The South African Revenue Service (SARS) is using computer algorithms, machine learning and other advanced technologies to ensure taxpayer compliance.
In its annual report published on Wednesday (27 October), the revenue collector said it had reviewed its methodology to detect risk and highlight cases where non-compliance was detected.
“SARS has significantly expanded the scope of detection, beyond data obtained through declarations, as well as the traditional third party data received which enabled the pre-filling of PIT returns, as well as auto assessments.
“Examples of such data sources include historical data on compliance behaviour as well as data regarding financial flows and assets held both locally and abroad,” it said.
SARS said it has also implemented several machine learning models that leverage multiple asset and income stream data sources to detect non-declaration and under-declaration.
“Thematic input is also considered to adequately reflect the impact of policy changes and relevant emerging dynamics in its risk detection process.
“Examples of this included abuse of donations in the context of Covid relief measures, fraud resulting from PPE tender contract awards as well as the flow of digital money.”
Finally, a more rigorous feedback loop was introduced to ensure that false positives are minimised, and assurance work is focused on where the most prominent risks are.
SARS said that the system has already detected several key issues, including:
- Taxpayers with a lifestyle that significantly exceeds parameters that are reasonably possible considering the taxable income they declare.
- Taxpayers with offshore investments that appear to have come from sources that were not declared as taxable income.
- Taxpayers who disposed of assets without declaring Capital Gains Tax.
- Taxpayers with income-generating assets, such as rental properties who do not declare this income as taxable income.
- Citizens with significant economic activity and assets who should be registered for tax but are not registered – more than 26,000 instances of citizens who had economic activity exceeding more than R1 million were identified.
- Taxpayers who claimed VAT refunds but failed to comply with CIT filing obligations or where a significant discrepancy of turnovers declared for different tax types were apparent.
- People who abuse the ease of registration for VAT and PIT with the deliberate intent to defraud the fiscus with refund claims.
- People who set up multiple fictitious entities to simulate business activity, and in so doing, attempt to disguise illicit economic activity.
“With the improvement of the risk engine rules, the percentage cases selected for verification has decreased from 21.57% to 14.11%. This equates to a 35% reduction in cases selected for audit intervention. With reference to the adjustment rate of selected cases,” it said.
“SARS improved from 35.10% to 47.79%. Total yield from automated risk engine activities for the year ending March 2021 was R56.9 billion, which was R3.9 billion higher than the R53 billion estimate.”