South AfricaPublic sector2020 – ongoing

South African Revenue Service (SARS)

AI and machine-learning risk screening of tax refunds

SARS said its risk management work, which uses AI and machine learning, prevented R101 billion in impermissible refunds in 2023/24

WorkingCompany's own claim

Problem

Refund fraud and abuse is a large leak in South Africa's tax system. The commissioner said in April 2024 that he remained concerned about it.

Approach

SARS's compliance programme uses data, AI and machine-learning algorithms to counter criminality and wilful non-compliance. It says the systems are meant to stop impermissible and fraudulent refunds without denying legitimate ones. The agency has been rebuilding its capability, including data science, since 2020.

Results

R101 billionimpermissible refunds prevented in 2023/24 through SARS risk management work, per SARS; not split out by AICompany's own claim[1]

1.8 millioncases completed by SARS risk management work in the same period, per SARSCompany's own claim[1]

Takeaway

SARS used AI within a wider risk management programme to screen refunds. Its preliminary year-end total did not separate the contribution of AI models.

Sources

Compliance efforts of SARS bear fruit and underpin the positive revenue results despite the tough prevailing economic conditions
South African Revenue Service2 Apr 2024Government publication

Source summaries

[1]South African Revenue Service2 Apr 2024

SARS's year-end revenue release says its compliance programme uses data, AI and machine-learning algorithms against criminality and wilful non-compliance. It says SARS prevented the outflow of R101 billion of impermissible refunds and completed 1.8 million risk-management cases. The commissioner says he remains concerned about refund fraud and abuse.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026