Bulgaria’s National Employment Agency (NEA) is currently working to digitalise its processes and services, including through the adoption of new tools. This chapter provides an assessment of the situation along with recommendations regarding the development and deployment of digital solutions to streamline the agency’s operations and services. Key goals include 24/7 service accessibility, improved user experience, and evidence‑based decision-making. Special attention is given to tools that support jobseekers, such as counselling and job-search platforms, skills assessment and profiling tools, and digital solutions that enhance collaboration with employers. Digital and AI-based tools can also significantly facilitate internal operations, by reducing administrative burdens and enabling staff to focus on their core responsibilities. The chapter also discusses such tools that support knowledge generation and analytics. Finally, it touches briefly upon key actions and considerations, including those related to potential challenges and risks to the NEA’s successful digital and AI transformation.
Optimising Processes and Services at Bulgaria’s National Employment Agency
Abstract
The previous chapters have highlighted several areas where the NEA’s operational processes can be improved or strengthened. A common element across these processes and functions is the significant potential for digitalisation and the adoption of digital and Artificial Intelligence (AI)-based technologies. This discussion is particularly relevant for the NEA, as it is currently in the process of upgrading its digital infrastructure. The gradual adoption of digital and AI-driven solutions would eventually allow the NEA to improve the overall efficiency and effectiveness of its internal operations and administrative processes, release significant burden from its frontline staff while enhancing the support it provides to jobseekers, workers and employers.1 Against this backdrop, the present chapter discusses such tools, drawing from good practices implemented by other EU and OECD PES, by key work stream areas.
Section 6.2 briefly discusses the current situation outlining the strengths and the key challenges the NEA is currently facing when it comes to the current digital infrastructure that supports key operational processes and services. These are also summarised in Table 6.1.
Source: Authors’ compilation.
Section 6.3 outlines key recommendations for the NEA to shift from isolated digital improvements to a unified, strategic transformation, building on a strong foundation of high-quality, rich data. All stakeholders involved in the discussions as part of this project emphasised that digitalisation must be a top priority, at the heart of the NEA’s new operating model. However, it also remains one of the most significant challenges. To ensure success and unlock the full potential of digital transformation, two critical requirements should hold: early and active involvement of staff in the design and implementation of digital tools and robust, ongoing training for all employees – both new and existing – as well as across all roles and levels.
This section is structured around the key pillars for the NEA’s digital transformation (Figure 6.1). It begins with a discussion around the need for the development of a comprehensive digitalisation strategy aimed at modernising the entire organisation. It then explores four core operational areas of the NEA’s work that would benefit substantially from the adoption of modern digital and AI-based tools. These areas include:
1. Digital tools designed to support jobseekers in their journey toward (re)employment – tools that are commonly used across EU and OECD PES. These include system for online registration, tools to enhance counselling and career guidance, personalised job search, online training and certification, job search and matching and vacancy listings (from the side of jobseekers), and skills assessments. A special reference is made on profiling, drawing on the experience of PES that have deployed advanced statistical and AI-based profiling systems.
2. Digital tools to enhance employer engagement, including improved matching services that facilitate connections between employers and potential candidates.
3. Digital tools to support administrative and back-office processes within the NEA.
4. Digital tools to support knowledge generation and analytics, such as data management platforms and Labour Market Information Systems (LMIS) including those with forecasting capabilities.
The chapter ends with a discussion of the importance of establishing robust monitoring and evaluation frameworks of these new digital solutions to ensure optimal performance and continuous improvement.
Source: Authors compilation.
Bulgaria’s Recovery and Resilience Plan (RRP) includes dedicated investment in the modernisation of the NEA (European Commission, 2022[1]). The objective is to enhance the efficiency and quality of the NEA’s services through the development and improvement of its IT infrastructure. The main planned digital developments include:
An updated National Database: The National Database is the primary information system of the NEA. It stores data related to the front-office activities, including the registration and mediation of clients. The NEA currently depends heavily on manual, paper-based processes, partly due to the lack of automated data transfer between the National Database (NDB) and other digital systems and registers.2 The updates to the National Database include enhancements in data management practices and establishments of data-sharing protocols with other public institutions such as the Ministry of Education, to verify enrolment in compulsory education. It is envisaged that components of the new National Database will involve the use of AI, including to assess and recommend suitable measures and services suitable for a given client.3
A virtual labour office, including a matching platform: The NEA is currently in the process of developing a new digital channel for clients – both jobseekers and employers – make its services accessible through a virtual labour office. The digital platform will have a variety of functions, some of which may be enabled by AI, including a virtual assistant and a matching platform to match jobseekers with relevant job vacancies posted by employers.
A professional compass application: This will likely be an orientation tool to recommend suitable training and job vacancies by profession and region to jobseekers.
The NEA has assigned much of this work to an external contractor, who is currently working on the updating of the National Database and the development of the matching platform.
Despite these encouraging developments, challenges remain. Current digital initiatives are often fragmented and implemented in an ad hoc manner. Some of the tools currently deployed are rather simplistic and outdated, the most prominent example is the tool for profiling/phasing jobseekers. The absence of integrated, dedicated tools means that many processes – for instance skills profiling and assessments – still rely on basic questionnaires, limiting their effectiveness. Furthermore, limited interoperability between internal and external databases increases the risk of human error and places a significant administrative burden on the NEA staff. The existing public-facing platforms, including the NEA’s website, are also in need of improvement, as they tend to be difficult to navigate and do not fully meet the expectations and needs of modern users. In addition, the lack of advanced labour market analysis and forecasting capabilities constrains the agency’s ability to anticipate future trends and support proactive, evidence‑informed policy planning. The recommendations for the development of new tools presented in the following section build on ongoing initiatives, to ensure their effective deployment and maximise their impact, while also addressing persistent challenges that dedicated digital solutions could help alleviate.
6.3.1. Underpinning digitisation through a comprehensive and dedicated agency-wide strategy
Building on examples from other PES (e.g. PES in Austria, Greece, Latvia, Norway, Sweden) the NEA’s digitalisation efforts should be guided by a comprehensive, dedicated digitalisation strategy to digitally upgrade the entire organisation. Currently, about a quarter of PES in OECD and EU countries have dedicated digitalisation strategies in place, with a further quarter opting to embed digital transformation into the overall PES strategy (Brioscú et al., 2024[2]). The NEA could get inspiration from these countries in developing its own digitalisation strategy. Relevant examples include the cross-disciplinary approach taken by the Norwegian PES to develop a dedicated strategy for responsible AI use, which helped in generating buy-in and ownership at all levels of the PES (Box 6.1) and the digitalisation strategy for the Latvian PES developed as part of an OECD/European Commission project.4
In many instances, PES strategies are closely aligned with broader national digital transformation plans, ensuring consistency and leveraging synergies across sectors and public agencies. Digitalisation strategies of PES typically articulate the long-term vision and key goals of the agency, such as improving service delivery, enhancing operational efficiency, streamlining operations and enhancing data-driven decision-making. They also outline gaps, inefficiencies, and areas where technology can make the greatest impact based on an assessment of the current situation. These strategies also define the tools and technologies to be adopted, often along with a roadmap detailing key milestones, timelines, and deliverables in line with the agency’s human and financial capacity, strategic priorities and feasibility considerations. They also establish clear guiding principles for implementation, including data security, data protection and other legal considerations, interoperability between internal and external systems, cross-agency collaboration and stakeholder engagement. Some digitalisation strategies also incorporate a robust monitoring and evaluation framework to measure and assess the performance of the new tools and platforms and identify areas for improvement for tools that don’t serve their intended purpose. Strategies could include plans for sustainability and scalability, focusing on maintaining and expanding digital solutions over time while ensuring they remain adaptable to technological advancements and the evolving needs of the labour market.
The planned introduction of several new tools and systems including with an AI component – along with wider digitalisation plans – will represent a significant change for the NEA; impacting its internal processes, the tasks and work of staff, how clients engage with the NEA and the services provided. Establishing a high-level strategy for this transformation would be useful in setting the objectives and vision for the NEA in its AI use. This would not only enhance transparency, but also promote awareness and understanding across NEA staff.
The Norwegian PES (NAV) has been using AI since 2017, following the creation of AI Lab to explore how AI could aid various aspects of its services and processes. With several AI tools in place and further ones in the exploratory phase, a strategy to set the objectives for the use of AI was launched in 2024. The AI strategy is structured around three components: ambitions, enablers and principles.
Ambitions: NAV’s aim is to use AI to contribute to: i) getting more people into employment, ii) improving benefit decisions and enhancing transparency about these; ii) reducing the rate of people not in education, employment or training (NEET), and iv) facilitating processes for NAV employees.
Enablers: The strategy sets out a variety of key enablers that impact the ability of the PES to achieve its AI’s ambitions. These include the need for clearly defined roles and accountability high-quality data, competence and experience with AI at all levels of the organisation, clear and efficient processes and frameworks for AI initiatives, easy-to‑use technology and the involvement of expertise from various disciplines.
Principles: Finally, the strategy establishes the six principles to ensure responsible use of AI within the PES: i) assess consequences, ii) privacy, iii) fairness, iv) explainability, v) security, and vii) transparency.
The development of the strategy began in the AI Lab before being brought to senior management. Consultations and discussions took place across the organisation to feed into the draft strategy, gathering inputs from cross-disciplinary expertise, including from IT, employment services and benefit departments. The plan is for the strategy to evolve over time with regular updates – allowing for lessons and experiences to influence the future direction.
Source: Presentation delivered by officials from the Norwegian PES during the international workshop on the use of AI by PES organised on 28 June 2024 as part of this technical assistance project.
6.3.2. Supporting the NEA’s work with jobseekers
Modern digital tools to identify the skills and needs of jobseekers and to guide and activate these
Providing information and resolving queries
The first step for a jobseeker is often to gather essential information about the PES, including how to register, the range of supports and services offered, and any eligibility criteria. The NEA’s website – and potentially a mobile app – should serve as the primary point of contact. As such, it must be easily accessible, user-friendly, and provide clear and comprehensive information.
AI-powered tools, such as chatbots or virtual assistants, can also be used to deliver information efficiently. Unlike traditional versions, AI chatbots can understand and respond to complex queries and continuously improve their responses through ongoing interactions with users. These tools are typically accessible to both jobseekers and employers, while also helping reduce staff workload within the PES. The matching platform currently being developed by the NEA is expected to include a virtual assistant, but the NEA can benefit from good practices of other PES that have been successfully using such tools for several years.
Since its launch in 2018, Frida – the chatbot used by Norway’s PES – has successfully resolved 80% of inquiries without human intervention, enabling the PES to make better use of its resources. Iceland’s PES launched its chatbot, Vinný, in July 2020 during the COVID‑19 pandemic. Vinný now assists with about one‑third of all queries, offering information on a range of services and benefits for PES clients. The PES in Finland employs two external chatbots: Tarmo, which helps with frequently asked questions, and Aino, which assists international workers interested in working or studying in Finland (and employers seeking foreign talent). Lithuania and Portugal introduced their AI-powered chatbots at the end of 2023 and in November 2024, respectively, to handle queries from jobseekers and employers about key PES measures. Greece has been using its AI chatbot, Daphne, since 2022 with plans to upgrade to a more advanced version that will serve as a digital employment counsellor. Other PES, such as those in France and Luxembourg, have implemented smart assistants to support staff in managing and responding to client queries. These tools enhance efficiency while preserving a human touch, as they require staff sign-off -allowing them to review, refine, and personalise the suggested replies as needed (Brioscú et al., 2024[2]).
Deliver counselling, career management, job-search orientation and matching services
Counselling services at the NEA are still delivered exclusively in person. In most PES though, counselling services are also accessible remotely through phone or digital channels (e.g. in Greece and Slovenia). These channels include the online user interface or dedicated communication platforms. Remote counselling offers significant benefits, particularly for individuals with disabilities and other vulnerable groups who face substantial barriers to labour market integration but have difficulties in physical access to services. To maximise accessibility and effectiveness, some of these digital platforms incorporate features like multilingual support and sign language interpretation. In general, it is important that digital tools for PES incorporate a range of features to ensure accessibility for populations with specific needs, such as people with disabilities and elderly citizens. Such features may include for example screen reader compatibility, large font options, zoom functionality, mobile‑friendly interfaces, captions/subtitles for audio or video content, visual alerts instead of only sound notifications. User support and/or training can be provided to these groups to ensure they can use these tools effectively.
In some PES, counsellors have the flexibility to select the most suitable mode of counselling – whether face‑to-face, by phone, via video, or through digital channels – based on the jobseeker’s preferences and needs (e.g. in Denmark and Estonia). Counsellors delivering these services online should be provided with guidelines, capacity building workshops and training sessions, to familiarise themselves with this new mode of service delivery (European Commission, 2021[3]). Some platforms designed for online counselling are enriched with additional features, publicly accessible through PES websites, and freely available for jobseekers. Such features may be self-assessment questionnaires to explore personality traits, interests, and competencies, along with resources for advice and guidance.
A core task of PES counsellors is to provide tailored career guidance and personalised job-search support to jobseekers. These functions can be supported by interactive tools that can assist jobseekers in setting clear career goals, developing effective job search strategies, and tracking their progress. Additionally, virtual workshops or webinars on key topics such as career development and job search strategies can provide valuable insights and equip jobseekers with the necessary tools to succeed in their job search. Jobseeker community forums and peer support platforms can provide online spaces where jobseekers can connect, share experiences, ask questions, and offer mutual support. They can also connect with industry professionals and mentors. The NEA’s webpage could host such a forum, fostering a sense of community and helping individuals to jointly navigate their job search journey. The NEA should consider adopting an online counselling tool or platform with career guidance and job search support features, in order to make its services more appealing to jobseekers, by offering easily accessible and interactive resources. This will help clients to become more proactive in their job search efforts, while empowering them to make better-informed decisions regarding their career paths.
AI methods can also help the PES to predict the hiring probability of companies. A notable example in this respect is the so-called La Bonne Boîte system of the French PES. The tool is operational since 2018 and can predict a company’s likelihood of recruiting – even before job vacancies are officially posted – based on past hiring behaviour and company characteristics. A company’s hiring potential is rated out of five stars. Jobseekers can access the tool directly through a dedicated website, search for jobs by location, and focus their job-search efforts on companies most likely to hire within the next six months, making job search experience easier and more efficient (Brioscú et al., 2024[2]). A similar tool used by counsellors to identify employers most likely to recruit new staff is deployed by the PES in the Netherlands, supporting jobseekers in their search.
In addition to job search, matching and vacancy listings can be facilitated through platforms that automatically match jobseekers with relevant job opportunities based on their skills, experience, and preferences (see also next section on digital tools to support employers). These platforms can feature searchable databases of current job openings, tailored to specific regions or sectors, making it easier for jobseekers to find suitable positions. Virtual career fairs and networking events offer jobseekers the opportunity to interact with employers, learn about job opportunities, and participate in interviews. These events can also include workshops designed to help jobseekers build valuable professional connections. In addition, online job application assistance tools can guide jobseekers through the application process, offering resume‑building support, cover letter templates, and interview preparation. These resources help improving job applications, increasing jobseekers’ chances of success. Remote job interviewing tools enable jobseekers to practice interviews and participate in real-time interviews with employers. These platforms often include feedback and performance assessments, allowing jobseekers to refine their interview skills and improve their chances of securing a job. Given the unique needs and complex profiles of vulnerable jobseekers, many countries and PES have deployed tools to support job matching and career development for these groups. In this context, an ongoing OECD / European Commission project is assisting authorities in Belgium and Greece in enhancing their capacity to develop digital solutions to support the job matching and career pathways of vulnerable populations, particularly Minimum Income Scheme beneficiaries (OECD, 2024[4]).
AI also presents opportunities to enhance the targeting of services and measures to meet the individual needs of jobseekers. Such tools can take the form of recommender systems. However, use cases within OECD PES are largely limited to date. The PES in France has implemented two such AI tools to aid the targeting of measures, each with a different end user. The first is a decision-support tool for counsellors called Mon Assistant Personnel (My Personal Assistant), which analyses a jobseeker’s CV in order to produce recommendations for suitable supports and job opportunities (Chapuis, 2018[5]). The second is a personalised recommendations tool for jobseekers within the personal space of the PES online portal. It provides tailored suggestions on potential services and measures, including training courses, workshops and PES events.
Lastly, digital platforms can play a key role in making labour market information easily accessible to jobseekers. These platforms can offer up-to-date insights on labour market trends, industry demands, salary expectations, and high-growth sectors. They may also provide detailed reports on regional employment trends, occupation forecasts, and emerging job opportunities, thereby enabling jobseekers to make more informed career decisions. For example, the PES in Slovenia provides data on past vacancy trends and current demand patterns for specific occupations by region, offering a valuable reference for both jobseekers and policymakers.
Identifying the skills and needs of jobseekers
Skills assessment tools can help both jobseekers and job counsellors. For jobseekers, these tools help identifying their skill profiles, fostering greater self-awareness regarding their abilities and preferences. For counsellors, they offer valuable insights into clients’ strengths and areas for improvement, enabling them to provide more personalised and targeted support, ultimately improving the overall efficiency and quality of the counselling process (OECD, 2024[6]). PES across the EU and the OECD increasingly integrate advanced digital and AI-driven tools and methods to support these tasks, including to help jobseekers identifying the skills or competencies they possess (Brioscú et al., 2024[2]). Such tools can be used to evaluate a broad range of skills. For example, in Flanders, Belgium, the PES has developed the Jobbereik (Job Reach) tool which allows jobseekers to see how their competencies and skills correspond to those associated with potential occupations, using data from job vacancies (Broecke, 2023[7]; Brioscú et al., 2024[2]).
In some cases, PES conduct a screening exercise to evaluate whether a job seeker possesses the necessary skills for a particular set of jobs. In other cases, adopted tools have a broader scope and are designed to assist jobseekers in discovering their talents and strengths and counsellors to get to know their clients better. Some self-assessment tools, such as Germany’s MYSKILLS and the OECD’s Skills Profiling Tool, help individuals identify and understand their skills and competencies and can be completed either at home or in an employment office. Jobseekers may use the test outcomes in job applications to demonstrate their skills to potential employers. The counsellors also use the tests results to guide jobs and ALMPs recommendations. There are also tools designed to specifically assess certain skills or categories of skills, such as the European Digital Competence Framework (DigComp) and the French Pix, which assess digital skills. Some PES deploy skills assessment tools for certain groups of jobseekers such as migrants and refugees and other vulnerable jobseekers who may not possess formal qualifications and their skills and prior experience cannot be easily showcased (e.g. Austria). Based on the results of these tools, job counsellors develop a labour market development plan that outlines intervention measures. The results from skills assessment tools can also serve as valuable input for statistical profiling tools, which calculate a jobseeker’s distance from the labour market (see next sub-section on profiling tools).
Offering online courses and certification programmes
PES also offer training through remote and asynchronous methods, often in collaboration with external providers. Online courses and certification programmes can cover a wide range of areas, including digital skills, job-specific training, and language courses. In some cases, such as in Greece, the PES partners not only with private training providers, but also with large technology and digital education companies such as Google, Coursera, Cisco, Amazon, and Microsoft to deliver high-quality upskilling courses to jobseekers (OECD, 2022[8]). Additional resources, including virtual workshops, training sessions, and videos, focus on developing soft skills aimed at improving workplace readiness. These resources cover essential skills such as time management, teamwork, and communication, helping jobseekers better prepare for the demands of the workplace. These are areas and resources that the NEA could consider, to enhance its services and support for jobseekers.
Source: Authors compilation.
A new profiling tool would enhance jobseeker assessment and improve service targeting while ensuring transparency, accuracy, and usability
As discussed in Chapter 3, the NEA currently employs an outdated and rather simplistic profiling tool to segment jobseekers into two main groups, depending on their job readiness. Counsellors can confirm or change the phase recommended by the tool and can also choose a third phase. The model only relies on information that is entered as categorial variables such as age, education level etc. Data entered as free text, including information gathered during interviews with the job counsellor, is not taken into account by the profiling tool. The NEA, in collaboration with an external contractor, is currently developing a new digital tool aimed at enhancing job matching. As part of this digitalisation initiative, the NEA is also considering the development of a modern profiling tool to improve predictive accuracy and facilitate the work of job counsellors, enabling more effective and tailored support for jobseekers and ultimately more efficient use of the agency’s resources.
The application of statistical profiling has significantly expanded in recent years among PES across the OECD (e.g. Flanders and Wallonia in Belgium, Estonia, Ireland, the Netherlands, Portugal, etc.). Statistical profiling tools are able to predict the likelihood of jobseekers becoming long-term unemployed5 (or re‑employment success) and distinguish those who are easy to place from those who are more challenging to place.6 They can also identify the key factors that may hinder jobseekers’ return to work. Some tools take a step further, by estimating a jobseeker’s likelihood of achieving specific career goals, such as securing a job in a particular occupation or sector. They can also offer tailored recommendations, including vacancies in related occupations or locations, and activation measures to enhance employability, such as the profiling tool deployed by the French PES.
The efficiency and usefulness of statistical profiling models depend on their predictive accuracy. This can be high when comprehensive underlying linked administrative datasets are used. In addition to socio-economic information, statistical models can use various types of data inputs including education, employment history, skills – both hard and soft (even though the latter is not systematically collected by PES), barriers to employment, unemployment duration, recipiency of benefits, job-search behaviour, as well as labour market information at the regional or local level. Some profiling models, particularly AI-driven ones, integrate data from additional sources, including online questionnaires or/and individual meetings with jobseekers, local labour market conditions, and even jobseekers’ activity on the PES website by tracking the number of clicks/hits (as is the case in the profiling tool used by the PES in Flanders, Belgium). Some PES complement profiling tools with other digital resources, such as tools that can generate labour market information – including at the local level – and skills assessment tools, to enhance accuracy and reduce bias. In most cases, the results suggested by the profiling tools work in a non-prescriptive manner. While they provide PES counsellors with valid information regarding the jobseeker’s distance from the labour market, and in combination with their own expertise they then make informed recommendations and support each jobseeker in the best possible way.
While this section focuses predominantly on aspects related to the tools per se, other important elements should be considered. Key considerations are how the new tool integrates into the agency’s existing processes, and how processes should be adjusted. This includes evaluating whether certain jobseeker meetings should be modified or eliminated, determining the most effective support channels – whether online, by phone, or in person – and deciding on the optimal frequency of jobseekers’ interactions with job counsellors. For example, in the Netherlands, only jobseekers identified as having a high risk of long-term unemployment receive a face‑to-face interview with a job counsellor early on, while those with lower risk are offered computerised services and are only invited to meet with a counsellor if they remain unemployed for six months. Profiling tools can also play a role in guiding resource allocation and service delivery. They can help determine the allocation of resources to different groups of jobseekers, and also the allocation of jobseekers to the most suitable services and interventions. Additionally, the PES can use these tools to monitor service effectiveness, gathering data to refine and improve interventions over time. Finally, the level of discretion granted to job counsellors in interpreting profiling results and supporting jobseekers is an important consideration (see section 3.5.2 on profiling in Chapter 3 for a more detailed analysis of this aspect).
Key suggestions based on the experiences of other countries in developing and adopting new profiling tools include (Figure 6.3):
1. To enable more accurate predictions and therefore maximise the effectiveness of a new statistical profiling tool, the NEA must prioritise data quality and richness (see Box on data preparation for a potential new profiling tool in Chapter 3). Ensuring interoperability with other relevant databases – both internal and external – is crucial for a comprehensive understanding of jobseekers’ needs and labour market trends.
2. The new profiling tool should be used in combination with other resources. Such tools may include skills assessment and tools that can generate labour market information and forecasting, along with other relevant contextual information. A holistic system will reduce the risk of bias, improve the segmentation of jobseekers, enhance service targeting, and support personalised career guidance across different offices and regions.
3. The profiling tool should support and facilitate the work of job counsellors, not replace them. By streamlining jobseeker assessment, it can reduce administrative burden, allowing counsellors to focus on their counselling/advisory role. Even powerful statistical tools are only a supplement to the counsellors’ experience and qualitative assessments, as many key factors are hard to capture in data or may be inaccessible due to privacy concerns. Therefore, job counsellors can combine the system results with their expertise to assign jobseekers to the most appropriate support stream.
4. To enhance its usefulness, the profiling tool should offer more than a numerical employability score. It should provide clear explanations of the factors affecting a jobseeker’s score as well as tailored recommendations for activation measures, and suggestions for job vacancies in related occupations and industries or different locations.
5. Throughout the development and implementation of a new profiling tool, NEA job counsellors must be actively involved, as they play a critical role in guiding and supporting jobseekers. Equally important is the involvement of jobseekers themselves, who are the ultimate end users of the tool. By engaging both counsellors and jobseekers in the process, the NEA can ensure the tool meets the practical needs of users and is designed in a way that enhances the user experience. Clear and effective communication about the tool’s benefits is also essential to foster understanding and trust among all stakeholders. Counsellors’ insights can help align the tool with real-world counselling practices, while jobseekers’ feedback ensures it is accessible, understandable, relevant, and effective in supporting their job search. The Austrian PES pilot of a new statistical profiling model in 2019 benefitted from securing stakeholder support to mitigate concerns and ensure a smooth and effective implementation. Proactively engaging key stakeholders helps mitigate issues related to transparency, data privacy, and potential bias, fostering trust and acceptance of the tool.
6. To maximise the tool’s effectiveness, counsellors should receive thorough training on how to interpret profiling results, use them in conjunction with the results of other tools and resources and translate them into actionable support strategies.
7. The labour market is constantly evolving, which makes it essential for statistical and AI-based profiling models to remain accurate and relevant over time. The NEA should make sure to regularly retrain the underlying prediction model using the latest labour market data. Continuous improvement will help maintain alignment with evolving trends, ensuring jobseekers receive up-to-date and effective guidance.
8. Public trust is essential for the success of AI-driven profiling tools. The NEA should develop from the outset a strong ethical framework to address concerns about potential discrimination, lack of transparency, and data misuse.
9. Prioritising data privacy, security, and compliance with national and EU data protection regulations will not only protect users but also mitigate potential concerns, thereby reinforcing confidence in the system. These measures should be communicated upfront, clearly outlining how privacy is protected, how data is used, and the steps taken to ensure fairness and transparency in order to proactively address potential client concerns.
10. If the NEA decides to implement an online questionnaire as part of the profiling process, it must be well-structured, clear, accessible, and easy to understand. The questionnaire should strike a balance between depth and brevity, ensuring that it captures essential information without being overwhelming or confusing. Thoughtful design will improve response accuracy and user engagement. The questionnaire should be tested with both counsellors and jobseekers before its full implementation and refined based on user feedback.
11. Profiling tools may be technically very advanced and predict with a high level of accuracy. Nevertheless, the outcomes and referrals that usually follow profiling require that well-designed and performing employment programmes and services are continuously available to support jobseekers depending on their needs.
Source: Authors’ compilation.
6.3.3. Strengthening matching and employer engagement
Chapter 4 outlined the actions and processes the NEA could implement to enhance employer engagement, an area that already has a solid foundation to build upon. An important aspect of this effort involves leveraging modern digital and AI-based tools to improve various areas of the NEA’s work with employers. Several PES across EU and OECD countries already use digital and AI tools to assist them in various interconnected ways (Brioscú et al., 2024[2]; Dromundo Mokrani, Lauringson and Xenogiani, 2024[9]). Examples of work areas that these tools can be of use include facilitation of job matching, support with the design of job vacancies and recruitment including for hard-to-fill roles, streamlined processes to attract engagement of employers with the PES, support with and provision of real-time labour market insights, and outreach to businesses (Figure 6.4).
Source: Authors’ compilation.
One of the key areas for improvement is AI-enhanced job matching and vacancy management. The NEA is currently in the process of developing a new digital tool, featuring an advanced matching component. The new tool is expected to enhance labour market mediation by matching jobseekers’ profiles with relevant job vacancies, assisting both jobseekers and employers. The experiences of other countries in developing digital tools for labour market mediation can offer valuable insights for the NEA as it creates its new matching platform (e.g. those from France, Greece, Slovenia and Sweden).
The new matching tool should be able to not only match jobseekers with vacancies based on their skills and qualifications, but also incorporate predictive analytics to suggest alternative roles and new career possibilities based on transferable skills and other information. Most PES vacancy matching algorithms leverage machine learning to incorporate a broader range of data, including not only the information collected during registration but also insights gained from the individual’s interactions with the PES and the services provided. By using competencies and skills taxonomies such as ESCO, in addition to traditional job classifications, the system can more effectively match job opportunities with registered jobseekers, ensuring that roles are aligned with the specific abilities and qualifications of candidates. If the tool proves effective, it could be further enhanced to not only suggest job opportunities but also recommend alternative pathways, such as training programmes and courses.
A dynamic vacancy management system within the tool’s environment will enable employers to post, update, and track job vacancies in real time. It should also allow the NEA to detect duplicate vacancies, as well as any illegal postings that are not compliant with labour regulation. Such systems typically send alerts to PES staff so that they can review the case and reach out to the employer if needed, as in the case of Canada, France and Sweden. By analysing employer and vacancy data, additional features can offer valuable insights into various aspects of the company’s current situation, such as its growth trends, seasonal employment patterns, and workforce composition, including demographic data like age and gender distribution. This information can help the NEA account managers and counsellors identify employers who are likely to recruit new staff. The new matching tool also envisions the development of an AI-assisted chatbot that would provide real-time information on services and benefits and address queries for both employers and jobseekers.
The new matching tool should be and transparent, user-friendly for both employers and jobseekers, which is essential for building trust and increase usability. To maximise effectiveness, the NEA should promote the tool to employers (both large and small businesses, business associations and chambers of commerce) and provide technical support and user guides. The NEA should also engage employers throughout the design and testing phases to ensure that the tool is aligned with real business needs. Promoting the tool to both employers and jobseekers is also important to ensure high adoption and engagement. This can be achieved by clearly communicating the tool’s benefits, such as improved job matching and career guidance, through targeted outreach campaigns. Additionally, offering support and guidance on how to use the tool effectively, such as through tutorials or job counsellor assistance (especially for those who lack digital skills), can help end users feel more comfortable and confident in using the tool. What is more, as already foreseen, training on how the tool works and how to interpret job matches generated by the tool should be provided to frontline staff. It is also important to regularly update the algorithms to reflect changing labour market conditions and needs and avoid biases in recommendations. Finally, data privacy, security, and compliance with national and EU regulations are critical aspects that the NEA already prioritises.
The PES in Korea has developed a machine learning matching tool known as the Work, which conducts matching across two dimensions (Brioscú et al., 2024[2]; OECD, 2024[10]). First, it conducts competency-based matching between the profile of the jobseeker and criteria of the job description, using job dictionaries. Second, click data from the jobseeker’s job browsing patterns is used to produce job-search behaviour-based matching to recommend similar vacancies to those viewed by the jobseeker, thus accounting for the preferences of the jobseeker. The tool also allows the user (either the jobseeker or PES counsellor) to filter job recommendations, such as by wage or location.
AI tools can assist the PES to support employers in meeting their recruitment needs. Several PES deploy AI tools to systematically identify and classify the occupation of a job offer and the associated competencies and skills required by the employer. Some of these tools propose the appropriate occupations and skills according to the ESCO classification. Beyond recommending the most relevant occupations and skills for job postings, PES can leverage AI tools to assist employers in drafting the actual job descriptions (e.g. in Luxembourg) and to provide suggested text to employers based on historical vacancy postings (e.g. in Canada). Additionally, AI can support PES staff in validating and refining postings before publication, thereby reducing their workload.
AI tools can also identify vacancies that are likely to be harder to fill, helping employers make these positions more appealing to jobseekers. For example, the French PES has implemented two AI algorithms in this domain to: i) predict the time it will take to fill a vacancy, and ii) to calculate the attractiveness of a vacancy. This allows PES advisors to reach out to employers to resolve any issues and to propose solutions to enhance the prospects of a vacancy (Brioscú et al., 2024[2]).
The NEA has a dedicated webpage for employers integrated into its main portal. However, this webpage is rather basic. Such an updated environment should serve as a centralised platform to register, accurately classifying vacancy postings by occupation and required skills, manage job postings, track applications, and access labour market information. Ideally, an updated employer portal would be integrated with the new job matching tool. Interactive dashboards should be able to provide real-time analytics on sectoral labour market trends and recruitment success rates. Additional features can be integrated and help employers in designing effective job advertisements. Employer satisfaction surveys could be conducted through this environment at occasional intervals.
The results of forecasting tools could also be a valuable source of information for employers, as they can provide insights into local labour market conditions, skill shortages, emerging industries and labour market trends (see also section 6.3.4. on strengthening labour market information). Having access to this information would allow employers who co‑operate with the NEA to gauge the labour market situation in their sector and region, proactively plan their workforce needs, tailor recruitment strategies, and invest in relevant training and development. By anticipating challenges such as skill gaps or labour shortages, employers can adjust hiring practices, compensation strategies, and talent pipelines to stay competitive. Ultimately, these tools can enable employers to make data-driven decisions, ensuring they are prepared for future workforce demands and better positioned for long-term growth.
Key horizontal considerations for successful integration and implementation of digital and AI technologies for employer engagement include:
Ensure strong interoperability between the new job matching tool and other digital platforms (employer-focused and in some cases jobseeker support tools).
Engage employers throughout the design and testing phases to align tools with real business needs.
Run employer satisfaction surveys at regular intervals and take into account the results for updating the tools and reconsidering their use.
Invest in training for NEA staff to maximise the benefits of new technologies.
6.3.4. Easing administrative processes and minimising reliance on in-person and paper-based procedures
PES have various back-office and administrative processes which can benefit from digitalisation and the use of AI. Across OECD countries, modernisation in this area has most commonly taken the form of digitalisation or automation, with AI use cases only gradually emerging. The high burden of administrative processes stands out as a central characteristic of the daily work of NEA staff. As already described in previous chapters, the registration process at the NEA still relies heavily on in-person document submissions. In addition, processes involving data entry into the NDB are particularly time‑consuming, as not all necessary data can be automatically retrieved from other databases and counsellors must manually enter information multiple times into the system.
The NEA is currently making major efforts to improve its data management. This transformation is expected to streamline the client experience, significantly reduce processing times, and alleviate administrative burdens on NEA staff, particularly job counsellors. For instance, counsellors should be able to access essential information for the jobseekers without the need to navigate multiple registers. Similarly, NEA staff would save time if they could automatically see the eligibility of jobseekers for different ALMPs without having to check compliance with all eligibility criteria manually. While these efforts are promising, it is important that relevant initiatives are fully implemented to deliver tangible improvements. The NEA should aim for a rapid implementation, e.g. within 12 months, to ensure that the benefits of the new system become available as soon as possible.
The integration of automatic data retrieval is a first step to improve the efficiency of the registration process, but expanding its use will be needed to further facilitate and accelerate the registrations while minimising the potential for human error during data entry. For example, automated data transfers between the NDB and other data systems, including data of the National Revenue Agency and of the National Social Security Institute, as well as platforms with vacancy information, would help streamline data processes and reduce manual entry errors. To this end, the NEA needs to continue investing in upgrading its IT infrastructure to support data exchanges and ensure system compatibility between different databases and platforms. It will also be key to standardise data formats across all connected systems to ensure smooth transfers and reduce discrepancies.
In its efforts towards modernisation, it is important reduce the NEA’s reliance on paper-based processes, by fully embracing digital communication channels. Automating processes like remote registration, electronic submission of documents, remote application processing, automated notifications for jobseekers, referral letter submissions and others enhance client experience, reduce processing times, and improve engagement.
Other countries, too, struggled with a high administrative load implemented reforms to make administrative tasks more efficient, thereby freeing up time for client-facing activities. Successful examples for the use of digital tools and more advanced data management systems to streamline performance reporting and reduce paperwork for staff include the Data Warehouse at Germany’s Federal Employment Agency and better workflow management systems in Belgium and Austria. AI tools can also assist PES in detecting fraud by analysing large datasets, identifying inconsistencies, and flagging suspicious claims from individuals. For example, Sweden’s fraud detection system is able to monitor jobseeker (and employer) data, helping to pinpoint potential fraudulent activity (Brioscú et al., 2024[2]).
6.3.5. Enhancing labour market information by using advanced forecasting tools and real-time data
Labor Market Information Systems (LMIS) are a critical component of any modern PES infrastructure. These systems provide real-time labour market data, skill demands, and sector growth, which are essential for informed decision-making. Some systems also encompass a forecasting component. Systematic sectoral and occupational forecasting skills anticipation can help governments and PES identify emerging trends in labour markets, predict skill shortages, and design relevant interventions such as job placement services, training programmes and other workforce development initiatives. Some PES have already incorporated AI methodologies for both ex-post analysis and labour market forecasting. However, with AI’s rapid evolution, its role in these areas is expected to expand significantly in the coming years (Brioscú et al., 2024[2]).
Generating labour market information, conducting in-depth analyses based on this information, and carefully interpreting the results, is instrumental for the NEA, as evidence‑based policymaking forms the cornerstone of effective labour market interventions. In addition, accurate and up-to-date labour market information enable counsellors to better support both jobseekers and employers. Counsellors are then able to provide job seekers and individuals who wish to move to another job with informed guidance, helping them make strategic career decisions, identify emerging job opportunities, and develop the skills most in demand. What is more, they can also support employers identify the right talent, make informed recruitment decisions and adapt their workforce strategies. Other stakeholders may also benefit from having access to this information, either directly or the results of relevant analyses. Such stakeholders may include training providers, employer associations, trade unions, regional and local authorities, NGOs and private employment services.
Understanding the profiles of jobseekers is essential for designing and implementing effective employment services and addressing the diverse needs of various target groups. By gaining a comprehensive view of the characteristics of the jobseekers including their strengths, weaknesses and the barriers they face, the NEA can develop targeted strategies to improve their employability, bridge potential skill gaps, and connect them to suitable job opportunities. Labour supply analyses can be conducted across different dimensions. Basic statistics analysis involves analysing demographic characteristics such as gender and age, educational backgrounds, unemployment spells, previous work experience, skill levels, and receipt of unemployment or/and other benefits. Cross-tabulations allow for a higher level of disaggregation, enabling the simultaneous analysis of data across multiple dimensions. Examples include distribution of age groups by unemployment duration, most frequent sectors and occupations among registered unemployed by gender and age group. Finally, analysing detailed profiles of jobseekers can be useful for planning targeted interventions. Profiles of interest may include youth, women, long-term unemployed, low-skilled unemployed and beneficiaries of social assistance.
The NEA already has good data to perform such analyses. Improvements in the National Database and interoperability with other databases would grant the NEA real-time access to additional information on jobseekers thereby enabling a more in-depth understanding of their profiles. The focus of such a potential analysis could vary, depending on whether planned interventions are designed for national or regional implementation. In contexts such as local labour markets with high seasonality or those experiencing shifts due to the green transition, the analysis should be carried out at the regional or municipal level. In such cases, additional information may be required, including on employed individuals at risk of dismissals in order to design targeted interventions proactively. In certain cases, qualitative information can complement administrative data, offering valuable insights that quantitative analysis alone may miss. This information can be gathered through individual interviews with job counsellors, or by conducting targeted surveys designed to explore the profiles of individuals who should be targeted. Timing is also important. In addition to short-term comparisons (e.g. monthly, quarterly), yearly comparisons are also necessary to identify and study longer-term patterns or shifts.
Analysing labour market conditions and trends is essential to enable evidence‑based policy decisions according to the needs of the labour market. Labour market analyses can be carried out at different levels including at the national, regional or even local level. Labour demand analysis should encompass both the current /more recent situation as well as longer-term trends. The former include information such as job creation, separations, levels of employment, average wages across various industries etc. This can be further disaggregated by key factors such as region, sector, skill level, and demographic characteristics (e.g. age, gender, education level). The latter type of analysis focuses on longer-term labour market trends from with data from the past few years, to understand how the labour market under examination has evolved. This can be reflected in terms of absolute numbers or growth rates for industries and occupations, helping to identify patterns, shifts, and emerging trends. In cases of major transitions or long-term structural changes, extending the timeframe of the analysis, may offer a clearer picture of sustained trends and transformations. For a general overview, a 2‑digit classification (e.g. broad industry sectors) is often sufficient to capture macro-level trends and make high-level policy decisions. However, for more detailed analysis, particularly when targeting specific sectors, occupations, or regions, a 3‑digit or even a 4‑digit classification (e.g. sub-sectors or specific occupations) is more appropriate. In Bulgaria, employment information is captured in the Employment and Social Security Information System, which compiles information on formal employment, wages, and contributions to the social security system.
Local employer surveys or consultations can be a useful complement to administrative data sources. The NEA can leverage its existing experience with such surveys and conduct them in a more systematic and regular way, trying to make the maximum out of these, including for validating results and trends identified by the administrative data. For instance, if administrative data suggest a need for specific training, the NEA can follow-up by surveying employers to validate these findings with them and learn more about their hiring and skills needs.
For certain labour markets, such as the ones under transition or with seasonal fluctuations, additional analysis could be undertaken. For example, in areas where mass layoffs are anticipated due to the green transition, the NEA should not only analyse labour demand in the areas directly affected but also in the neighboring regions and nearby urban centres. The same methodology could be applied to seasonal markets for the identification of sectors that recruit outside the pick times including in nearby areas. More generally, local labor market surveys and consultations can help the NEA collect additional useful information that may not be captured through administrative data.
In Bulgaria, systematic sectoral and occupational forecasting is still in its early stages. These efforts are largely ad hoc and project-based, often conducted as part of CEDEFOP’s European skills forecasting exercises, which provide quantitative projections of future employment trends by sector and occupation. As a result, the NEA faces challenges in accurately anticipating future skills gaps, sectoral shifts, and labour market needs. Even though these insights can still be useful to the NEA, Bulgaria’s forecasting processes should be further strengthened and integrated into policymaking, as done in other EU and OECD countries.
To improve the forecasting process in Bulgaria, it would be beneficial to invest in advanced labour market information systems that can provide real-time data and predictive insights. The NEA can learn from the experience of PES and other public institutions in EU and OECD countries that are already advanced in this domain. Some PES have developed dedicated systems in-house or use the results of existing systems hosted by other public agencies that leverage a combination of methodologies and integrate data from various sources, including administrative records, surveys, and reports (see Box 6.2). Another notable example is Estonia’s OSKA forecasting system that anticipates future labour market trends and skill demands across various sectors. Managed by the Estonian Qualifications Authority, OSKA utilises quantitative analysis derived from national registries, sector surveys, and labour demand forecasts to identify skills gaps essential for the country’s economic development. The system helps policymakers and the PES to better align their strategies and policies with future labour market needs. OSKA’s results are particularly insightful for the development of educational and training programmes that align with labour market demands, ensuring the workforce remains adaptable and well-prepared for emerging challenges and opportunities (European Commission, 2022[11]). These systems can continuously provide real-time updates and monitor labour market trends.
For the NEA, adopting or partnering with relevant institutions to develop a similar system would enhance its ability to forecast labour market trends and promote more proactive, data-driven decision-making. Additionally, strengthening collaboration with educational institutions, industry stakeholders, and employers could help enhance the accuracy and relevance of sectoral and occupational forecasts, ultimately improving the responsiveness of the labor market and increasing alignment between jobseekers’ skills and employer demands. The results from the monitoring of the labour market outcomes for individuals participating in NEA’s programmes can also provide some labour market information, though indirectly (see section on monitoring and evaluation later in this chapter).
The Greek Labour Market Diagnosis Mechanism (M.D.A.A.E.) is a foundational tool for designing data-driven labour policies in Greece. It collects, analyses, and visualises labour market data – such as employment, unemployment, vacancies – sourced from administrative databases and surveys. Using innovative diagnostic methods, it identifies the professions and skills most in demand across sectors, regions, and municipalities. By assessing labour supply (registered unemployment) against demand (newly created jobs), it pinpoints mismatches to optimise training and resource allocation. Its focus is on regional dynamics enables tailored solutions, and the design of policies enhancing adaptability and efficiency of labour market in Greece. It is developed by the Unit of Experts in Employment, Social Insurance, Welfare and Social Affairs (M.E.K.Y.) of the Ministry of Labour and Social Insurance and is extensively used by the Greek PES. Key functions include:
Monitoring employment trends, highlighting the most dynamic professions and industries across national, regional, and local levels, with detailed breakdowns by key factors.
Identifying current labour market needs in occupations and skills in relation to the skills offered by the workforce.
Forecasting labour needs that will arise in the immediate future and the medium term, based on economic activities dynamics.
Conducting surveys, business studies, and focus groups to evaluate workforce gaps at national and local levels.
Maintaining an information system and website for data storage and publishing findings, including an Annual Report, to support policymakers in various Ministries and the PES, the National Workforce Skills Council, and researchers.
Recently, the Mechanism has expanded to sectoral and occupational forecasting, aiming to identify the technical skills required for high-skill jobs in the ICT sector and the green skills in in the tourism, financial services, and pharmaceuticals sectors. Similar analyses are planned for additional sectors in the future.
Future initiatives also include the development of a tool to assess labour supply-demand imbalances – by profession, gender and age – at multiple geographic levels; the creation of an occupational guide; and the implementation of biannual business surveys on their staffing and skills needs with a focus on digital and green skills.
Source: (Mechanism of Labour Market Diagnosis, 2025[12]) and Presentation by officials of the Unit of Experts in Experts in Employment, Social Insurance, Welfare and Social Affairs (MEKY) of the Greek Ministry of Labour and Social Insurance during the 2nd capacity building seminar organised on 28 May 2025 as part of this technical assistance project.
More recently, a number of PES have adopted AI-driven tools for job classification and skills identification purposes. In these cases, new sources of labour market information such as information from job postings can complement conventional labour market data sources. The AI model used by the PES in Luxembourg for example, can classify occupations and identify skills from job descriptions, using the ESCO framework. Some of this information is publicly available through an online dashboard. There are plans to integrate these models into job-matching and counselling services (Brioscú et al., 2024[2]). Similarly, in Flanders, Belgium, neural network models predict occupational and other skills demand based on job postings. The Swedish PES leverages AI and natural language processing to analyse job postings, track labour market trends, and improve job matching (Brioscú et al., 2024[2]).
While AI use to aid the generation of such knowledge by PES is largely in its infancy with only a few limited examples, this is likely an area where AI deployment will increase in the coming years (Brioscú et al., 2024[2]). In generating labour market information, AI use by PES at present largely focusses on labour market monitoring activities and to understand changing skills demands using vacancy information. These insights can then be used to inform the design and targeting of PES support, including by aiding the job-search support and guidance given to jobseekers by counsellors.
AI can also contribute to the production of evidence on the impact of ALMPs on participants. This can be achieved by deploying machine learning methods to counterfactual impact evaluation methodology to enhance the determination of sub-group impacts (Athey, 2019[13]; Lechner, 2023[14]; Cockx, Lechner and Bollens, 2023[15]) or by developing AI-powered system to automate the production of policy evaluations. The latter is seen in Estonia, where the PES has implemented the MALLE tool to conduct systematic and regular counterfactual impact evaluations of a number of key ALMPs. These evaluations focus on the employment status and earnings of programme participants and visualise results in interactive dashboards for PES staff.
The NEA has so far little experience in monitoring and evaluating the impact of its ALMPs on key labour market outcomes. While it tracks basic employment indicators for participants following the completion of programmes and conducts surveys to assess client satisfaction – or, in the case of employers, their hiring needs, these efforts provide only partial insights and do not measure the actual effectiveness of ALMPs in improving employment prospects or reducing unemployment. Developing a robust monitoring and evaluation framework is crucial for evidence‑based policymaking and resource allocation. Ideally, the NEA should also invest in building a comprehensive monitoring and evaluation system for ALMPs with integrated feedback loops to allow for continuous programme improvement. This requires not only investing in data collection and analysis capabilities, but also ensuring access to high-quality administrative data while addressing privacy concerns and building internal expertise.
The first step in establishing a comprehensive monitoring and evaluation framework for the NEA’s ALMPs is to develop clear results chains and well-defined indicators for its priority programmes. A results chain outlines the logical sequence of inputs, activities, outputs, outcomes and expected long-term impacts of each ALMP, providing a structured approach to measuring its effectiveness. In parallel, defining the appropriate indicators, identifying their sources and establishing the frequency of their collection, is essential to track progress and assess the impact of ALMPs. In addition to the indicators, the NEA can also set targets, depending on its priorities and objectives. Result chains, indicators and targets should be reviewed on a regular basis and adjusted as necessary to ensure their continued relevance, accuracy, and alignment with evolving labour market dynamics and the agency’s policy priorities.
Effective monitoring is crucial at every stage of programme implementation to ensure that everything is proceeding as initially planned. It also allows for timely adjustments. Evaluation is the systematic and objective assessment of a programme or policy, focusing on its design, implementation, and/or outcomes at a specific point in time. Process evaluations can generate significant insights as their aim is to understand how processes are implemented in practice compared to how they are supposed to work. These evaluations generally rely on administrative data, but also usually incorporate inputs from qualitative methods such as interviews, consultations and focus groups discussions with various relevant stakeholders. Such stakeholders may include policymakers and programme designers, end users such as frontline staff, jobseekers and employers as well as intermediaries such as training providers. The results can identify areas that work well, areas for improvement, address any gaps in service delivery, and ultimately enhance the overall effectiveness of the programme. Impact evaluations and particularly counterfactual impact evaluations (CIEs) are generally the most reliable way of measuring the effects of ALMPs. Their key aim is to measure the causal impact of a specific policy or programme. This involves determining the difference between the outcomes after the measure is implemented and what those outcomes would have been in the absence of the measure.
The technical assistance project underlying this report places a strong emphasis on impact evaluation and capacity building. In particular, it includes an impact evaluation of one of the NEA’s core ALMPs namely professional counselling services (see Box 3.6 in Chapter 3). Additionally, the project includes a guidance note to help the NEA prepare for impact evaluations of ALMPs, accompanied by a series of capacity-building webinars on impact evaluations related issues (see Box 3.6 in Chapter 3). The NEA should further invest in evaluations, particularly CIEs, to rigorously assess the effectiveness of its operations and programmes. The NEA can also learn from the experiences of countries that participated in the OECD-EC project on impact evaluation of different ALMPs. They demonstrate how the use of linked administrative and survey data provides valuable information for making informed policy choices (OECD, 2024[16]). Strengthening its capacity for evidence‑informed policymaking will eventually enable the NEA to design more targeted and efficient interventions that address labour market needs.
Several PES have either established monitoring and evaluation systems for their ALMPs, or streamlined impact evaluations of ALMP, by leveraging statistical computing software and BI tools. The Greek PES has introduced a monitoring and evaluation system to assess policy effectiveness and is currently exploring ways to untap its potential by enhancing integration with additional data sources and optimising data utilisation. The Estonian PES has adopted tools and technologies that enable near-real-time assessment and visualisation of the labour market effects of training programmes, employment incentives, and work-related rehabilitation schemes. Similarly, the PES in the Slovak Republic has outsourced the development of automated reports evaluating ALMP impacts.
Information for monitoring needs to be presented clearly and concisely, ensuring it is easily readable and understandable. To achieve this, the NEA should adopt suitable data visualisation and reporting tools. Various Business Intelligence (BI) tools, including Microsoft Power BI, Tableau, Qlik Sense, Oracle Business Intelligence, SAP BusinessObjects and others, can offer powerful analytics capabilities. These tools can connect to multiple data sources, visually represent key data, and generate automatic reports that integrate tables, figures and text into a single document. Dashboards can provide the NEA with real-time or regularly updated insights, enabling quick assessments of programme performance, but also serve external stakeholders. Internal reports can be generated at periodical intervals (e.g. monthly, quarterly or annually) depending on the needs and priorities of the NEA. These reports should be concise, highlighting key findings and trends to support informed decision-making. Customisable reporting options should be available to address specific needs, such as tailoring programmes for particular target groups or regions.
The NEA register already has a good foundation of data for monitoring and evaluation purposes, as it contains detailed information on all registered jobseekers, such as personal and family characteristics, data related to registration and termination of registration, unemployment spells, information from profiling and APs, ALMP participation as well as incomplete data on vacancies to which counsellors direct jobseekers. Enhancing the monitoring and evaluation of ALMPs would greatly benefit from access to a broader range of data. This would be particularly effective if there were continuous, systematic, and real-time interoperability with other key databases in Bulgaria, such as the Social Assistance Agency register, the National Social Security Institute, the employment register and the tax registry managed by the National Revenue Agency and possibly other administrative registers.
6.3.6. Monitoring and evaluation mechanisms for assessing the performance of new tools and mitigating risks of bias
Just like the ALMPs implemented by the NEA, digital and AI solutions implemented by the NEA should also be subject to rigorous and regular monitoring and evaluation to ensure model performance and assess the impact on end-users (OECD, 2022[8]; Brioscú et al., 2024[2]).
To assess the impact of the NEA’s AI systems, evaluations can be undertaken both prior to roll-out by piloting measures in the form of randomised control trials or following implementation through counterfactual impact evaluations. Furthermore, process evaluations (to assess the effectiveness of the integration of new solutions) or cost-benefit analyses (to systematically estimate value for money) offer additional routes for knowledge generation surrounding the planned AI developments of the NEA.
The NEA should also closely monitor its digital and AI solutions to ensure continued quality of outputs and to detect any potential model drift or degradation. In addition, monitoring AI systems is crucial to ensure that outputs remain equitable and that biases do not emerge over time.
In addition to these monitoring and evaluation activities, the NEA should ensure that feedback loops are established to ensure learnings and experiences are fed into future AI initiatives and to avoid making the same mistake twice. This includes taking necessary steps to fix or abandon poorly performing solutions.
EU PES have implemented monitoring and evaluation frameworks to measure the performance of their tools. For example, the French PES has implemented a rigorous monitoring framework to evaluate digital tools, ensuring they are user-friendly, impactful, and sustainable. Similarly, a recent OECD/European Commission technical assistance project supported the Spanish PES by rigorously evaluating a digital tool designed to assist counsellors in providing personalised job search and training advice (OECD, 2023[17]).
The adoption of digital technologies, including those powered by AI, presents many opportunities for PES. However, it also involves challenges and risks. Based on the NEA’s responses to an OECD questionnaire on digitalisation and AI use in PES, the most prominent challenges include the lack of the necessary skills among staff, issues related to data, resistance from staff, understandability and explainability, concerns around unfair and biased decision-making, national concerns about more elaborate digital solutions, the objective within the organisation to maintain a human-centred model of service delivery and concerns regarding transparency.
The use of AI presents a number of specific risks that are particularly relevant for the NEA. It is essential therefore to be aware of these and take proactive steps to mitigate them (Brioscú et al., 2024[2]):
Accountability is key when AI systems impact people’s lives. For the NEA, this means ensuring proper system performance through strong governance frameworks and conducting ongoing risk management throughout the AI lifecycle to identify and address potential issues.
Making AI easier to understand is essential. This means promoting transparency about how AI systems are used and how they produce results – both for staff and clients – to build trust and understanding. For example, in the case of an AI job-matching tool, this means having full transparency around how the algorithm works, what data it uses, and how decisions are made.
The success of an AI model depends on the quality of input data. For the NEA, the development of AI solutions should go hand-in-hand with work to ensure data readiness. The second important point regarding data use in AI systems is that of data security and privacy. This is particularly critical given the significant volumes of personal and sensitive data thar are held in the NEA’s databases.
AI systems carry risks of bias and discrimination, which could greatly impact clients’ job opportunities and prospects. Therefore, it will be important for the NEA to take steps to incorporate fairness into AI design and take proactive steps to prevent such risks.
The modernisation journey being undertaken by the NEA can be met with reluctance and opposition from end users – both staff and jobseekers. In addition, lack of skills also presents a challenge in the adoption of AI. Successful adoption of AI systems by the NEA will hinge on the deployment of measures to close the skills gap generated by these new technologies among all end-users (staff and clients).
The NEA’s digital transformation may face resistance from staff and jobseekers. Lack of skills among both staff and clients presents a further challenge in the adoption of AI. As digitalisation continues to accelerate, the NEA should ensure that no one is left behind in the process. This requires targeted efforts to close skills gaps among NEA staff as well as focused attention to individuals lacking digital skills or access to digital infrastructure, as well as vulnerable groups for whom digital methods may not be suitable for engagement or service delivery.
AI models can degrade over time, especially when new data are not integrated, risking poor outcomes. For the NEA, it will be important to closely monitor AI models and their outputs and detect potential model drift. More generally, it will also be important for the NEA to evaluate the impact of AI systems, particularly on end-users (both staff and clients).
In seeking to seize the benefits of AI, the NEA must take some practical steps to mitigate related risks and to prepare the organisation for the associated change. In doing so, the NEA should consider existing frameworks to guide their AI use, such as the OECD’s AI Principles which seek to promote the responsible use of trustworthy AI by AI actors (OECD, 2024[18]).7 The most important steps are the following:
1. Prioritise transparency and explainability: The NEA should opt for clear and easy to understand AI systems – so-called “white box” models – over complex ones with hidden logic and unknown decision-making mechanisms. This will not only help the NEA staff to use the new tools more effectively but will also foster trust among jobseekers. For example, an AI-powered tool to profile jobseekers can help a counsellor decide on suitable services and measures. However, if the counsellor doesn’t know what information the tool used or how it reached the recommended profiling score, they might make incorrect suggestions or overlook important issues. Transparency and explainability on how the NEA is using AI should apply both internally and externally. In the Netherlands, for instance, the PES publishes information about how AI is used, including data sources, safeguards against bias, and human oversight (UWV, n.d.[19]).
2. Ensure close co‑operation with providers and sufficient internal expertise: Since the NEA will outsource AI development, close co‑operation with providers is essential to ensure tools meet user needs. Other PES, like in Norway and Luxembourg, achieve this by bringing developers into the agency so that they can work alongside the relevant experts and enhance their understanding of the context for the developments or promoting close collaboration between technical and operational teams in order to encourage closer and regular communication and more efficient solving of any roadblocks or technical issues. Even if physical proximity is not possible for the NEA, regular communication with the providers and strong project management remain crucial. The NEA should also ensure a sufficient level of internal expertise to guide development, monitor progress, and handle future updates effectively.
3. Involve and support end-users from the outset and on a regular basis: Engaging both staff and jobseekers in the design and testing of AI solutions would help ensure they are useful, user-friendly, and trusted. Collecting feedback prior to finalisation and rollout – through consultations and customer satisfaction surveys – can greatly improve adoption. In the case of the NEA staff, it would be important to provide relevant support including through training, the provision of informational materials and guidelines, assignment of a dedicated contact point for technical assistance. Staff should also be involved in decision-making. In France, for example, jobseekers and counsellors took part in a working group that helped draft a “Charter for Ethical AI” for the PES.
4. Ensure human oversight: AI solutions should support – not replace – human decision-making. The NEA should make clear that final decisions still rest with staff and allow them to override AI recommendations when necessary. This will not only help NEA clients trust AI systems more, but also reassure NEA staff that they still have control and can make their own decisions. In some cases, this is a legal requirement – for example, Article 22 of the general data protection regulation (GDPR) gives individuals the right not to be subject to decisions made solely through automated processing (Council of the European Union and European Parliament, 2016[20]).
5. Establish a strong governance framework surrounding AI use: The NEA should put in place a governance framework to guide AI use and ensure compliance with rules like GDPR and the EU AI Act. Examples from other PES include Ethics Committees and oversight Boards that review and monitor AI systems (e.g. PES in France and Flanders in Belgium).
References
[13] Athey, S. (2019), “The Impact of Machine Learning on Economics”, in The Economics of Artificial Intelligence, University of Chicago Press, https://doi.org/10.7208/chicago/9780226613475.003.0021.
[2] 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.
[7] 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.
[5] Chapuis, R. (2018), “Les impacts de l’intelligence artificielle sur l’emploi : comment favoriser la complémentarité avec l’humain et faire émerger de nouveaux types de métiers ?”, Les Annales des Mines: Enjeux Numériques, Vol. 1, https://www.annales.org/enjeux-numeriques/2018/en-2018-01/EN-2018-03-06.pdf.
[15] Cockx, B., M. Lechner and J. Bollens (2023), “Priority to unemployed immigrants? A causal machine learning evaluation of training in Belgium”, Labour Economics, Vol. 80, p. 102306, https://doi.org/10.1016/j.labeco.2022.102306.
[20] Council of the European Union and European Parliament (2016), Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC, (General Data Protection Regulation) (Text with EEA relevance), http://data.europa.eu/eli/reg/2016/679/oj.
[9] 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.
[11] European Commission (2022), PES Practices Database, Estonia: Anticipating labour market needs via the “OSKA” skills forecasting system, https://ec.europa.eu/social/main.jsp?catId=1206&langId=en (accessed on 4 November 2024).
[1] European Commission (2022), Proposal for a Council implementation decision on the approval of the assessment of the recovery and resilience plan for Bulgaria, https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52022PC0172 (accessed on 5 July 2024).
[3] European Commission (2021), PES Practices Database, Estonia: Blended employment counselling: combining face-to-face and distance counselling of the jobseekers, https://ec.europa.eu/social/main.jsp?catId=1206&langId=en (accessed on 30 October 2024).
[14] Lechner, M. (2023), The Value Added of Causal Machine Learning for Public Policy, EC-OECD Technical Workshop: Evaluating labour market policies through the use of linked administrative data. Paris, France, https://www.oecd.org/els/emp/Session3_Value_Added_Causal_Machine_Learning.pdf.
[12] Mechanism of Labour Market Diagnosis (2025), The Mechanism, https://mdaae.gr/en/o-michanismos/ (accessed on January 2025).
[6] OECD (2024), Modernising Latvia’s Public Employment Service through Digitalisation, Connecting People with Jobs, OECD Publishing, Paris, https://doi.org/10.1787/07e830f5-en.
[16] OECD (2024), OECD-EC project on policy impact evaluation through the use of linked administrative and survey data, https://www.oecd.org/en/about/projects/oecd-ec-project-on-policy-impact-evaluation-through-the-use-of-linked-administrative-and-survey-data.html.
[18] OECD (2024), Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449, https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449 (accessed on 18 September 2025).
[10] OECD (2024), Strengthening Active Labour Market Policies in Korea, Connecting People with Jobs, OECD Publishing, Paris, https://doi.org/10.1787/44cb97d7-en.
[4] OECD (2024), Using AI to improve job matching tools for minimum income scheme beneficiaries, https://www.oecd.org/en/about/programmes/dg-reform/using-AI-to-improve-job-matching-tools-for-minimum-income-scheme-beneficiaries.html (accessed on 6 June 2025).
[17] OECD (2023), Impact Evaluation of the Digital Tool for Employment Counsellors in Spain: SEND@: Report on the Design and Implementation of an Impact Evaluation of the Digital Counselling Tool for Spain’s Public Employment Services, OECD Publishing, Paris, https://doi.org/10.1787/fe1ec3c3-en.
[8] OECD (2022), Harnessing digitalisation in Public Employment Services to connect people with jobs, Policy Brief on Active Labour Market Policies, https://www.oecd.org/en/publications/harnessing-digitalisation-in-public-employment-services-to-connect-people-with-jobs_367a51f3-en.html.
[19] UWV (n.d.), Onze algoritmes, https://www.uwv.nl/nl/over-uwv/organisatie/algoritmeregister-uwv/onze-algoritmes (accessed on 14 May 2024).
← 1. Other activities under this technical assistance project are related to the development of digital tools that can assist the NEA to provide more efficient services to its clients. These activities include a note on the constraints and risks associated with the use of digital tools supported by AI by PES as well as the development of a monitoring and evaluation plan for the new matching tool that the NEA is currently developing. Further, a technical note has been prepared to help the NEA develop the data for a new profiling tool that is in NEA’s future plans. In addition, an international workshop held in June 2024 showcased the experiences of a selected number of OECD and EU PES in the use of AI to assist their services and processes. This included lessons learned during the course of developing and implementing such solutions. Another international workshop organised in February 2025 briefly touched upon some of these tools (e.g. profiling, digital platforms for counselling), placing special emphasis on issues related to implementation.
← 2. Jobseekers still need to provide physical documents when registering with the NEA, especially if data cannot be obtained ex officio from existing registers. While remote registration is possible, it’s rarely used and still requires documents to be sent physically. Most data entry is done manually by counsellors at different times, as only some can be retrieved through Bulgaria’s inter-register exchange system (see also Chapter 3).
← 3. The current version of the NDB is based on a structured, relational model where data is stored in predefined tables with strict schemas. This means that any changes to the structure can be complex and time‑consuming. The new system will be based on NoSQL databases, allowing for flexible schemas, meaning one can store different types of data without predefined structures, making it easier to adapt to changing requirements.
← 4. The digitalisation strategy for the Latvian PES was the outcome of an OECD and European Commission project on the modernisation of Latvia’s PES (OECD, 2024[6]).
← 5. Long-term unemployment is defined differently across countries. In most cases, the threshold is 12 months, but in some countries, it is shorter, such as 6 months in Belgium and Sweden, or 26 weeks in Denmark.
← 6. Other outcome variables include the probability of labour market (re)integration, exiting the unemployment registry due to employment and reducing benefit dependency. Most outcome variables are of binary format, with only a few cases being processes in a continuum way.
← 7. The OECD AI Principles were adopted by Member countries in May 2019 as part of the OECD Council Recommendation on Artificial Intelligence. The Principles were then subsequently updated in May 2024. The Recommendation sets out five values-based principles for the responsible stewardship of trustworthy AI, alongside related recommendations for governments to include in their national policies and international co‑operation in this domain.