This chapter looks at advances in tax rule management, including the use of assurance frameworks to validate the integration of tax rules into third party products and systems, the automation of registration processes, and the role of artificial intelligence.
Digitalisation and Digital Transformation in Tax Administrations
Regional Trends and Insights
Abstract
Central to the future vision of tax administration as described in the 2020 report Tax Administration 3.0: The Digital Transformation of Tax Administration (OECD, 2020[1]), is a shift away from tax rules and calculations being executed within the tax administration’s own IT environment, towards embedding tax rules and relevant information for tax processing directly into taxpayers’ natural systems. Under this model, compliance is increasingly designed into business processes, enabling tax obligations to be calculated, reported, and in some cases settled automatically at the source. This approach is already well‑established in many economies through PAYE systems, where payroll software integrates tax rules directly into employers’ payroll processes.
Given the inherent complexity of tax systems, tax administrations are increasingly leveraging technology to simplify administration and improve accuracy. One important development in this context is the use of machine‑readable law, whereby legislation is translated into a structured, standardised digital format that can be processed by software used in taxpayers’ natural systems. This enables automated tax calculations, real‑time or near‑real‑time compliance checks, and seamless integration with accounting, payroll, and financial systems.
When applied in appropriate cases, machine‑readable tax rules can enhance transparency, reduce errors, and improve taxpayer services, while also enabling faster and more consistent implementation of legislative and policy changes. By reducing ambiguity and limiting divergent interpretations, they can also help mitigate risks associated with inconsistent or uncertain application of tax law.
As can be seen in Table 5.1., economies are at different points regarding the use of machine-readable tax law, with one third indicating that they publish all tax laws in machine-readable format and 14% indicating that they do this for some tax laws. The highest adoption is seen in Latin America and the Caribbean, where 63% of economies publish all tax laws in machine‑readable form, followed by Central and West Asia (60%). Those figures seem very high and might be a result of how administrations completing the underlying survey interpreted the question. Future work will need to confirm this.
Percentage of economies with the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
The Swedish Tax Agency (STA) has explored how the Rules as Code concept can be applied to facilitate tax compliance for small to medium sized businesses.
The STA now provides a technical solution that creates opportunities for business system developers to enable legal guidance in their systems. The STA creates rule-based machine-readable files, as well as written specifications in different areas of tax specific legislation supplied as open data. When used together with a rule engine, the rule files combined with existing data in the business system could create a more automated compliance process.
A key element of this work is that STA only makes the rule files available for use, and is not obliged to develop any application or services of its own. The internal work process enables legal experts within the STA to independently create rule files and make them publicly available without the need to involve technical assistance. This is made possible by the development of a rule editor that the experts can use to convert decision trees into machine readable code. To date, the STA has made rule files and specifications in three different legal areas as well as an Application Programme Interface available for use.
The concept, even though already available and in production, is still in a development phase and continuously improving with feedback from business system developers.
Source: OECD (2024), Tax Administration 2024: Comparative Information on OECD and other Advanced and Emerging Economies, https://doi.org/10.1787/2d5fba9c-en.
Alongside technological advancements and the widespread use of digital tools, a growing number of third parties are developing software products that support tax compliance, reporting, and calculation. While these products play an increasingly important role in modern tax administration, their use also introduces new risks related to consistency, accuracy, security, and trust. To address these risks, many tax administrations have established assurance frameworks or similar mechanisms.
These frameworks are designed to provide confidence that third‑party software applies tax rules correctly, consistently, and in line with administrative and legal requirements, producing outputs that can be relied upon or accepted by the tax administration. Assurance mechanisms can also support taxpayers and intermediaries in selecting appropriate software products and help maintain a level playing field among providers.
As can be seen in Table 5.2., 29% of administrations have a dedicated assurance framework with the highest percentage being in Africa (50%) and Latin America and the Caribbean (44%). The vast majority of those that do not have a framework have implemented other mechanisms to ensure that third-party software packages incorporate tax rules correctly:
62% of administrations provide technical documentation to enable product development by software providers. Providing technical documentation to software developers is most widespread in Europe, followed by the Latin America and the Caribbean.
58% of administrations implement controls around the authorisation and validation of information sent or received from those software products, notably high in Latin America and the Caribbean.
20% of administrations require software providers to follow a software development cycle, particularly in Africa (50%) and the Pacific Islands (43%).
Percentage of administrations with the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
Trustworthy, reliable, and secure software is essential to maintaining confidence in automated tax processes. As part of this, some tax administrations are increasing transparency by publishing lists of approved, certified, or recognised software products, as well as clear criteria for assurance.
Table 5.3. shows that overall, close to one third of administrations publish names of approved or recognised software products (31%) and / or maintain a corresponding register (32%). Where registers exist, the majority are public, although this varies by region. Central and West Asia shows the highest percentages among the regions, at 50% for both measures.
As regards the approval or recognition process for publishing the names of software products or for including them in a register, on average, certification by the tax administration is the most common approach (67%), particularly in Africa, Est and South Asia and Latin America and the Caribbean.
Other mechanisms, such as certification by another government body, are used less frequently, see Table 5.3. In addition, some administrations reported using other mechanism for approving or recognising software products, see the Inventory of Tax Technology Initiatives (OECD et al., 2025[2]).
Percentage of administrations with the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
Taxpayer registration and identification are foundational components of an effective tax system. They support compliance, enhance transparency and trust, and, in some cases, enable access to wider public services, benefits, or grants. Accurate and timely registration data underpins core tax administration activities, including assessment processes, withholding regimes, and third‑party reporting and data matching.
Automating registration and deregistration processes, or parts of thereof, at key moments in the taxpayer lifecycle – such as the creation of a business, entry into employment, or registration as a seller on a gig or sharing‑economy platform – can significantly reduce administrative burdens for taxpayers and businesses. Automation can also increase efficiency and data accuracy by reducing manual interventions and duplication, while ensuring that taxpayers are brought into the system at the appropriate time.
When supported by strong data governance and inter‑agency data sharing arrangements, automated registration processes can improve compliance outcomes and reinforce trust in the tax system. Safeguards are essential to ensure that automation is transparent, lawful, and responsive to changes in taxpayers’ circumstances.
Table 5.4. illustrates that around 60% of tax administrations can automatically register taxpayers based on data received from third parties, while 39% can also automatically deregister taxpayers. The highest levels are observed in North, South and West Europe, where around 70% of administrations automatically register and deregister taxpayers. Some tax systems go further and allow third parties to automatically register or deregisters taxpayers. However, in most cases this must be with taxpayer consent.
Percentage of administrations with the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
Withholding regimes are a powerful tool for supporting overall tax compliance while significantly reducing administrative burdens for large numbers of taxpayers. This is particularly effective where taxpayers are not required to undertake post‑payment adjustments, for example when the withholding is final or accurately reflects the taxpayer’s individual tax position.
Effective withholding often relies on third parties, such as employers, financial institutions, or digital platforms, applying the correct tax treatment at the time of payment. To do so accurately, these parties may need timely and reliable information on changes in taxpayers’ individual tax circumstances, such as tax rates, thresholds, or eligibility for reliefs. This highlights the importance of secure, well‑governed data flows between tax administrations and withholding agents.
As can be seen in Table 5.5., only 15% of tax administrations are providing such information on an automatic basis. The highest uptake is found in North, South and West Europe (26%) and Latin America and the Caribbean (25%).
Where such arrangements are in place, third parties most commonly calculate withholding based on the individual tax position of the taxpayer (around 63%), rather than applying a flat tax rate. Withholding under these arrangements is primarily driven by legislative requirements, reported by nearly all administrations that provide automated information, while voluntary arrangements play a more limited role.
Percentage of administrations with the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
AI is being increasingly adopted by tax administrations to improve both the effectiveness and efficiency of administrative operations, notably in areas such as compliance risk management. At the same time, AI is supporting enhancements in taxpayer services by enabling more timely, real‑time, and personalised interactions, helping tax administrations better tailor processes and communications to taxpayers’ individual circumstances. The data from the Inventory of Tax Technology Initiatives shows that nearly 49% of tax administrations use AI (OECD et al., 2025[2]), see Figure 5.1.
Percentage of administrations with the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
There are a range of uses for AI by tax administrations, as can be seen in Table 5.6. The most common is using AI to detect tax evasion and fraud, which is done by 74% of tax administrations. Other common use cases include AI as part of risk assessment processes (70%), virtual assistants (57%) as well as aiding officials in making administrative decisions (49%) and making recommendations for actions (45%).
Across regions, and in relation to administrations that use AI, Africa reports some of the broadest use across functions, with high adoption in risk assessment and fraud detection, followed by North, South and West Europe and Asia. Administrations from the Pacific Islands did not report any use of AI.
Less common uses of AI include automated provision of personalised information to stakeholders, ensuring the integrity of tax administration systems/processes, and dispute resolution. Some administrations reported using other approaches for software‑product approval mechanisms, see the Inventory of Tax Technology Initiatives (OECD et al., 2025[2]). The survey underlying the inventory also asked whether administrations use AI to make final administrative decisions, but none indicated that AI is used in this way.
Percentage of administrations that use artificial intelligence and have the respective process in place by region
Note: Percentages relate to the 49% of administrations that indicated using artificial intelligence. AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
Armenia – Detecting tax fraud and identifying high-risk taxpayers
By implementing an Artificial Intelligence (AI) & Machine Learning (ML) system, the Armenian Tax Administration (State Revenue Committee, SRC) aims to ensure the most efficient use of the accumulated legacy data and the continuously increasing volume of taxpayer-related information. The objective is to enhance the detection of tax fraud and to improve the identification of high-risk taxpayers, thereby strengthening risk-based audit selection.
For the development of the AI & ML system, a predictive gradient boosting trees model was selected. The model was trained using data from tax audits conducted in 2023 and subsequently tested on 2024 data. The results were validated against known outcomes from the 2024 tax audits.
The AI model was applied to a subset of 698 taxpayers identified through the SRC’s existing risk-based system. The assessment showed that within the first decile (285 taxpayers), the probability of fraud was 1.8 times higher compared to the remaining identified taxpayers.
These results demonstrate the effectiveness of integrating AI & ML tools into the SRC’s risk assessment framework and to support their further use in enhancing compliance and audit targeting.
Chinese Taipei – Generative AI-powered business, tax assistant with agentic workflows
The Business Tax Assistant system was developed to address critical pain points in business tax operations. Traditionally, tax officials have relied heavily on personal experience to conduct complex judgments and analyses - a process that is both labour-intensive and time-consuming. These challenges are often exacerbated by labour shortages and the steep learning curve faced by newcomers when resolving practical cases. Consequently, the integration of GenAI has become a high-priority expectation for modernising tax administration.
Designed using an agentic workflow, the Business Tax Assistant implements Retrieval-Augmented Generation (RAG) combined with Large Language Models (LLMs). This architecture mitigates the risk of AI hallucinations and ensures high accuracy by using natural language analysis to interpret tax laws and explain corresponding legal reasoning. To maintain decision integrity, the system adheres to a "human-in-the-loop" principle. While the AI performs the heavy lifting of screening keywords and cross-referencing statutes, legislative rationales, and administrative rulings, human officials provide the final review and verification.
Before this digital transformation, integrating diverse information sources into a coherent analysis typically required one to five working days. By simulating the professional workflow of a tax official through an AI agent, the Business Tax Assistant has revolutionised these operations. The system reduces overall processing time to under three minutes and achieves a 91% accuracy rate. This not only significantly enhances administrative efficiency but also ensures greater procedural consistency and service precision across the organisation.
Sources: Information provided by Armenia (2026) and Chinese Taipei (2026).
However, while AI offers significant potential benefits for tax administrations, its adoption must be guided by robust governance frameworks to ensure responsible and trustworthy use. Careful management is required to address ethical considerations, safeguard data privacy and confidentiality, and mitigate risks related to bias, transparency, and accountability.
Table 5.7. highlights that a large majority of administrations (85%) that use AI have put limitations in place on its use. This trend is consistent across most regions, particularly in Africa and Latin America and the Caribbean where all administrations that use AI report having such limitations.
In contrast, ethical frameworks for the application of AI are less common, with 40% of administrations reporting having such frameworks. Adoption varies significantly across regions.
Percentage of administrations that use artificial intelligence and have the respective process in place by region
Note: Percentages relate to the 49% of administrations that indicated using artificial intelligence. AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
One of the most significant risks associated with the use of AI is ensuring that its outcomes remain objective. To address this, tax administrations are putting in place a range of processes and controls. These processes are either carried out by the tax administration or by a third party, with Table 5.8. and Table 5.9. providing a breakdown by approach.
Around 80% of administrations indicated that they themselves carry out reviews of AI input information, the probing and testing of AI responses, and the monitoring of AI outputs. These approaches reflect a focus on validating the quality of the data feeding AI systems as well as stress‑testing outputs to detect bias or errors, indicating an emphasis on ongoing oversight once AI models are deployed. The review of the AI source code is done by slightly more than 50% of tax administrations.
Regional patterns show some differences. For example, all administrations in Central and West Asia report reviewing the inputs and outputs, and around 70% of administrations using AI in Africa, Central and West Asia, and North, South and West Europe report reviewing the AI source code.
Percentage of administrations that use artificial intelligence and have the respective process in place by region
Note: Percentages relate to the 49% of administrations that indicated using artificial intelligence. AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
The use of third parties to ensure that decisions by the AI are objective is less common across all AI‑using administrations. On average, the most reported measures are probing and testing of AI responses (21%) and reviewing AI input information (19%). This suggests that tax administrations largely retain responsibility for ensuring the objectivity of AI systems in-house, with external assurance playing a more limited role.
Percentage of administrations that use artificial intelligence and have the respective process in place by region
Note: Percentages relate to the 49% of administrations that indicated using artificial intelligence. AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).
For more information on the use of AI in tax administration, see also the sub-chapter “AI in tax administration” included in the 2025 OECD report Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions (OECD, 2025[3]).
With the management and application of tax rules being automated to a large extent and with vast amounts of data being internally and externally generated, tax administrations continue to look at new and innovative approaches to reduce burdens, provide certainty and ensure compliance.
Table 5.10. highlights some of those innovative approaches. For example:
Using IT systems that automatically detect errors or inconsistencies in the data provided by stakeholders is reported by 68% of administrations, with particularly high adoption in Europe and Central and West Asia.
Provide taxpayers or businesses with the opportunity to test upfront tax treatments of scenarios to inform their decision making and to allow them to understand their future tax liabilities, is reported by 31% of administrations and relatively similar across all regions.
Using Distributed Ledger Technology (DLT) in tax administration processes remains limited overall, with only 8% of administrations reporting this. However, 20% of administrations in Central and West Asia and in Latin America and the Caribbean report using DLT.
Percentage of administrations that have the respective process in place by region
Note: AFR – Africa, AS-CW – Central and West Asia, AS-ES – East and South Asia, EU-CE – Central and East Europe, EU-NSW – North, South and West Europe, LAC – Latin America and the Caribbean, PAC – Pacific Islands.
Source: OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 20 January 2026).
References
[3] OECD (2025), Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, OECD Publishing, Paris, https://doi.org/10.1787/795de142-en.
[1] OECD (2020), Tax Administration 3.0: The Digital Transformation of Tax Administration, OECD Publishing, Paris, https://doi.org/10.1787/ca274cc5-en.
[2] OECD et al. (2025), Inventory of Tax Technology Initiatives, https://www.oecd.org/content/oecd/en/data/datasets/inventory-of-tax-technology-initiatives.html (accessed on 15 June 2026).