Nine Months Evaluating an AI Vendor While Collections Uses a Free Chatbot

A bank will spend nine months evaluating an AI vendor. Security review, model risk committee, procurement, a pilot in one department, a steering group to watch the pilot. While that is happening, somebody in collections is pasting customer statements into a free chatbot to draft follow up letters faster. Both of those things are true at the same time, in the same building. I have watched it in rooms across Lagos and beyond, and the second one almost never comes up in the first one. ## The thing about regulated data I call this BROAI. Bring Your Own AI, after the Bring Your Own Device problem that kept corporate IT awake a decade ago. Staff quietly using their own tools on company data because nobody told them which tools were approved. In financial services the material is account balances, transaction history, BVN and identity records, credit files, and the internal memos that discuss all of it. The Central Bank of Nigeria regulates what happens to that. The Nigeria Data Protection Act regulates the personal data inside it. Neither of those frameworks has a clause for a staff member trying to be helpful with a free account. What makes it harder in banking than elsewhere is that the people doing it are often your better employees. Slow processes push capable people toward shortcuts. Punishing them is the wrong response. Giving them an approved tool is the right one. ## Where the value actually is in this sector Most of the AI conversation in African banking has collapsed into customer service chatbots. That is the front door. The value is behind it. **Credit decisioning for thin file customers.** The majority of adults across the region have no conventional credit history. That is usually described as a problem. It is also the clearest opportunity in the sector, because alternative signals exist in enormous volume: transaction behaviour, airtime and data purchase patterns, mobile money flow, device and location stability. Fintechs like Kuda, Moniepoint and OPay grew by serving customers the traditional scorecard could not see. The modelling is not exotic. The willingness to act on it is the differentiator. **Transaction monitoring and fraud.** Rules based systems generate enormous false positive volumes, which costs you twice: analysts drown, and genuine customers get blocked. Pattern based detection reduces both. In a market with high volumes of instant transfer through NIP, the window to catch something is very short. **Collections.** Most collections operations treat every late account the same way. Some customers need a reminder, some need a restructure, and some are gone. Sorting them early changes recovery rates and costs almost nothing to try. None of those three is a chatbot. All three sit inside operations and change what the business can see. ## What good looks like One strategy rather than nine department experiments. A written list of approved tools. A named owner for each model who is accountable when it produces a wrong answer, because it will. Customer data staying inside systems the institution controls. That sequence is why SWITCH, the six phase framework I use with clients, opens with Scan and Score. Most institutions either overestimate their readiness or pick a use case because it sounded good in a conference session. You cannot buy your way past either of those. The institutions that sort this out will move faster on credit and lose less to fraud at the same time. The ones that do not will discover the gap through an incident rather than a competitor. One question for your next executive meeting. If a member of staff pasted a customer statement into a free AI tool this morning, what in your current setup would tell you? ## Frequently asked questions **What is the biggest AI risk for a bank or fintech in Africa?** Ungoverned use by staff. Customer financial data leaving the institution through free consumer tools that the business does not own, control or audit. **Where should a bank start with AI?** Usually fraud or collections, because both have clean historical data and a commercial result you can measure inside a quarter. **Can AI help lend to customers with no credit history?** Yes, and it is the strongest opportunity in the sector across Africa. Alternative signals carry real predictive power where conventional scorecards see nothing. **How does CBN regulation affect AI adoption?** It governs how customer data is handled and how decisions affecting customers are made and explained. Both need answering before a tool is selected, not after a pilot has run. **Does AI replace credit officers and analysts?** It changes what they spend time on. The work moves from processing toward judgement on the cases the model flags as uncertain. *Stanislaus Martins advises and trains enterprise teams across Sub Saharan Africa on the practical application of marketing, technology and AI. Formats are on the [speaking and training page](/speaking), or start a conversation on the [work with me page](/work-with-me). Sector regulation sits with the [Central Bank of Nigeria](https://www.cbn.gov.ng).*