Kathy Gibson reports from the Microsoft AI Tour – Paying tax and submitting returns has become a relatively easy process for most South Africans.
This is due largely to the fact that the South African Revenue Services (SARS) is embracing artificial intelligence (AI) and machine learning to execute on its mandate.
“We don’t think AI is the future,” says Edward Kieswetter, commissioner of the South African Revenue Services (SARS). “In fact, we are playing catch-up – AI is alive and well.”
At more than 200 SARS sites, with employees using 14 000 individual devices and serving 14-million taxpayers, there is a tremendous amount of work that goes into making tax season efficient.
In the last tax year, the organisation processed 230-million messages, managing 6 600 databases and 1,2-billion records amounting to 30 petabytes of data on 3 000 servers.
“We could not do that without using data science and AI technology,” Kieswetter stresses.
SARS operates on the theory that most taxpayers are honest, he adds, which is why SARS embraces the international benchmark of voluntary compliance.
The first tenet of this is to steer away from trying to “catch people out” but rather making processes clear and simple to understand. The second tenet is to make systems seamless and easy to comply with; and the third tenet is a credible threat for non-compliance.
As an example, Kieswetter points out that in this fiscal year, five million individuals didn’t have to do anything about their tax returns. “They just happened. People received an assessment and, if they didn’t have other information to submit, they received a refund in under three days.
“Taxpayers who chose to do their own returns received assessments in under five seconds.”
Key to the efficiency of the process is the fact that SARS is able to harvest third-party data from other data sources – and analyse it using AI and other models.
The agency is also improving its interactions with taxpayers through its contact centre.
“We have always had a chatbot that is merged with Azure Cloud Services and enabled with OpenAI,” Kieswetter says. “This provides generalised guidance.
“We are now adding hyper-personalisation so every taxpayer can received individual responses based on their own records.”
So far, about 300 000 responses have been generated.
SARS is also experimenting with biometrics, which will make systems easier for taxpayers to use while increasing security for SARS.
“Bear in mind that the registration of a taxpayer is not risk-free for us,” Kieswetter says. “You constantly have to balance risk and service.”
He adds that implementing intelligent systems that deliver on their mandate requires organisations like SARS to have a clear vision and employ the right people to implement it.
“It is clear and ridiculously simple to understand what we are trying to achieve.”