Technology in the SAT with AI

The Tax Administration System (SAT) uses Artificial Intelligence (AI) to increase tax collection, prevent tax evasion, and classify taxpayers.

---

In May, the tax authority reported that revenue in the first four months of 2024 reached 1 trillion 766,612 million pesos, an increase of 5.5% compared to the same period in 2023.

The SAT (Mexican Tax Administration Service) modernized its technology and made its tax collection processes more efficient with the 2024 Master Plan. This plan integrates, for the first time, the use of AI and machine learning to address tax evasion and be more accurate in identifying high-risk taxpayers.

Governments that have adapted to information technologies, digital platforms, and artificial intelligence systems have managed to increase tax revenue.

The SAT is developing graph analytics and machine learning models. These advanced AI tools allow for the analysis of large volumes of data and the discovery of complex patterns that would be impossible to detect manually.

The main objective of the SAT during the Fourth Transformation has been to dismantle tax evasion networks, such as the case of shell companies.

---

Through graph analysis, a map of related parties and taxpayer transactions is created, revealing networks of shell companies and simulated operations, thus helping authorities to identify key players and patterns of evasive behavior.

Inconsistencies can be detected in digital tax receipts (CFDI) linked to smuggling operations and shell companies. AI identifies patterns by analyzing transactions in real time, facilitating intervention by the authorities.

Meanwhile, machine learning technology is an Artificial Intelligence technique where the algorithm improves autonomously as it is trained, which serves to classify taxpayers in terms of risk.

By using machine learning algorithms, the SAT creates predictive models that assess the probability of tax non-compliance by a taxpayer, considering pre-programmed factors such as payment history, income patterns, and economic activities.

The classification allows the SAT to focus its audits and reviews on the highest risk taxpayers, optimizing resources and being more efficient in tax collection.

Artificial intelligence can monitor high-risk economic activities, such as fuel trading and the construction sector, where simulated transactions are common. It can also identify patterns that suggest fraud attempts when requesting VAT refunds. Furthermore, it can detect irregularities in import and export declarations to prevent smuggling, as well as identify fraudulent practices in payroll and pension management.

However, the use of these technologies raises a legal question for high-risk taxpayers, as it must be ensured that taxpayers' rights are respected, but risk classification leads to greater scrutiny and potentially to penalties if the algorithms are not accurate or if there are biases in the data.

To mitigate these risks, the tax authority must be transparent. As a public institution, AI models must be constantly audited to ensure that the data used to train the algorithms is representative and free of bias. Furthermore, legal procedures must be in place for taxpayers to appeal decisions based on artificial intelligence.

---

One of the biggest challenges in implementing AI for tax collection is preserving taxpayer privacy and respecting bank secrecy. Best practices in this area include data anonymization (the data used to train the algorithms is anonymized to protect taxpayers' identities), data encryption to prevent hacking or theft of sensitive information, and ensuring that only authorized personnel have access.

The SAT collected 4.51773 trillion pesos in 2023, a 12.3% increase. The 2024 Master Plan has the potential to improve tax collection, reduce evasion, and contribute to increasing this percentage. However, it is crucial that AI (Automated Tax Collection) does not stigmatize taxpayers and that they have legal certainty that it is implemented ethically and transparently.