Short answer. Most banks already use AI. Few can point to it in their results. The European Central Bank found that about 70% of euro area firms use AI, yet only 7% use it intensively, and only the intensive users show a measurable productivity gain. The gap is not the model. It is the workflow around it. Turning adoption into results means choosing a small number of high-value banking workflows, redesigning them end to end, and running them in production under clear governance. Ableneo ships 4 of 5 projects to production, which is where results are made.
Adoption is now the easy part. A bank can buy copilots, connect a chatbot, and spin up a dozen pilots in a quarter. Results are harder, because value appears only when AI changes how work actually flows, not when it sits beside the existing process.
Three patterns separate banks that see results from banks that only see activity:
Ableneo shipped 34 production AI projects in 2025, with 22 AI agents running in production, including more than 10 agents in a single Tier-1 bank. The pattern behind those numbers is the same every time: redesign the work, run it in production, measure one number.
For a bank, the adoption-results gap is also a governance problem. A workflow redesigned without controls will not pass an audit, and a model that cannot be explained will not reach production. Under DORA, in force since January 2025, AI systems that support critical functions need operational resilience, incident reporting, and third-party risk controls. Under the EU AI Act, credit scoring and insurance pricing are high-risk uses, and those obligations apply from 2 December 2027. A bank that builds results into its AI from the first workflow, with human oversight and an audit trail, answers both the board’s demand for a number and the supervisor’s demand for control.
This is why intensive, governed use beats broad, shallow adoption. The ECB’s analysis links productivity gains to intensive use in core activities, not to peripheral or routine tasks. In a regulated bank, the core activities are exactly the ones under the most scrutiny, so redesign and governance have to happen together rather than in sequence.
About 70% of euro area firms use AI, but only 7% use it intensively, and only intensive users show a measurable gain (ECB).
Because most AI is added to a process instead of changing it. A copilot that drafts an email faster saves a few minutes per person, but those minutes do not add up to a line in the results while the surrounding workflow, handoffs, and legacy systems stay the same. Researchers call this the micro-productivity trap: task-level gains that stall at the firm level. Banks feel this harder than most, because the work runs across siloed functions and a core system that is often 20 years old, so a faster step in one team just waits longer at the next handoff. The banks that escape it redesign the work. Industry surveys in 2026 report that high performers were close to three times more likely to have fundamentally redesigned their workflows, while only about one in five adopters had redesigned any workflow at all.
It means changing the sequence of steps, decisions, and controls across a complete banking process, not swapping one manual step for a model. Take credit. Redesign covers application intake, data gathering, scoring, the credit memo, the approval decision, and ongoing monitoring, with AI and people assigned to the steps each does best. The model drafts and flags. The person decides and signs. The audit trail records both. Redesign also settles where the data comes from and who is accountable at each step, because an AI step with no clear owner and no clean input is the step an auditor stops at. The aim is to remove handoffs and rework across the whole journey, which is where a bank’s cost and cycle time actually sit.
Pick one workflow that is high volume, rule heavy, and already measured. High volume means the saved time compounds. Rule heavy means AI has clear ground to stand on. Already measured means you can prove the result against a baseline you trust. Document validation, KYC checks, and parts of the credit process usually fit. Avoid starting with a workflow that touches a final customer-facing decision with no person in the loop, because the governance load will slow the first win you need to show the board.
Agree one metric per workflow before the build, measure the baseline, then measure the same thing in production. Time per case, error rate, or cost per case are the clearest. State the number in production terms, not pilot terms, because a pilot result rarely survives contact with live volume. A credit operations lead who can say a validation step now takes 30 minutes instead of 7 hours, at over 90% accuracy, has a board-ready result. Ableneo reached exactly that outcome on a document validation workflow for a regulated operator, 14 times faster after earlier suppliers had failed to deliver. A team that can only show a demo does not have a result.
Expect the first measurable result from a single redesigned workflow in one to two quarters, not years, when the workflow is scoped tightly and runs in production. Broad adoption programs that touch many processes shallowly tend to show activity for a long time and results never. The faster path is narrow and deep: one workflow, redesigned, in production, measured, then the next. Ableneo reaches production on about 4 of 5 projects, and repeat clients average 1.7 projects, which is the pattern of results compounding workflow by workflow.
Ableneo builds AI into the workflow and runs it in production, which is where results are made. In 2025 that meant 34 production AI projects and 22 AI agents live, including more than 10 agents in a single Tier-1 bank, delivered with the governance a regulated environment demands. The discipline is the same every time: redesign the work, ship it to production, measure one number. For the related question of moving a single use case across the line, see our guide on how to get a stalled AI pilot into production.
Key takeaways
Planning AI in a regulated business? Ableneo takes systems from classification to governed production.