Safaricom has reported an 87 per cent reduction in fraud after deploying artificial intelligence to identify fraudulent activity. The technology is shifting fraud detection from manual reviews towards real time monitoring and risk scoring.
The company reported that more than 1.4 billion cyber threats were detected and mitigated during 2025. The figure reflects the scale of digital activity being monitored across Safaricom's technology environment.
Fraudsters are adopting more sophisticated methods to target customers and businesses through digital channels. Safaricom is responding by using AI to examine transaction data at much greater scale.
Esther Wanjira, Senior Officer, Fraud Strategy & Analytics at Safaricom, said AI can analyse millions or billions of transactions simultaneously. This allows systems to detect subtle patterns that may be difficult for human investigators to identify.
The technology also helps Safaricom identify changes in customer behaviour and emerging fraud patterns. Grace Muhindi, Fraud Operations Officer at Safaricom, said AI has helped detect online Ponzi schemes faster than before.
Traditional fraud investigations often began after customers had already lost money through fraudulent activity. Investigators then had to examine individual cases and reconstruct how the deception occurred.
AI changes that process by allowing transaction data to be analysed continuously as activity takes place. Suspicious transactions can be flagged earlier, while reviewed cases can provide information for training fraud detection models.
The system does not operate without human oversight, since unusual transactions can have legitimate explanations. A sudden Ksh 50,000 transfer to an older customer could appear suspicious if the account has been inactive.
The transaction could still be legitimate if the money came from a close family member. Human investigators can therefore assess relationships and circumstances that automated systems may not capture.
This combination of automated monitoring and human review gives Safaricom a broader approach to fraud management. AI can process enormous transaction volumes, while investigators provide context before suspicious cases are escalated.
The reported reduction also points to a shift from reactive investigations towards preventative fraud controls. Instead of relying mainly on post incident analysis, Safaricom can identify unusual behaviour earlier.
The approach also allows detection models to adapt as criminals develop new techniques. Reviewed fraud cases can help improve the models and strengthen their ability to recognise emerging patterns.
Safaricom's experience shows that AI can extend the capacity of fraud teams without removing human judgement. Its reported 87 per cent reduction demonstrates the operational value of combining automated analysis with contextual investigation.
