Executive View | AI, Banking Strategy, and Practical Transformation

1. Kenya does not need AI theatre in banking.

The Central Bank of Kenya’s 2025 AI survey shows the sector is interested in AI, but not mature enough for blanket transformation claims. Half of surveyed institutions said they had adopted AI, yet 70 percent still had no formal AI strategy. More tellingly, only 41 percent of AI adopters had implemented AI policies. That is not a sign that AI is unimportant. It is a sign that deployment has moved faster than governance.

In practical terms, that means Kenya should be careful about “AI everywhere” thinking. The sector is still building the foundations: data strategy, model explainability, auditability, third-party oversight, and human review for critical decisions. If those foundations are weak, the institution may automate error faster than it automates value.

2. The most practical AI use case in Kenya is credit risk, not AI branding.

This is the clearest signal in the research. CBK found that the top AI application already in use was credit risk assessment at 65 percent, and 83 percent of institutions said they were likely to adopt AI for credit risk assessment in future. That makes sense in a market where risk quality is under pressure and credit decisions increasingly need to be faster and more data-aware.

The case for practicality becomes stronger when placed against the wider sector numbers. KBA’s 2025 State of the Banking Industry report says the gross non-performing loans to gross loans ratio rose to 16.4 percent by December 2024, with gross NPLs reaching KSh 672.7 billion. In that context, practical AI is not an avatar in a banking app. It is better early-warning signals, smarter portfolio monitoring, improved collections prioritization, and cleaner credit segmentation.

3. Fraud, cybersecurity, and transaction monitoring are the next obvious priority.

CBK’s survey shows cybersecurity was the second most common AI use case at 54 percent, while fraud risk management stood at 40 percent. This is exactly where AI is practical: it can combine many weak signals, flag suspicious behaviour faster, and help institutions focus investigators where the risk is real instead of overwhelming teams with low-value alerts.

For the Kenyan market, this is especially relevant because digital channels are already central to the customer relationship. As transactions move faster and more of the banking experience shifts to mobile, fraud and cyber controls have to become more adaptive. That is where AI can be useful without becoming reckless.

4. e-KYC and onboarding are practical because customers already care about speed.

Not every AI use case is glamorous. Some of the highest-value ones are simply about removing friction. CBK found that 41 percent of institutions were already using AI in e-KYC, and 82 percent expected future adoption in e-KYC. KBA’s 2024 customer survey also showed that account opening simplicity matters, and that mobile app functionality is the single most prioritized banking feature at 42.86 percent.

That means practical AI in onboarding should focus on document extraction, identity consistency checks, exception routing, assisted verification, and clearer next steps for customers. The point is not to make onboarding feel futuristic. The point is to make it feel easier, faster, and more trustworthy.

5. Customer service AI is practical only when it solves, not when it performs.

Customer service is a real AI opportunity, but only if institutions are honest about what customers actually want. CBK found customer service among the top current and future AI applications. KBA’s 2024 survey adds the commercial reason: 56.49 percent of customers prefer self-service channels, while 47.3 percent said poor customer service was the reason they switched banks.

So the practical AI question is not whether to build a chatbot. It is whether AI can help the bank resolve issues faster, guide customers better, summarize cases for agents, improve complaint handling, and reduce the number of times a customer repeats the same issue. A chatbot without system integration is theatre. A service copilot that improves first-contact resolution is practical.

The practical AI test is simple: if the customer still feels delay, confusion, and repetition, the bank has not solved the problem.

6. Back-office AI is likely to matter more than many public-facing AI ideas.

Kenya’s banking sector still has many labour-heavy workflows. KBA’s 2025 report notes that, in microfinance banks especially, personnel and administrative costs make up the bulk of operating expenses and that cost-to-income ratios remain elevated. That is exactly why document-heavy and exception-heavy processes deserve attention.

Practical AI in operations can help classify complaints, summarize files, prepare internal reports, flag missing documents, support reconciliations, and guide staff through routine but time-consuming review work. This is less visible than a customer-facing AI launch, but often more valuable because it improves control, speed, and unit economics at the same time.

7. What is not practical for Kenya right now?

• Black-box credit decisions where the bank cannot explain why a loan was approved, priced, or declined.
• AI inside core ledger decisions without clear controls, audit trails, and manual override points.
• Generic chatbots that answer politely but cannot actually resolve customer issues.
• Replacing relationship managers outright instead of equipping them with better context and decision support.
• Large AI programs with no data readiness, no business owner, and no measurable operating target.

CBK’s own findings justify that caution. Only 56 percent of AI adopters had measures to ensure model explainability, only 56 percent had auditability mechanisms for machine learning models, and 93 percent of surveyed institutions said CBK should issue AI guidance. That is a strong signal that Kenya should prefer assistive and governed AI over uncontrolled autonomy.

8. So what should Kenyan banks do over the next 12 months?

1. Start with two or three business problems, not a grand AI agenda. Credit monitoring, fraud detection, onboarding, and service resolution are better starting points than broad “transformation” language.
2. Put governance in place early. Data strategy, model explainability, audit trails, human review, and third-party oversight should not wait until after deployment.
3. Measure operating outcomes. Track false positives, approval turnaround time, complaint resolution time, early-warning quality, and cost-per-case changes.
4. Keep humans in the loop for high-impact decisions. AI should strengthen institutional judgment, not replace accountability.
5. Use AI around the bank before using it at the bank’s center. Start with support, risk, and workflow layers before pushing AI into high-stakes core decisions.

Executive takeaway

Kenya’s banking sector should stop asking where AI looks exciting and start asking where AI is operationally justified. The strongest near-term use cases are credit risk, fraud and cyber monitoring, e-KYC and onboarding, service copilots, and back-office workflow support. Those are the areas where AI can be practical now because they align directly with the sector’s current pressures.

The institutions that win will not be the ones that say “we use AI” the loudest. They will be the ones that use AI carefully enough, clearly enough, and usefully enough to improve the economics and trust of banking in Kenya.