Artificial Intelligence applied to auditing

Increasingly, Tax Administrations (TAs) use new ICTs to be more effective and efficient in their management, and the digitalization process has accelerated exponentially in the current circumstances.

Within this new technology, Artificial Intelligence (AI) presents multiple benefits for TAs, since it transforms data into a knowledge and impact asset for tax and customs management, and thus can achieve the intelligent use of such data and the way it interacts with taxpayers.

The combination of AI, Internet of Things (IoT), Data Analysis and Data Analytics, will give exponential benefits through the collection and analysis of a large volume of taxpayer data in real time for better decision making that will positively impact several administrative areas of the TAs.

Today, many TAs already use AI for information and assistance, with virtual conversation assistants and chatbots[1]. In the collection function, AI is used to predict the collection, in customs at airports with facial recognition systems, among many other uses that will surely continue to be enhanced in the future.

In this commentary, I would like to share some concrete examples of AI applied in audits or audits, both in massive or extensive controls and in intensive controls.


It can be stated that auditing is the action through which the TA seek to prevent taxpayers from incurring in tax evasion or fraud and, if they do commit such evasion or fraud, they seek to detect it, prove it and settle it.

Its objective is to maximize subjective risk and thereby modify taxpayer behavior so that voluntary compliance is increasingly achieved. This is due to the fact that the percentage obtained as direct collection from inspection actions does not exceed 2% to 3% of the total collection.

Generically, AI can be used to analyze relationships between taxpayers to identify hidden or simulated relationships or potentially high-risk tax non-compliance networks, which can generate new sources of information for selection rules that are not obvious.

Also, in risk analysis processes, to analyze more complex tax credit applications or in Customs to analyze import and export declaration forms.

AI can be applied in audits, which allows time reduction since there is real-time information.

By combining an audit with AI (which employs automatic learning and algorithms), the classification of transactions is done automatically, providing a detailed report of possible risks.

One possible use of AI in audits is also to compare companies’ pricing structures to obtain more accurate transfer pricing.[2]

Going to the specific cases of its application, the Norwegian TA (NTA) uses data analysis and automatic learning techniques to improve efficiency in the selection of the cases to be inspected.[3]

The algorithm is trained with historical data to predict the possibility of errors in each VAT return. Each case is assigned a score and tax officials begin inspecting taxpayers with the highest scores.

The more declarations are audited, the more data the algorithm will obtain for using in the model, thus improving its accuracy. The percentage of successful inspections practically doubled in relation to the manual process.

In France[4], almost a quarter of the tax audits carried out in2019 are the result of AI algorithms, where 11 billion Euros were collected after controls, an annual increase of 30% compared to 2018.

Chile’s Internal Revenue Service (SII) is using AI to study the notes employees take when answering the taxpayers ‘questions and test which combinations are most likely to yield a taxpayer who will default.

Peru’s SUNAT[5] for the electronic control of the General Sales Tax (IGV) sends alerts to taxpayers through text messages when invoices have been received for expenses or costs that the AI systems consider unusual for the type of business that has made the purchase.

In Colombia, DIAN[6] seeks to connect national ports, through technology with AI systems and robots, to detect evasion and verify commercial operations.

In Costa Rica,[7] through the use of Big Data, tax collection was improved with a predictive data mining model that detected the simulation of payments to third parties for more than $31 million.

In Brazil, the SISAM[8] is an AI system that learns from the history of the forms submitted and estimates the probability of around 30 types of errors that may occur in each line of each new import declaration form, and calculates the expected value of the revenue for each error detected.

In the United Kingdom, the HMRC developed since 2017 the Connect System, a computerized data mining system of social network analysis that crosses the tax records of companies and individuals with other databases to establish fraudulent activities. It basically looks for the correlation of the declared income with the lifestyle, comparing it with multivariate statistical models using AI.

Data comes from a variety of sources[9], including banks, land registry, credit cards, vehicles, municipal taxes paid, VAT registration, tax returns, tax investigations, employer income, online platforms, social networks, web browsing and email records.

Since 2017 Finland has been developing robots for tax audit processes, which will allow – potentially – to reduce the workload of these tasks in 52 years of effort per person, and to improve the quality of work and reduce errors.[10]

In Canada[11] an AI solution was developed to evaluate whether a person should be classified as an employee or an entrepreneur in a given business context. It not only presents an answer with its respective percentage of confidence, but also the reasons why such answer was reached (legal provisions, jurisprudence and comments that justify it).


I am convinced that AI is here to stay and in the coming years, it will be used permanently in auditing for signs of fraud, patterns of behavior, or organized evasion practices.

In order to deal with the most complex cases of fraud, TAs must process large amounts of information from different sources and not necessarily consistent in terms of format, language, etc. For example, information obtained from social networks is a new source of knowledge that offers great potential, and some TAOs have already incorporated it.

This information can also be analyzed with software with data geo-localization capacity and subjected to AI processes to subsequently generate products that, in the opinion of analysts, are a valid tool for managing the taxpayers’ risks and feeding the audit/investigation function.[12]

Information must be able to be exchanged with other institutions (prosecutors’ offices, FIU, police, customs, financial authorities, etc.) for complex investigations into illicit financial flows and the fight against the financing of terrorism thanks to the interoperability of systems.

I understand that we must never lose sight that ICTs are tools to obtain better results, that is to say they are not an objective in itself, the technology “as a fashion” is useless, we must always ask which will be the strategic objective of its incorporation.

Of course, AI also implies risks for[13] which specific regulation is required, especially for the adequate protection of taxpayers’ rights and guarantees.

It is clear that AI does not act by itself but depends on how it is “trained or programmed” by humans, which is why they are and will be responsible for its proper functioning.

In short, governments should work together with the different intervening actors to guarantee the adequate use of AI, in an ethical, transparent and equitable way, protecting the fundamental rights of citizens and always seeking that ICTs be an integrating element with human resources.

[1] To expand on this topic, see our commentary Virtual Conversational Assistants in the Tax Administrations: the future is today. CIAT Blog 22/09/2020.
[3] The ICTs as a strategic tool to leapfrog the efficiency of the Tax Administrations. Page 541
[5] CIAT ICT Book, page 326.
[8] CIAT ICT Book page 207.
[11] com/global/en/pages/tax/articles/artificial-intelligence-in-tax.html
[12] Information obtained from the book ICT as a Strategic Tool to Leapfrog the Efficiency of Tax Administrations. Page 213
[13] To expand on this topic, see Artificial Intelligence in TAs, benefits, and risks of its use. Alfredo Collosa. Mercojuris 22/09/2020.

Disclaimer. Readers are informed that the views, thoughts, and opinions expressed in the text belong solely to the author, and not necessarily to the author's employer, organization, committee or other group the author might be associated with, nor to the Executive Secretariat of CIAT. The author is also responsible for the precision and accuracy of data and sources.


  1. Sheila Kua Reply

    Great article. Very helpful

  2. Alfredo Collosa Reply

    Dear Sheila thanks a lot for your comment.

Leave a Reply

Your email address will not be published.

CIAT Subscriptions

Browse through the site without restrictions. Consult and download the contents.

Subscribe to our electronic newsletters:

  • Blog
  • Academic offer (Only in spanish)
  • Newsletter
  • Publications
  • News alert

Activate subscription

CIAT Members

Representatives, Correspondent and Authorized staff (TA)