African Banks Are Spending On AI Before Measuring ROI

African banks are moving deeper into artificial intelligence, but the uncomfortable part of the story is that many are still spending before they can properly measure what the technology is returning.

A Backbase and African Banker report surveyed 277 senior financial executives across 37 African countries and found a sector in what it calls the accountability phase of AI adoption. Banks are not walking away from AI. In many cases, they are increasing budgets. But ROI measurement, legacy-system integration and governance are still uneven.

Fintech News Africa notes that 82 percent of respondents who lack formal ROI measurement still plan to expand AI spending over the next 12 months. That is the tension at the centre of the report. Executives see AI as strategically necessary, but many institutions have not yet built the measurement discipline to know whether the money is being spent well.

The report says 85.1 percent of institutions that do measure AI ROI say results have met or exceeded financial projections. That is encouraging. It means AI is already paying off for banks that track outcomes properly. But it also sharpens the problem for the rest of the sector: if measurement is what separates useful AI from expensive experimentation, then banks cannot treat ROI as an afterthought.

The governance gap is especially clear at the top. The article summarising the study says only half of C-suite executives are measuring AI ROI, compared with 82 percent of finance teams. That disconnect matters because the people approving AI budgets are not always the people left to explain whether those investments improved margins, fraud detection, lending, customer service or operational efficiency.

Legacy systems remain the biggest practical obstacle. Backbase says 50.2 percent of respondents cited legacy integration as the primary bottleneck, and close to 58 percent of banks that do not measure AI ROI said legacy integration is the greatest challenge to scaling AI internally. That makes sense. AI depends on clean data, connected systems and clear workflows. If a bank still operates across fragmented core systems, spreadsheets and manual handoffs, AI may amplify the mess rather than fix it.

The most useful banking use cases are also becoming clearer. Conversational AI is the most common entry point, but fraud detection and transaction monitoring are seen as the most impactful because the benefits are easier to measure. Credit scoring and alternative credit assessment are also important, especially in Africa where many people and small businesses still lack formal credit histories.

This links with a wider African fintech theme. We recently covered how Kenya is moving crypto firms into a licensing regime and how Cloud9 is combining business banking with social commerce. Across the market, financial technology is becoming more regulated, more operational and more accountable. AI in banking has to follow that same direction.

There is also a risk angle. Data privacy was cited by 48.5 percent of respondents, while risk and regulatory compliance followed at 40.2 percent. Those are not small concerns. Banks hold sensitive financial data, identity data and behavioural data. If AI systems are plugged into that environment without strong governance, the downside can include privacy breaches, unfair credit decisions, model drift and regulatory exposure.

The answer is not for African banks to slow down for the sake of caution. AI can reduce fraud, improve service, expand credit and cut operating costs. But the banks that will benefit most are likely to be the ones that measure from the beginning: what use case, what baseline, what cost, what risk, what operational improvement and what customer outcome.

African banking is not in the AI hype stage anymore. It is entering the stage where boards, regulators and customers will ask harder questions. The banks that can answer those questions with evidence will move ahead. The ones that cannot may discover that AI budgets can grow very quickly without becoming AI value.