, skills needs and salary trends. These insights ensure that the platform remains aligned with actual market conditions rather than relying on static or self-reported data. They also enable better, more forward-looking matching and career guidance decisions and sets realistic expectations. Insights can be differentiated and tailored depending on the users – whether jobseekers, employers, counsellors, social workers and training providers. This pillar can also inform evidence‑based policymaking, as it enables better targeting, the design of more effective interventions and therefore more efficient resource allocation. These insights can be visualised through interactive dashboards and tailored outputs, making them accessible and actionable for both policymakers and different types of end users. This final pillar interacts with all components of the platform through bidirectional flows. On the one hand, it informs each function while on the other it continuously learns from their own results outputs via feedback loops. For clarity and readability, not all flows related to labour market intelligence are depicted in the figure.
5.4.3. The new AI job matching and career recommendations platform should be able to interact with and support various users
A new AI-driven matching and career recommendations platform can generate substantial benefits to multiple user categories and types. In practical terms, these users may encompass all individuals and organisations that either engage directly with the platform or are influenced by its operation.
Potential users can be grouped into two primary categories, based on how they engage with the platform:
1. Direct users, who interact with the platform to achieve their own goals (e.g. jobseekers and employers).
2. Indirect users, who may not interact with the platform directly but are affected by or benefit from its operation (e.g. policymakers).
Within each category, the following breakdown provides deeper segmentation by user role.
Direct users:
MIS beneficiaries: Individuals receiving the GMI (Greece) / RIS (Belgium) who access the platform to connect with job opportunities.
Unemployed: Any unemployed individual who uses the platform to search for job opportunities, receive information or benefit from its different functionalities.
Employees: Individuals currently employed who wish to explore new job opportunities, change employers, or advance in their careers by using the platform’s features.
Employers (and hiring managers): Company representatives posting vacancies, connecting to and reviewing candidates and benefiting from the platform’s various functionalities. For typically larger firms, access may be granted to internal HR units/teams/specialists managing recruitment.
Firms offering internships, apprenticeships and work-based learning opportunities: Representatives of companies who express interest and offer for internship, apprenticeship and work-based learning placements.
Job counsellors: Professionals at DYPA (Greece) / Le Forem, VDAB, Actiris and ADG (Belgium) who support jobseekers by offering counselling, guidance, and personalised recommendations on available ALMPs.
Employer counsellors: Professionals at DYPA (Greece) / Le Forem, VDAB, Actiris and ADG (Belgium) who engage with employers to understand their recruitment needs, secure positions for jobs and ALMPs, as well as to facilitate effective matches between candidates and vacancies.
Social workers: Professionals in the Community Centres (Greece)/ CPAS/OCMW (Belgium) who support MIS beneficiaries and other vulnerable individuals on their path towards self-sufficiency.
Training and certification providers: Organisations delivering training, reskilling and upskilling programmes, courses and certifications and use the platform to promote their programmes, connect with potential learners or understand labour market needs.
Indirect users:
National Ministries responsible for social and employment policies: In Greece, this includes the Ministry of Social Cohesion and Family – responsible for policies related to social cohesion, family welfare, and poverty reduction – and in particular the Directorate for the Fight Against Poverty, which oversees the management and supervision of the MIS. It also includes the Ministry of Labour and Social Security, which is responsible for employment policy, labour market regulation, social insurance, and social security policies. In Belgium, responsibilities are divided across federal and regional levels. At the federal level, social protection and labour regulation fall under: the Federal Public Service Social Security (FPS SS), responsible for social security benefits, disability, health insurance, and oversight of social-protection schemes; the SPP IS, responsible for the Minimum Income Scheme (RIS), social inclusion, and supervision of CPAS/OCMW and the FPS Employment, Labour and Social Dialogue, which regulates labour law, working conditions, and labour market governance. At the regional level, ministries supervise employment policy, training and activation services within their jurisdictions (Flanders, Wallonia, and Brussels).
Specialised policy and analytical units: Such as Greece’s Unit of Experts in Employment, Social Insurance, Welfare and Social Affairs (MEKY)11 and the CBSS in Belgium. These groups support policy development, co‑ordinate labour market diagnostics, and use platform insights to guide evidence‑based decision making.
PES: Covers DYPA in Greece and the regional PES in Belgium (VDAB, FOREM, Actiris and ADG). Multiple departments and units within the PES may benefit from the platform’s insights and functionalities. These may include the central management team, strategic planning, employer services, training, vulnerable groups, data and analytics, finance and communications.
Heads of local employment offices: Local directors of DYPA offices (KPAs and EKOs) in Greece and regional/local PES branches in Belgium who are responsible for overseeing service delivery, staff co‑ordination, and local operational strategies. They may rely on the platform’s dashboards and reports to plan resources, monitor performance, and target services more effectively.
Local social-welfare institutions: Heads of Community Centres in Greece and CPAS/OCMW in Belgium. These actors are responsible for co‑ordinating social welfare services and implementing the MIS scheme at the local level.
IT developers/External contractors: Technical teams and external service providers involved in maintaining, updating, integrating, or expanding the platform’s digital infrastructure, ensuring functionality, security, and interoperability with other systems.
Evaluators: Independent experts or institutional bodies responsible for assessing the effectiveness, impact, and quality of the platform, including its matching algorithms, user experience, and policy relevance. This could be done by analysing aggregated or anonymised information, without access to any personalised data (partial and controlled access granted by the platform’s managing authorities).
Community Organisations / NGOs: Civil-society groups supporting vulnerable populations, jobseekers, or MIS beneficiaries. They may use the platform’s aggregated or anonymised data to guide their services and collaborate with the PES or social-welfare institutions (partial and controlled access granted by the platform’s managing authorities).
Research Institutes/Researchers, analysts, and public-policy experts: Experts who may use anonymised data, platform insights, and evaluation results to examine labour market trends, assess matching methodologies, or support evidence‑based policymaking (partial and controlled access granted by the platform’s managing authorities).
This section discusses the proposed options for designing and implementing an AI-powered matching and career recommendations platform for MIS beneficiaries and potentially beyond this very specific target group. As the situation in Greece and Belgium differs substantially, the options are analysed separately for each case.
5.5.1. Greece can either build an entirely new platform or extend an existing one
For developing a matching platform enriched with AI-enabled functionalities in Greece, two options can be considered, starting from the fact that a matching tool is already available at DYPA.
Option 1: Leverage the existing matching platform of DYPA
The first option is to extend DYPA’s matching platform, currently operating within a testing environment. The starting point should be to examine whether upgrading the existing infrastructure is technically and operationally feasible before committing to a full rebuild. This would require a comprehensive feasibility and interoperability assessment to evaluate whether the current DYPA platform:
Can technically support new AI modules and advanced features (described in detail under Section 5.4.1 of this report) such as advanced matching features, vacancy management related functions, counsellor and social workers dashboards, skills profiling, and labour market analytics.
Can be securely integrated with public systems such as the MIS platform and other registries including DYPA’s various databases, ERGANI as well as external platforms such as LinkedIn and support secure data flows across ministries and organisations.
Has the necessary architecture for gradual scalability and can accommodate multi‑agency, role‑based access.
Meets data governance and security requirements under GDPR regulations and the EU AI Act.12
Conducting this analysis first will help avoid redundant investments, duplicating systems, legal and compliance risks, the creation of tools that cannot be implemented or cannot be used effectively by frontline staff and overly long development times.
Should the assessment findings show that the DYPA platform can be adapted, upgrading it would likely be less costly, less disruptive, and consistent with current operations and processes. This means the existing platform will be enhanced and expanded with additional AI functionalities and integrated with multiple data sources, eventually operating as the single unified national matching platform for all users.13
Option 2: Build a new platform from scratch
If the feasibility study shows that the necessary upgrades cannot support the country’s strategic needs for matching MIS beneficiaries and other vulnerable groups, then developing a completely new platform becomes the appropriate alternative. If constraints, including funding, capacity, institutional co‑ordination and time pressure, make it unrealistic to incorporate all functionalities from the outset, then a prioritisation exercise will be necessary to guide gradual development and the addition of new features. This approach allows Greece to progress depending on technical feasibility and available resources, while ensuring that essential functionalities are in place from the beginning.
If this option is chosen, at the long-term, it would not be efficient to maintain two separate systems, namely the existing DYPA platform and a new one for MIS beneficiaries and vulnerable groups. The new platform should eventually replace the existing DYPA system and operate as a single unified national matching platform serving all direct user groups, including MIS beneficiaries and vulnerable jobseekers, the unemployed individuals in general, employees, employers, as well as job and employer counsellors and social workers.
Figure 5.6 presents the two options at a high level, starting from the premise that a matching tool is already in place.
The figure outlines two strategic options for developing an AI-enabled matching platform in Greece. The first option is to build on DYPA’s existing platform, provided that a feasibility and interoperability assessment confirms it can support the required upgrades. If the assessment is negative, the strategy shifts to Option 2, namely developing a new platform from scratch
Note: In cases where funding or capacity constraints prevent the development of a new platform, selected functionalities can be prioritised.
Source: Authors’ elaboration.
A summary presenting the two main options is included in Table 5.1.
The development of an AI-enabled matching platform can follow two main options/scenarios.
Note: The first option is to upgrade DYPA’s existing system, provided a feasibility and interoperability assessment confirms it can support new AI functionalities and integrate with MIS and other registries. If the assessment indicates that such upgrades are not viable, the alternative is to develop a new platform from scratch, which would then replace the existing DYPA system and serve as a unified national platform for all users.
Source: Authors’ elaboration.
5.5.2. Belgium can consider developing a new unified matching platform or opt for a federally co‑ordinated umbrella platform connecting the existing regional platforms
Belgium’s labour market governance is highly decentralised. Each region already operates its own PES and digital tools, with VDAB being the most advanced in terms of digital innovation. Forem, Actiris and ADG (to a lesser extent) also maintain mature digital systems. The organisational implications for CPAS are extremely important, as far as the human and operational impacts are concerned. From a “technical” point of view, the first steps should be strengthening the digital maturity and integration of CPAS, as well as improving the quality and standardisation of CPAS data.
Given this institutional setup and the strong regional digital capabilities already in place (particularly in Flanders), the following two options can be considered:
Option 1: A new single unified national matching platform
This option would involve creating a new unified matching platform used across all regions and public services. Such a system would provide a single‑entry point for MIS beneficiaries and potentially later on for all jobseekers and employees, as well as for employers, eliminating the existing fragmentation across the different already existing platforms.
A new platform developed from scratch could:
Replace regional matching systems with one platform for the entire country.
Provide a unified interface for jobseekers and employers and unify all vacancy data and candidate profiles.
Integrate RIS, social inclusion and activation pathways through federal mechanisms.
Establish one national labour market database and a single governance structure.
This approach offers several advantages – at least at the theoretical level. It would enable full harmonisation of matching services across the country and support the creation of a coherent national labour market intelligence ecosystem. A single platform would also provide one access point for all jobseekers and employers, simplifying the user experience, and would make national monitoring and evaluation more straightforward.
However, there are substantial challenges associated with this option. Employment competences lie with the regions, imposing both legal and political constraints on full centralisation. Moreover, regional PES already operate advanced systems. Developing a new national platform would entail significant investment, complex co‑ordination, and long development timelines. Furthermore, during the transition period, it might duplicate existing systems, and adoption could be lower if regional platforms remain more advanced or more trusted by the end users.
Overall, while theoretically attractive, a fully centralised national platform built from scratch would require substantial political agreement, structural change, available funds and a proven feasibility case, making it a considerably challenging option under Belgium’s current institutional framework.
Option 2: A federally co‑ordinated matching platform integrating regional systems
Building a new platform from scratch at the federal level would likely be disproportionate, costly, and duplicative. A rather more viable path is creating an umbrella platform that integrates the existing regional platforms through a federally co‑ordinated interoperability layer, shared standards and common AI components, but only to the extent that the regions collectively agree to these elements, and while allowing regions to retain full autonomy over their operations and service delivery. In practical terms, this means that each region will keep its own digital platform. These systems, however, will be connected to an umbrella platform of the federal level, provided that the regions explicitly consent to such a connection. Jobseekers and employers would continue to use their regional PES interface, but the systems would “talk to each other” behind the scenes. Under this option, the federal level would support:
Common rules, data, interoperability and metadata standards, agreed upon by all regions.
Shared AI building blocks (e.g. skills taxonomy, matching logic, etc.) developed jointly and used only with regional approval.
A harmonised framework for AI ethics and compliance with the EU AI Act, co-created with and validated by the regional authorities.
A unified access layer for MIS beneficiaries or inter-regional mobility programmes, implemented only where the regions endorse such arrangements.
A joint governance structure involving the federal authorities and the regional PES and social services, with regions maintaining all existing responsibilities and their decision-making power.
A technical assessment would first need to confirm that the existing regional systems can be indeed connected through common standards, APIs, and data‑exchange mechanisms without disrupting current operations and services. It would also assess whether the existing platforms can support the required upgrades in performance, data quality, and interoperability.
This approach offers several advantages. It builds on the strong digital infrastructures already in place – thereby avoiding unnecessary duplication – while fully respecting regional competences, established practices, and autonomy. By linking systems rather than replacing them, it also improves the consistency of labour market data at the national level and strengthens inter-regional mobility, which today remains limited. Because it relies on existing infrastructure, this option would involve lower implementation and maintenance costs and could be deployed more quickly.
At the same time, the approach comes with certain challenges. It would require voluntary political agreement among the regions on data governance and common technical standards, as well as close co‑ordination across the separate regional platforms. Significant and sustained co‑ordination and oversight would also be required to ensure consistent application of AI ethics and compliance with GDRP regulations and the EU AI Act across all regions, based on mechanisms jointly approved by the regional authorities.14
Considering Belgium’s institutional structure and the advanced digital capacities already in place, an interoperable platform co‑ordinated at the federal level, subject to regional approval and without modifying regional responsibilities, seems to be the most realistic and efficient option to pursue. Such a model would allow Belgium to build on the strengths of existing regional systems while ensuring that key functionalities are aligned and compatible across the country, only where the regions choose to collaborate.
Figure 5.7 presents the aforementioned two options at a high level.
The figure outlines two strategic options for developing an AI-enabled matching platform in Belgium. The first option is to build a new unified matching platform from scratch, only if the regions collectively endorse this approach. If this option proves not viable, the strategy shifts to Option 2. This includes retaining the existing regional platforms and connect them through a federally co‑ordinated interoperability layer, provided that the regions agree, and a technical assessment confirms that such integration is feasible
Note: In cases where funding or capacity constraints prevent the full development of a new platform, selected functionalities can be prioritised.
Source: Authors’ elaboration.
A summary presenting these two options/scenarios is included in Table 5.2.
The development of an AI-enabled matching platform can follow two main options/scenarios
Note: The first option is to develop a new unified platform at the federal level only if the regions collectively endorse this approach. The second option is to maintain the existing regional platforms and connect them an umbrella platform of the federal level, provided that the regions explicitly consent to such a connection, and a technical assessment confirms that such integration is feasible.
Source: Authors’ elaboration.
Well defined governance structures are essential to ensure that AI-powered matching platforms operate securely, transparently, and in alignment with national (and regional in the case of Belgium) labour market policies. The respective governance models clarify who owns the system, who manages and administers it, who has access, and how responsibilities are divided across the different institutions involved.
5.6.1. The proposed governance model in Greece involves joint ownership across three agencies
Given the platform’s importance for MIS beneficiaries and other vulnerable groups, and its role in linking the MIS with activation policies and labour market integration through its third pillar, joint ownership between the Ministry of Social Cohesion and Family and the Ministry of Labour and Social Security and its implementation arm – DYPA – appears to be a well-aligned and suitable governance option. Table 5.3 presents the proposed governance model.
The proposed governance model outlines a co‑ordinated framework in which strategic leadership, operational management, IT development, data protection, scientific oversight and local implementation are clearly distributed across the competent ministries, DYPA and supporting institutions to ensure an effective, secure and inclusive platform for MIS beneficiaries and other vulnerable jobseekers
Source: Authors’ elaboration.
5.6.2. A shared governance model aligns with Belgium’s decentralised system, balancing a key role for federal agencies with the autonomy of regional actors
Given the decentralised nature of Belgium’s social assistance and employment system and the key role that both federal and regional actors play in supporting MIS beneficiaries, a shared governance model is well aligned with the country’s institutional reality. At federal level, SPP IS, FPS SS, FPS Employment, Labour and Social Dialogue are well placed to oversee the interoperability layer, co‑ordinate shared rules and common AI components. At the same time, all regional PES retain their full autonomy and responsibility for service delivery and for operating their own digital platforms.
The recommendation for a federal interoperability layer requires a clear legal framework stipulating which actors collect, modify, use and supervise data and who bears responsibility for automated decisions, particularly under the AI Act (high-risk systems).
Table 5.4 presents the proposed governance model. That model corresponds to Option 2 described in the previous section.
The proposed governance model reflects Belgium’s institutional structure and offers a coherent approach for developing a platform that serves MIS beneficiaries effectively while respecting regional competencies
Source: Authors’ elaboration.
Under this second-best scenario, the career development component would operate as an independent digital tool enhanced by AI, designed specifically to support career guidance, skills development, and informed career decision making. This approach would allow users to benefit from personalised career insights, training recommendations, and labour market intelligence even in the absence of a fully integrated matching platform.
Given that no similar tools currently exist for MIS beneficiaries in either country, this solution would need to build on emerging digital functionalities while requiring substantial further development and integration. This implies designing the underlying architecture, defining the relevant data sources, and ensuring interoperability with existing systems. Particular attention would also need to be paid to how the tool interacts with existing job matching and skills identification systems, so that career recommendations and labour market insights remain consistent with available employment opportunities.
This section explores this alternative scenario and outlines specific considerations for Greece and Belgium. Similarly to the case of a comprehensive matching platform, a standalone tool can be developed incrementally, starting with core functionalities and expanding over time through modular add-ons, conditional on several factors.
5.7.1. Greece can build a comprehensive career recommendation tool or embed it in DYPA’s existing matching platform
While Greece does not currently have a comprehensive career recommendation tool of the type envisaged here, relevant functionalities have started to emerge. To make the best possible use, this new tool should be able to interact with multiple systems such as DYPA’s systems, the MIS registry, employment and social security registries and even training providers’ systems or formal education system even if it operates as a separate digital environment. The development of such a tool would involve designing the technical architecture, defining the relevant data sources, and ensuring that the platform can access and process information in a structured and secure manner. Particular attention would also need to be given to governance arrangements, including defining the institutional responsibilities for the ownership, operation, and data management of the platform. Users should be able to access a separate environment that provides guidance, training pathways, and labour market insights.
The development of a comprehensive career recommendation tool would offer several advantages. On the positive side, building such a platform would allow for greater flexibility in terms of design. A dedicated architecture could be developed specifically for career guidance and progression functionalities, allowing these features to be fully integrated from the outset rather than being added to an existing system. It would also ensure connections with multiple sources, enabling a more integrated approach to career support. In addition, a newly designed system could rely on modern technological standards and scalable architecture, making it easier to expand functionalities over time and adapt to future needs or technological developments.
However, this approach wouldn’t come without challenges. Developing a comprehensive platform would require significant financial investment and a longer design and implementation timeline. It would also require co‑ordination across multiple institutions. In addition, users might need to navigate several different digital environments, as the career recommendation tool would operate separately from existing job matching systems.
An alternative approach would be to integrate the career development functionalities into the existing DYPA matching platform, provided that this is technically feasible. Such integration would allow users and counsellors to access both job matching and career development services within a single digital environment. This option corresponds to the first option discussed earlier, which proposed leveraging the existing DYPA matching platform following a dedicated feasibility and interoperability assessment. Conducting such an analysis at an early stage can help avoid redundant investments, the duplication of systems, legal or compliance risks, the development of tools that cannot be effectively used by frontline staff, or excessively long implementation timelines. If the assessment shows that the DYPA platform can be adapted to support these functionalities, upgrading the existing system would likely represent the most practical option. In this case, the platform could be gradually enhanced with additional AI-based functionalities and connected to multiple data sources, eventually evolving into a unified national platform supporting both job matching and career development services.
If constraints such as funding limitations, institutional capacity, co‑ordination challenges or time pressures make it unrealistic to incorporate all functionalities at once, a prioritisation exercise would be necessary to guide gradual development. This would allow Greece to introduce the most essential functionalities first and progressively add new features as technical feasibility and available resources allow.
Given the platform’s importance for MIS beneficiaries and other vulnerable groups, and its role in linking the MIS with activation policies and labour market integration through its third pillar, joint ownership between the Ministry of Social Cohesion and Family and the Ministry of Labour and Social Security and its implementation arm – DYPA – appears to be a well-aligned and suitable governance option. The proposed governance model would be the same with the one explained earlier (Table 5.3), while it would apply regardless of whether the career development functionalities are implemented as part of the DYPA’s existing matching platform or developed as a standalone tool hosted within DYPA.
5.7.2. Belgium can consider developing regional career recommendation tools which should ideally be co‑ordinated at the federal level
Given Belgium’s highly decentralised system, in an ideal scenario, the development of digital career recommendation tools could take place within a broader umbrella platform that connects existing regional systems through a federally co‑ordinated interoperability layer. In practice, this would mean that the regional PES would continue to operate their own platforms, while the interoperability layer would allow information to be exchanged across systems.
If this is not feasible, an alternative and more pragmatic solution would be to develop career recommendation tools at the regional level within each PES. In this scenario, each regional PES would integrate career development functionalities within its existing digital platforms and services. The key idea is that the career recommendation tools would be developed at regional level, but designed to interoperate with each other where necessary, rather than being managed through a single central platform (which would be considerably challenging under Belgium’s current institutional framework). The key enabling factor would be early agreement on common technical standards for data exchange and interoperability rules to ensure that the different tools remain compatible and capable of exchanging information when needed.
Under such an approach, each region would maintain full autonomy and control over its platform, operations and service delivery, jobseekers and employers would continue to use their regional PES interface, but the career recommendation tools would be able to “talk to each other” behind the scenes through agreed interoperability mechanisms. This interoperability could be limited to specific types of information related to the career development functionalities, rather than encompassing all data held by each PES. Such an approach would allow systems to exchange relevant information while fully respecting regional governance arrangements and institutional responsibilities. For example, the types of information that could reasonably be shared across PES systems include educational and training opportunities, short courses, vacancy data and broader labour market insights. At the same time, information related to ALMPs would remain within each PES. This is because access to and participation in such programmes are typically managed at the regional level and are linked to individuals registered with the respective PES. The required interoperability standards could be defined at the federal level in close co‑ordination with, and with the agreement of, all regions.
Such an interoperable system could be beneficial to users who may wish to move across regions or explore opportunities outside their area. Career recommendations and skills profiles could remain usable across systems, helping individuals navigate labour market opportunities more effectively regardless of the region and the PES in which they are registered.
This approach offers several advantages. First, it is well aligned with Belgium’s decentralised governance structure, allowing regions to retain full control over their employment services and digital platforms. Each PES can design and implement career development functionalities that reflect regional labour market conditions, institutional priorities and its existing digital systems. This approach also builds on the strong existing digital capabilities, avoiding the need to replace or duplicate existing systems. In addition, interoperability between the regional tools would allow certain types of information to be shared across systems, enabling users to benefit from broader labour market intelligence and potentially explore opportunities beyond their region. Another advantage of this approach is that it reduces the political and institutional complexity associated with creating a single national career recommendation platform. Rather than requiring a comprehensive redesign of governance arrangements, regions can develop their own tools while agreeing on a limited set of common standards that allow them to communicate. This allows flexibility in implementation, as regions can adopt new functionalities at their own pace depending on available resources, technical capacity and policy priorities.
At the same time, this approach also presents certain challenges. Ensuring interoperability between multiple regional systems requires early agreement on certain issues such as common technical standards, data formats and exchange protocols. Without such co‑ordination, there is a risk that the different tools evolve in ways that make future interoperability more difficult and complicated. Another limitation is that users may still experience some level of fragmentation if the different systems are not sufficiently aligned in terms of design, functionalities or user experience. While interoperability can allow systems to exchange certain information, it does not fully eliminate differences between regional platforms. As a result, individuals moving between regions or exploring opportunities outside their region may still need to connect and interact with multiple systems.
The proposed governance model would be the same with the one explained earlier (Table 5.4). Given the decentralised nature of Belgium’s social assistance and employment system and the key role that both federal and regional actors play in supporting MIS beneficiaries, a shared governance model is well aligned with the country’s institutional reality. At federal level, SPP IS, FPS SS, FPS Employment, Labour and Social Dialogue are well placed to oversee the interoperability layer, co‑ordinate shared rules and common AI components. At the same time, all regional PES retain their full autonomy and responsibility for developing and operating the new career recommendation tools as well as service delivery.
References
[1] Brioscú, A. et al. (2024), “A new dawn for public employment services: Service delivery in the age of artificial intelligence”, OECD Artificial Intelligence Papers, No. 19, OECD Publishing, Paris, https://doi.org/10.1787/5dc3eb8e-en.
[6] Broecke, S. (2023), “Artificial intelligence and labour market matching”, OECD Social, Employment and Migration Working Papers, No. 284, OECD Publishing, Paris, https://doi.org/10.1787/2b440821-en.
[2] Dromundo Mokrani, S., A. Lauringson and T. Xenogiani (2024), “The Role of Public Employment Services in Enhancing Labour Market Matching”, in Carcillo, S. and S. Scarpetta (eds.), Handbook on Labour Markets in Transition: Promoting Resilience in a World of Flux, Edward Elgar Publishing, https://doi.org/10.4337/9781839106958.00030.
[8] Köver, C. (2024), AMS erntet Hohn mit neuem KI-Chatbot (AMS faces ridicule with new AI chatbot), https://netzpolitik.org/2024/diskriminierung-ams-erntet-hohn-mit-neuem-ki-chatbot/ (accessed on 6 October 2025).
[5] Krishnan Somos, A. (2025), Berufsinfomat unleashed: Experiences, Challenges and Impact in Austria, https://www.oecd.org/content/dam/oecd/en/about/programmes/dg-reform/using-ai-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries/Concept-Note-and-Agenda-International-workshop-Digital-and-AI-powered-job-matching.pdf.
[3] TEM (2023), Job Market Finland - A meeting place for people and jobs, https://tem.fi/documents/1410877/177691663/THU%2BKokkola%2BJob%2BMarket%2BFinland.pdf/9d1a71b4-66c7-33f4-3c07-9eee27cf0498/THU%2BKokkola%2BJob%2BMarket%2BFinland.pdf?t=1698222604759.
[4] Vainio, M. (2025), Job Market Finland and Expert´s Job Market Finland: New customer information system of the Finnish PES, OECD, Paris, https://www.oecd.org/content/dam/oecd/en/about/programmes/dg-reform/using-ai-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries/Job-Market-Finland-and-Experts-Job-Market-Finland.pdf.
[7] Yeo, S. (2025), Integrating the most vulnerable jobseekers: The example of Korea’s one-stop shops, OECD, Paris, https://www.oecd.org/content/dam/oecd/en/about/programmes/dg-reform/using-ai-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries/Integrating-the-most-vulnerable-jobseekers-The-examples-of-Korea-one-stop-shops.pdf.
← 1. The proposed concept recommending long-term career pathways including ALMP support draws on previous project activities and outputs. These include the fact-finding meetings in both countries, the identification of promising data that could potentially be used to develop digital solutions matching MIS beneficiaries with labour demand and recommending career pathways, the findings of the data analysis on MIS beneficiaries and employers in both countries as well as analysis of long-term career pathways of MIS beneficiaries, including the role of key ALMPs to facilitate their career progression. It is also informed by the experience of existing tools in EU and OECD countries. Some of these were presented during the two international knowledge‑sharing workshops held in May and November 2025, as well as during the study visit in Sweden in February 2026 and its corresponding note. Additional examples have been provided, in the note which summarised lessons learnt on matching MIS beneficiaries to labour demand and recommending career pathways.
← 2. The detailed technical and functional specifications of the matching tool fall beyond the scope of this project and this specific chapter.
← 3. This chapter focusses on job-matching platforms used by governments and public agencies. It does not cover private‑sector platforms or online marketplaces that connect freelance workers with clients, such as those operating in the gig economy.
← 4. A detailed description of good practices on job matching and career recommendations tools is included in a dedicated report produced under this project and can be accessed on the project’s webpage (Output 3).
← 5. A detailed table with all the functionalities of the AI-powered matching and career recommendation platform has been developed as part of the project and can be accessed on the project webpage. The functionalities are organised by pillar and classified by their priority level (as essential, optional, or nice‑to-have). For each functionality, a brief description and its intended purpose are also provided.
← 6. The processing of health-related data is subject to strict conditions under the GDPR and requires appropriate legal bases, safeguards, and, where applicable, explicit consent.
← 7. Berufsinfomat has been subject to criticism for producing socio-economic biases and hallucinating information (Köver, 2024[8]). AMS tries to mitigate these impacts through making clear the tool is not a replacement for personal appointments with counsellors, and the bot itself directs users towards in-person services.
← 8. Recommendations may also extend beyond immediate job placement to include ALMPs, relevant training and courses to address skills gaps and improve employability, as well as sustainable employment opportunities. These broader aspects, are examined in detail in Chapter 6 which discusses digital solutions enhanced by AI to recommend career pathways for MIS beneficiaries, including ALMP support.
← 9. Early warning systems are data-driven tools designed to identify individuals who may be at risk of experiencing negative outcomes and hence require timely and targeted interventions. In education, EWS are often used to flag students who may be at risk of dropping out from school based on indicators such as attendance, grades, or behaviour. In PES, similar systems help identify jobseekers who are likely to become long-term unemployed by analysing factors such as work history, skills, age, benefit duration, or engagement patterns.
← 10. In Greece, the establishment of an individual learning accounts system is foreseen by Law 4921/2022; however, the system is not yet fully operational. Belgium has recently experimented with such a system, but it was eventually abolished in January 2026.
← 11. A specialised scientific and advisory unit operating under the remit of the Greek Ministry of Labour and Social Security providing scientific support and policy formulation in the fields of employment, social insurance, welfare and social affairs. It also supervises and co‑ordinates the Labour Market Diagnostic Mechanism (MDAAE), an online labour market monitoring tool that combines big data analysis with job demand and skills data offering solutions for more effective employment services and labour market policies.
← 12. A brief reference to the legal considerations for making relevant data available for digital matching and career recommendations solutions is included in Chapter 6.
← 13. Key considerations regarding the data sources and the data required for the operation of the matching and career recommendations platform is included in Chapter 6.
← 14. A brief reference to the legal considerations for making relevant data available for digital matching and career recommendations solutions is included in Chapter 6.