How can AI help detect non-compliance and improve tax enforcement?

This post is part of a new series on intelligent taxation and AI, which brings together evidence, policy lessons, and challenges of using AI to strengthen tax administration. Explore the Tax for Growth initiative’s policy toolkit: Harnessing AI and data for tax administration, which synthesises lessons for governments and Commonwealth tax administrators.

How do we know if taxpayers are trying to evade paying their taxes? This is a major question for revenue authorities around the world, and being able to spot evaders without inadvertently punishing compliant taxpayers is vital for building trust with citizens and funding public services. 

This is where revenue authorities’ enforcement strategies can play a crucial role. Artificial Intelligence can help smooth the process of tax enforcement by spotting inconsistencies, identifying possible evaders, and making the process of managing and digitalising data easier.

Information is the cornerstone of tax enforcement

Regardless of AI adoption, high-quality information and data are at the heart of effective tax enforcement. For example, when taxes are enforced via audits, information on how money has been spent and collected will be the basis on which an auditor makes their decision on whether a taxpayer has been compliant. 

One of the reasons why VAT has become one of the most important taxes for raising revenue in low- and middle-income Commonwealth countries is because of the information and data that is naturally created in the VAT system. 

The expansion of other sources of information, such as verifiable third-party information, has also been very important for tax authorities to cross-check the information they collect.

How are Commonwealth tax administrators using AI to enforce taxes?

Through the IGC’s partnership with the Commonwealth Association of Tax Administrators (CATA), we surveyed 25 Commonwealth tax administrations to best understand how they might use AI in their tax enforcement. From the responses, we see that AI has the potential to be useful throughout the auditing stage of tax enforcement.

Figure 1: How Commonwealth revenue authorities are currently using AI in tax enforcement

AI can highlight inconsistencies in tax returns

In our survey, 50% of respondents stated that they use AI to detect tax fraud. For example, if a firm imported a small number of goods, but reported VAT invoices for more goods over the course of the year than it had initially declared at the border, AI could spot that inconsistency at a much faster rate than a human. However, the extent and effectiveness of using AI to flag tax fraud may vary drastically between countries, depending on their data landscape and data maturity.

Identifying characteristics of potential non-compliant taxpayers

AI is very good at spotting patterns. This means that revenue authorities can use machine learning and AI to identify characteristics of taxpayers that are more likely to be non-compliant. 

Why is this important? Identifying the characteristics of evaders can help authorities predict much more accurately whether a taxpayer is engaged in fraudulent activities. When we surveyed revenue authorities, over 30% said that they are already using AI to identify firms involved in fraudulent activities in some capacity.  

As AI improves and is used by more revenue authorities, identification of likely evaders will only improve, which should place a smaller tax burden on compliant firms. 

If compliant firms are less likely to be audited, it can lead to a positive cycle wherein more trust is fostered in the tax system. However, revenue authorities will need to consistently make processes easier and more transparent in order for AI to be an effective tool.

AI can help select cases to audit

Using AI for audit selection could be faster, more accurate, and lead to a higher ‘hit rate’ without impacting compliant taxpayers. 

In Commonwealth countries, around 30% of revenue authorities have some form of automated audit selection, but AI could be the next step towards a more evidence-driven selection method. 

In an ongoing study in Paraguay, a machine learning model that helped select which taxpayers at the border should be investigated was found to not only have a higher hit rate of selecting fraudulent cases, but also selected far fewer taxpayers to be investigated - showcasing how intelligent audit selection can lead to more effective enforcement. 

Lessons on adopting AI in tax enforcement from Ecuador and India

In Ecuador, machine learning was used to flag ‘ghost firms’ that were issuing fake VAT invoices, allowing their clients to falsely claim back input VAT. The algorithm identified the characteristics of the clients of these ghost firms, which helped tax auditors select cases to investigate and ultimately led to an 81% increase in additional taxes collected. 

Similar tools were also used in India to identify bogus firms. This led to a 15 percentage point increase in auditors’ hit rate in identifying non-compliant firms. 

There is potential for machine learning and AI to help detect and select firms to audit if these tools are trained on high-quality and accurate data, which were already present in India and Ecuador.

What can revenue authorities do with AI next?

AI can be a powerful tool for making tax enforcement fairer and more effective. For tax authorities to make the best use of AI in enforcement, they need to:

1. Determine what level of data maturity the revenue authority is at. A new IGC toolkit on harnessing AI and data for tax administration includes a checklist which can be used as a starting point. 
2. Build a strong information landscape. AI lives and dies by the information it is fed. Maintaining a strong data and information landscape will be an ongoing task for all Commonwealth tax authorities.
3. Continuously evaluate how effective AI tools are at enforcing taxes. AI is not magic, and it cannot solve all enforcement problems. It can be susceptible to biases, which can lead to unintended consequences. This is where evaluation will be key in identifying where it can be best used and where work will be needed to make it most effective.

Editors note: AI has the potential to help revenue authorities strengthen tax administration through better data management, more targeted enforcement, and effective audit selection. As the blogs in this series have collectively highlighted, governments seeking to modernise tax enforcement while maintaining fairness and accountability should therefore harness AI and machine learning tools - not as a standalone solution, but alongside high-quality data, transparent processes, and continuous evaluation to build more effective tax systems.

The International Growth Centre’s Tax for Growth initiative supports tax administrators and policymakers in generating effective approaches to make taxation work for development.

Explore Tax for Growth's research and country-focused evidence