The State of AI in African Banking 2026

The era of largely unconstrained AI experimentation in African banking is giving way to a harder discipline. Foreign-exchange pressures, rising dollar-denominated cloud costs and tightening data-localisation rules are concentrating board minds.

Responses from 277 senior banking leaders across 37 nations confirm a sector still committed to AI, yet increasingly insistent that it justify the capital being deployed against it. Which leaves the sector with one question it can no longer defer: ROI or no AI?

African banking’s AI experiment is entering its accountability phase. Foreign-exchange pressures are squeezing margins, with the pressure most acute where currency volatility has been highest. The Nigerian naira lost over 40% of its value against the dollar between 2023 and 2025; the Kenyan shilling hit multi-decade lows in the same period. Across Francophone West Africa, WAEMU-zone institutions face a different constraint: the CFA franc’s peg to the euro creates exchange-rate stability but limits the monetary policy tools available to manage dollar-denominated technology costs. Dollar-denominated cloud costs are rising. Data-localisation rules are spreading, complicating the cross-border architectures on which AI depends. Boards that once tolerated open-ended AI budgets are increasingly asking what they are getting for it.

The answer is less settled than the sector’s confidence suggests. African banks are building credit models, automating processes and installing conversational interfaces, but on different terms from their Western counterparts. Balance sheets are tighter. Talent markets are distorted by currency depreciation and persistent demand for remote work from Western technology firms. Legacy infrastructure predates the cloud era by decades in many cases. Ambition and architecture are not always aligned. This report surveys 277 senior executives across 37 nations to examine where that alignment is working and where it is not. It is the first systematic assessment of the return on investment of AI in African banking. The findings cover adoption levels and use cases, ROI measurement discipline, infrastructure constraints, governance maturity, and the structural factors that distinguish institutions that meet their AI targets from those that fall short.

What it reveals is a sector in transition: uneven in maturity, contradictory in some of its behaviours, but moving in one direction. For institutions that have built the discipline to measure what they deploy, AI is delivering. For those yet to build that discipline, measurement is the bottleneck.

African Banker conducted this survey in partnership with Backbase in Q2 of 2026. Responses were gathered from 277 senior banking executives across 37 African nations, spanning five sub-regions: West Africa, North Africa, East Africa, Central Africa and Southern Africa.

West African respondents, the largest cohort in this survey, skew toward the Early Adopter tier, with national commercial banks in the region heavily concentrated in this tier. Central African respondents show a higher-than-average rate of self-identified Innovators, which may reflect the presence of regionally significant pan-African institutions in markets such as the DRC.

Within the context of agentic AI adoption, African banks generally fall into one of three categories: early adopters, early majority, and innovators.

45.5% of respondents self-identified as early adopters, institutions that are characterised by programmatic, department-wide adoption at the workflow level. Meanwhile, 28% of respondents qualify as part of the early majority, piloting agentic AI across one or more teams and departments at the task level. Finally, a significant minority of 26.5% of banks surveyed are innovators, spearheading transformative adoption at the systemic/organisational-wide level.

Notably, at the highest level of AI adoption — those who have achieved innovator status — the difference between the type of institution is marginal: 29.9% of international-affiliated banks are innovators, compared to 29.7% of regional banks, and 28.1% of pan-African banks. At lower adoption levels, over half of regional banks and national commercial banks identified as early adopters. This suggests that although institutional structure is a contributing factor to AI adoption, it is by no means a defining variable. Institutional leadership and investment appetite play a major role in AI maturity.

In fact, institutions with higher levels of AI adoption are more likely to have formal governance frameworks in place. For example, 44.1% of Innovators have board-level involvement in the strategic direction of AI use within the institution, compared with 25.6% of Early Adopters and 29.2% of the Early Majority. Across all governance frameworks and processes, only 7.4% of Innovators reported a total absence of governance protocols, compared with 39.3% of Early Adopters and 23.6% of the Early Majority. This suggests that governance scales with AI maturity, indicating it likely serves a dual role as both a driver and a signal of deeper adoption.

What does AI adoption entail in practice? The survey reveals that although there are multiple use cases for AI in Africa’s banking sector, conversational AI serves as a universal entry point, with Innovators and Early Adopters — operating at both ends of the maturity spectrum — deploying it at roughly 50%, give or take a few percentage points. Meanwhile, operational efficiency/automation and risk management/fraud prevention are the two other most widespread use cases for AI, with 39.3% and 36.2% of respondents, respectively, reporting that they use AI for these purposes.

However, the true reflection of maturity lies in the diversity of AI deployment across different categories of adopters. Whereas Early Adopters concentrated their efforts on conversational AI (47.9%), followed by operational efficiency (31.6%) and risk management and fraud prevention (30.8%); Innovators have a much stronger distribution across multiple use cases, including advanced financial services (39.7%), credit scoring and loan automation (36.8%), and hyper-personalisation and marketing (27.9%). The differentiated AI portfolio is the Innovator’s signature.

Although conversational AI was the most popularly cited use of AI, the one identified as most impactful is actually fraud detection and transaction monitoring — which also has the most directly measurable return on investment. The second most impactful use case is credit scoring and alternative credit assessment, especially important for thin-file customers using mobile money data. The scale of the opportunity remains vast: while the World Bank’s Global Findex Database 2025 shows financial account ownership in Sub-Saharan Africa has climbed to 58%, it leaves 42% of the adult population entirely unbanked. AI-powered alternative credit scoring represents the most credible mechanism yet for bringing this population into the formal financial system at sustainable cost.

Finally, conversational AI and chatbots still emerged third among the most impactful AI use cases, with WhatsApp and USSD deployments cited across multiple institutions. An example of this operational shift is Nedbank’s integration of Kasisto’s KAI platform, which reduced live-agent chat volumes by more than half within just twelve months of launch. By absorbing high-frequency, low-complexity queries, the technology accelerates resolution times for the client while freeing up human capital to handle complex, high-touch interactions where human intervention remains vital.

The 3 most impactful AI use cases, as ranked by respondents.

The data on AI return on investment does not tell a simple story of progress. What it reveals is a sector in active contradiction: institutions that are simultaneously expanding AI budgets and failing to measure the returns, expanding governance frameworks and leaving the C-suite outside them, claiming infrastructure readiness while naming integration failure as their primary obstacle.

One of the most remarkable points of tension centres on the ROI of AI. At first blush, the data suggest a straightforward picture: a little over 67% of respondents across all AI adoption categories measure ROI. Among those who do measure ROI, over 52% report that returns exceed their expectations, over 30% say they are broadly on target, while a minority of just under 15% claim AI is not meeting their projected returns.

Meanwhile, the executives who are at the helm of investment decisions are the least likely to track their returns. Specifically, the C-suite is half as likely as Finance teams, which are responsible for managing profit and loss (82%), to track AI ROI. Simultaneously, Risk & Compliance professionals show the highest non-measurement rate (48%) after the C-suite, despite being the primary implementers. Although this is likely to be a governance issue rather than a structural one, it does complicate the understanding of how governance and AI maturity need to be sequenced — which is the cart and which is the horse? If governance does not underpin AI adoption, institutions risk racking up AI debts without returns.

Although AI ROI measurement follows maturity — Innovators track more than the Early Majority, who in turn track more than Early Adopters — the same logic does not apply to institution types. Pan-African banks score the lowest, with just over half of respondents reporting that they track the ROI of their AI investments. Regional banks perform best on this metric, with over 90% of respondents measuring ROI.

African banks plan to continue investing in AI, even in cases where there is limited evidence that existing investments are paying dividends. Arguably, there are two possible explanations: overconfidence that AI will serve as a panacea for existing limitations, or fear that investing in AI is the only way to avoid falling irrevocably behind. A third possibility is, of course, some combination of the two. Barriers to adoption exist at two levels — institutional and industry-wide. In both cases, legacy integration tops the list.

Aymen Daoud · Regional Vice President Africa, Backbase

African banking has rarely looked stronger on paper. Average return on equity across the continent’s banks reached 19% in 2024 and held at 17% in 2025, against a global average of around 10%, according to McKinsey’s most recent banking review. But look past the headline return, and a less comfortable picture emerges. African banks’ cost-to-asset ratio sits at roughly double the global average, and McKinsey is explicit that the recent improvement in cost-to-income ratios has come from revenue growth, not from any underlying gain in operational efficiency. Strip out elevated interest rates and foreign-exchange trading gains — which alone delivered Nigeria’s five largest banks over $1.7 billion in 2023, close to 40% of their total operating income that year — and the efficiency story looks considerably thinner.

That gap matters more now than it did five years ago because the technology banks are being asked to adopt is the one that punishes inefficient architecture the hardest. This friction is sharpened by three features of the African market. Currency volatility sits at the centre: operating costs for cloud hosting, software licensing, and AI compute are mostly denominated in dollars, while revenue is earned in naira, cedi, or shillings — every depreciation cycle quietly inflates the technology budget. Regulatory data sovereigntyis tightening rather than loosening: Nigeria’s central bank this month issued a circular requiring that payment data generated within the country be hosted and managed domestically, with compliance due by January 2027. Supervisors are shifting from observation toward active rule-making:South Africa’s Reserve Bank and Financial Sector Conduct Authority published a joint report on AI in the financial sector in November 2025 and are now consulting on folding AI governance principles directly into their Culture and Governance standard.

This is the context in which a new variable has entered banking: autonomous AI agents capable of executing workflows rather than merely answering questions. Backbase calls this shift Agentic Banking, and unveiled the architecture built to support it — the AI-native Banking OS — publicly for the first time at this year’s Backbase ENGAGE conference.

For decades, banks were designed around two actors, employees and customers, interacting through systems of record such as core banking, CRM, payments, cards, KYC, and fraud detection. Each system was built to do its own job well, in isolation from the others. Autonomous agents are a third actor, and they do not tolerate that isolation the way humans do. A human employee can mentally reconcile the fact that the core system, the CRM, and the fraud engine disagree slightly about a customer’s status and route around it. An agent acting on a write-permission across systems has no such judgment to fall back on, relying instead on whatever version of the truth it is given.

The implications for risk and compliance officers are concrete. If an autonomous agent approves a transaction or fails to flag a customer for enhanced due diligence, based on a CRM record that has not synchronised with the core banking ledger, the result functions as a KYC or AML control failure rather than a software bug — and the regulator will treat it as one regardless of which vendor’s model sat behind the decision. An institution cannot demonstrate auditable, governed decision-making to a regulator if the underlying systems cannot agree on what happened.

In practice, this is what the Banking OS is built to solve. It sits above the systems of record rather than replacing them, coordinating the work that currently falls between core, CRM, payments, and KYC. Every system and every actor, human or agent, draws on a single, consistent definition of the customer through Nexus, its semantic layer. A governance layer, Sentinel, authorises actions deterministically before they execute: no action runs, by any actor, without a Decision Token, producing an auditable record by default rather than as an afterthought. And workflows that must run identically every time are clearly separated from those that benefit from adaptive, judgment-based execution — the deterministic and agentic modes the Banking OS runs side by side. In time, regulators will require something functionally equivalent to this kind of immutable authorisation record for any AI-initiated action that moves capital or touches a customer’s record.

This remains, fundamentally, an issue of foundations rather than model capability, and it is one African banks are better placed to answer now, while the continent’s AI deployments are still young, than they will be in five years with a decade of point solutions layered on top of each other. The institutions that treat this as plumbing to fix before scaling agents will spend less, comply more easily, and be the ones still standing when the current generation of models is, inevitably, replaced by the next.

Aymen Daoud is Regional Vice President for Africa at Backbase, working with financial institutions across the continent on AI-native platform transformation.

The state of AI readiness and adoption within Africa’s banking sector is variable and complex, influenced by the interplay of several variables — some complementary, and others contradictory. That said, the presence of tension points suggests a state of evolution, rather than failure. If anything, the overarching conclusion of the study is that AI pays off, but only for banks that build the infrastructure to govern it. The vast majority of respondents who are measuring their ROI are seeing their expectations at least met, if not exceeded — but that result tracks closely with whether the underlying architecture was in place to support it.

Within this occasionally ambiguous landscape, who is winning? Which institutions are turning ROI measurement into ROI leadership? And what separates them from the rest? These questions will be the subject of the next report.