All Case Studies
Financial Services

Safaricom / M-Pesa

AI-powered fraud detection achieving 89% F1 score and financial ecosystem serving 40 million active customers across Africa

AI Readiness AuditAI Pilot ProgramAI Transformation Program
89% F1 score
Fraud Detection Accuracy
40 million
Active Customers (Kenya)
KES 161 billion
Annual Revenue Contribution

The Challenge

Safaricom's M-Pesa processes over 37 billion transactions annually across seven African markets, serving 40 million active customers in Kenya alone. As transaction volumes surged, so did fraud threats — social engineering, SIM swaps, and account takeovers exploited the platform's scale. Traditional rule-based detection generated excessive false positives while missing evolving attack patterns, and fraud losses spiked 344% to KES 810 million in 2024.

Operating across Kenya, Tanzania, Mozambique, and other territories with divergent regulations and infrastructure constraints — intermittent rural connectivity, heterogeneous handset ecosystems from feature phones to smartphones, and data-sovereignty statutes mandating in-country storage — Safaricom needed AI solutions that could deliver real-time fraud scoring under unreliable network conditions while complying with each market's regulatory framework.

The Approach

Safaricom partnered with Amazon Web Services to deploy graph neural networks that analyze transaction relationships and user behavior patterns, achieving an 89% F1 score in identifying social engineering attacks. The system scans millions of transactions in real time using Amazon Bedrock and SageMaker, detecting SIM swap fraud, identity theft, and coordinated attack networks before funds are transferred.

To address infrastructure constraints, Safaricom implemented phone number masking across M-Pesa transactions, reducing the data available for social engineering and impersonation scams. The company also deployed geo-locking technology that restricts M-Pesa agent tills to specific geographical locations, dropping till-theft and asset misappropriation cases to zero in covered areas.

Beyond fraud prevention, Safaricom expanded M-Pesa into a comprehensive financial ecosystem with AI-powered products including Fuliza overdraft facilities, KCB M-PESA loans, and Ziidi money market fund investments. AI-driven customer service is targeted to handle 70% of customer resolutions by 2027, part of a broader $500 million digital transformation strategy positioning Safaricom as Africa's leading technology company.

Results

89% F1 score
Fraud Detection Accuracy
Graph neural network system built with AWS SageMaker achieves 89% F1 score identifying social engineering attacks across millions of daily M-Pesa transactions
40 million
Active Customers (Kenya)
M-Pesa reached 40 million monthly active customers in Kenya as of March 2026, with 10.5% year-over-year growth — the highest in four years
KES 161 billion
Annual Revenue Contribution
M-Pesa revenue grew 15.2% to KES 161 billion in FY2025, contributing 44.2% of Safaricom Kenya's service revenue across payments, credit, and wealth management

This is an industry case study based on publicly available information. Safaricom / M-Pesa is not a Pertama Partners client.

Want results like these for your organization?

We help enterprises across Southeast Asia design and deliver AI transformation programs. Let’s talk about what’s possible for your team.