According to a 2026 report surveying 277 banking executives across 37 African nations, 83.2% of institutions plan to increase their AI spending over the next 12 months. Yet nearly a third do not formally track the return on investment. More strikingly, among the lenders that do not measure returns at all, 82% still plan to expand their budgets anyway. Between dollar-denominated cloud hosting bills, legacy core banking integration bottlenecks, and an acute regional deficit of machine learning engineers, African lenders are doubling down on an expensive technology whose ultimate fiscal contribution is still an article of faith. In this edition of Delve into AI, EMMANUEL NWOSU examines the uncomfortable disconnect between operational speed and audited balance-sheet yield inside Africa’s multi-million-dollar banking automation race.
Loan documentation has historically been an administrative bottleneck, even within well-capitalised financial institutions. At Standard Bank, a vetting process that previously took roughly an hour now wraps up in about 60 seconds, powered by artificial intelligence.
The metric, disclosed in the South African lender’s Capital Markets Day filing in March 2026, applies to its commercial asset finance division. It serves as a flagship case study for how Africa’s largest bank by assets is operationalising machine intelligence.
Standard Bank reported that its conversational bot handles 65% of all incoming digital customer queries, while over 39,000 employees—approximately 72% of its total workforce—were regular enterprise users of generative AI by August 2026.
These workflows showcase the immediate operational promise of automation: administrative backlogs evaporate into single-minute tasks, customer friction drops, and bank staff are freed to focus on high-touch advisory roles requiring human judgement.
Yet the net financial dividend remains murky. Deploying AI across regulated financial environments requires rigorous data governance protocols, expensive software licencing, complex systems integration, and widespread workforce reskilling. For these outlays to be commercially defensible, banks must prove that bottom-line yield outweighs implementation overhead.
Industry data reveals an uncomfortable divergence between capital expenditure and performance tracking. According to The State of AI in African Banking—a 2026 joint report by banking software provider Backbase and African Banker, surveying 277 executives across 37 African nations—83.2% of institutions plan to increase their AI allocations over the next 12 months. Yet only 67.1% formally track the return on investment (ROI) of those deployments.
More strikingly, among the third of respondents whose banks do not track AI returns at all, 82% still plan to expand their budgets. African banking leaders are aggressively capitalising a technology whose commercial contribution they have yet to quantify.
This absence of measurement does not signify failure. Algorithmic systems can compress turnaround times, sharpen fraud monitoring, and elevate customer experience without immediately shaving costs from the income statement.
But the distinction between operational efficiency and financial return becomes harder to ignore as AI moves beyond isolated experiments and into systems that banks use every day.
However, as AI transitions from experimental sandbox pilots into core banking architecture, the distinction between unmonetised operational speed and audited fiscal return becomes impossible to ignore.
Faster work, uncertain financial returns
Standard Bank has been building its AI capabilities on technology investments made over several years. In its March investor presentation, the bank said conversational AI was handling 65% of digital queries on its personal banking app, reducing the workload associated with customer service.
It also reported that engineering productivity had improved by more than 20% in the previous year, while AI was beginning to accelerate work in lending. Its commercial asset finance example, in which loan documentation processing fell from roughly an hour to a minute, underscores the extent of the improvement possible in some administrative processes.
The bank has also expanded the technology across its workforce. In August, it reported more than 39,000 active users of generative AI, supported by approved enterprise tools.
Consider the loan documentation process. Reducing processing time from an hour to a minute frees up the employees responsible for that work. The bank may then be able to process more applications with the same workforce, reduce overtime, or redirect staff to other responsibilities.
Those benefits become measurable financial savings only when they result in expenses avoided or additional income that can reasonably be attributed to the investment.
Standard Bank said it plans to redirect the time employees save using AI towards customer relationships and work requiring more complex decisions. This suggests that the bank is looking beyond cutting operating costs to extracting more value from its existing workforce. On its own, it does not specify revenue generation or cost savings.
Sticking with Standard Bank’s lending business, AI could help employees process more loan applications without a corresponding increase in staff costs. If used to assess borrowers’ creditworthiness, it could also help the bank make better lending decisions and reduce the risk of defaults. These are two different benefits, and Standard Bank has not established whether its AI investments have delivered either financially.
The bank’s non-performing loan ratio stood at 5.6% at the end of June, down from 6.2% the previous year. Credit impairment charges, which reflect expected losses from loans, also fell by 12% to R7.13 billion ($435.5 million) in the first half of 2026. The lender has not attributed the improvement to AI usage.
At Nedbank, another tier-1 South African lender, the improvements are also measurable.
According to its 2025 integrated report, the bank said it implemented 30 advanced AI use cases during the year. Its broader programme, combining AI, generative AI, analytics, machine learning, and robotic process automation, helped save approximately 312,000 employee-hours across media and content creation, document translation, and personal productivity. Global payment processing times also fell by about 30%.
Nedbank has applied predictive models to identify customers more likely to purchase investment products, alongside analytics that help it target potential credit card and loan customers.
These uses illustrate how banks expect to earn returns from AI in two broad ways: spending less to provide existing services and earning more by serving customers more effectively.
The cost of making AI work
AI does not operate independently of a bank’s existing technology. Models need access to reliable data, secure computing systems, and the applications through which employees and customers use them. Banks also need processes for testing outputs, controlling access to sensitive information, and monitoring decisions made with AI.
Those requirements can make the expense of deploying AI considerably larger than the cost of a software subscription.
At its 2026 Capital Markets Day, Standard Bank reported total technology spending of R23.5 billion ($1.44 billion) in 2025, equivalent to 27.7% of its operating expenses. It said 88% of its legacy servers had been decommissioned and that 71% of computing workloads eligible for migration had moved to the cloud at the time of the presentation.
The bank subsequently reported that cloud migration had reached 78% of eligible computing workloads by the end of June 2026.
Standard Bank said its technology investments helped improve efficiency over several years, contributing to a decline in its group cost-to-income ratio from 59.1% to 50.2%. This means that for every R100 ($6.09) the bank earned in operating income, it spent about R50 ($3.05) on operating expenses, compared with roughly R59 ($3.6) previously.
Standard Bank’s technology spending has continued to rise alongside its expanding AI operations. The bank spent R23.51 billion ($1.43 billion) on technology in 2025, up 4.9% from R22.4 billion ($1.36 billion) in 2024. In the first half of 2026, its software, cloud, and technology-related expenses rose 6% to R7.16 billion ($436.5 million). The bank attributed the increase partly to expanding its AI capabilities, alongside investments in specialised technology expertise, core banking systems and cybersecurity.
Yet, its overall technology spending grew by only 2% to R11.83 billion ($721 million) during the six months, as lower amortisation and other technology costs offset part of the increase. Standard Bank has been following what it calls a save-to-invest strategy, using efficiencies from earlier technology investments to fund newer capabilities, including AI.
The bank’s investment in modernising its technology systems comes amid the backdrop of its AI-readiness ambition; it illustrates the work required to introduce new capabilities into an established banking business.
To prove this point, African banks face AI integration challenges. In the Backbase and African Banker survey, 50.2% of respondents identified integrating AI with existing technology as their biggest internal obstacle to adoption.
Banks may have to connect customer records, transaction histories, and other information stored across systems developed at different times. A lender introducing AI into credit assessment, for example, needs reliable access to the financial information used to evaluate borrowers. Bringing those records together can require upgrades to existing systems, additional computing infrastructure, and specialised employees to maintain them.
These requirements add to the initial and ongoing cost of using AI. Where banks purchase cloud services or software in foreign currencies, exchange-rate movements can also increase their expenses in local currency.
The financial case for AI depends on more than how much time a system saves. Banks must weigh those gains against the full cost of introducing and operating the technology, including the infrastructure needed to support it.
Why the returns remain difficult to measure
The Backbase and African Banker research suggests that some African banks are getting the financial results they expected from AI. Among respondents whose institutions formally measure returns, 85.1% said their investments had met or exceeded their projections.
The banks, however, were measuring performance against their own targets. The survey does not establish how much they invested, the financial value generated, or whether returns were calculated consistently across institutions.
Consider AI-powered lending. If a bank approves more applications after introducing AI, the additional loans could increase its interest income. But the financial benefit would depend on funding costs, the performance of those loans, and whether the technology improved lending decisions. Similarly, an AI fraud detection system’s value lies in the losses it prevents, not simply the number of suspicious transactions it flags.
The survey also found that 71.7% of respondents working with third-party AI providers measure returns, compared with 31% of those developing AI entirely in-house, underscoring how banks differ considerably in their ability to deploy proprietary AI—another cost centre.
In June 2026, Evident’s inaugural AI Index for Banks in the Middle East and Africa assessed 25 major banks on AI talent, innovation, leadership, and transparency. Standard Bank ranked second and Nedbank fourth, making them the only two South African banks among the four regional leaders. Other South African lenders, FirstRand, Capitec, and Absa, rounded off the top ten list.
Evident attributed their comparatively strong performance partly to larger pools of AI talent, proximity to research universities, and engagement with emerging central bank regulation.
The concentration suggests that South African banks have built stronger AI capabilities than many of their regional peers. Across the 25 institutions assessed, Evident found that specialist AI development talent, relative to workforce size, was only half the level recorded among its benchmark of major global banks.
This shortage has financial implications. Banks without sufficient internal expertise must either invest in building their own AI capabilities or rely on external providers, while still paying to integrate and maintain the technology.
Some banks outside Africa have begun providing a clearer account of the financial value they derive from AI. Singapore’s DBS Bank reported approximately S$1 billion ($783 million) in economic value from data analytics and AI in 2025, up from S$750 million ($587 million) in 2024. By the end of 2025, it had deployed more than 2,000 models across over 430 use cases.
DBS has gone beyond reporting how many tasks AI completes or how much time employees save. It has placed a monetary value on the technology’s contribution to its business, although it is unclear whether the Singaporean lender is now profitable on its AI usage.
The comparison brings the financial question facing African banks into clearer focus. Standard Bank and Nedbank have demonstrated improvements in productivity, while many institutions surveyed by Backbase report meeting their AI investment targets. But they provide little indication of how much individual AI programmes cost or what they return financially.
Productivity is a key use case for AI in banking, and across other non-financial organisations. But the opportunity here is establishing what that productivity is worth.
True scale demands moving beyond surface-level integrations to robust execution. We’ve filtered the noise out of Moonshot 2026, optimising the conference strictly for high-calibre connections between startup founders, global financial operators, enterprise leaders and individuals rewiring Africa’s technical frameworks. Get 20% off Early Bird tickets for a limited time.
