Short answer. Adoption is a delivery discipline, not a launch event. Roughly 80% of enterprise AI projects fail to deliver their intended value, and the cause is rarely the model. It is staff who never make the system part of daily work. Banks that succeed do three things before and after go-live: they train the people who use the system, they measure real weekly usage, and they give one named owner accountability for adoption. Ableneo ships 4 of 5 AI projects to production, and adoption planning is a large part of why those systems stay live.
A working model is about a third of the job. The rest is getting a credit analyst, a claims handler, or a call-centre agent to open the tool every morning and act on what it says. When that behaviour does not form, the project shows green in the delivery report and red in the value report. RAND’s review of more than 2,400 enterprise AI initiatives put the failure-to-value rate near 80%, roughly twice the failure rate of ordinary IT projects. McKinsey found around 70% of AI initiatives miss their intended outcomes, and in most cases the block is not technical.
In a bank the pattern is specific. A pilot proves a fraud-scoring model catches more true positives than the old rules engine. Six months later the fraud team still overrides most of its flags, because nobody redesigned the review workflow or explained how the score is produced. The model works while the adoption does not.
The banks that get past this treat adoption as a set of concrete, dated actions:
Ableneo shipped 34 production AI projects in 2025, and around 4 of 5 of the projects reach production. Clients run an average of 1.7 projects with the team, which happens only when the first system is used and trusted enough to justify the second.
For a bank in the EU, staff enablement is now a legal obligation, not a nice-to-have. Article 4 of the EU AI Act requires every organisation that deploys an AI system to ensure a sufficient level of AI literacy among the staff who operate it. The obligation has applied since 2 February 2025, and supervised enforcement starts on 2 August 2026. Breaches sit in the middle sanction tier, up to EUR 7.5 million or 1% of global annual turnover, whichever is higher. Untrained staff are no longer only an adoption risk. They are a compliance exposure with a number attached.
DORA adds a second reason. Operational resilience depends on the people who run a system knowing how it behaves and what to do when it drifts or fails. A frontline team that does not understand an AI output cannot recognise an incident, escalate it, or keep a critical function running. In Central Europe, where banks such as those Ableneo works with in Slovakia, Czechia, and Austria run on long-lived core platforms, adoption also means fitting the new system into workflows that staff have used for a decade. That is a change-management problem before it is a technical one.
Around 80% of enterprise AI projects fail to deliver value, and the usual cause is non-adoption, not the model.
Resistance is usually rational. Staff fear the tool is there to replace them, they do not trust an output they cannot explain, and the system adds a step without removing an old one. Surveys bear this out: a large share of employees quietly work around sanctioned AI tools, and a meaningful minority actively undermine initiatives they see as imposed. None of that is fixed by a better model.
The fix is to remove the specific cause. Show the team that the system handles the repetitive part of their work so they spend more time on judgement calls. Make every AI output explainable in plain language, with the reason for the score visible. Retire the manual step the tool replaces, so the net workload drops. When staff see the system as an ally that reduces drudgery and keeps them accountable for the decision, use follows.
Article 4 does not prescribe a curriculum. It requires measures proportionate to the role, the AI system, and the context in which it is used. For a bank that means a credit officer using a scoring model needs deeper literacy than a back-office user of a document summariser. A defensible programme covers what the system does, its known limits, how to read and challenge its output, and when to escalate. It is documented, role-specific, and refreshed when the system changes. The literacy record also feeds the human-oversight and accountability evidence that a DORA audit and an EU AI Act conformity file both expect.
Activity is not adoption. A login count tells you the tool was opened, not that it changed a decision. The metrics that matter are behavioural: the share of eligible cases where staff act on the AI output, the override rate and whether it falls as trust builds, time saved per case, and the number of weekly active users against the target the owner committed to. Track these weekly for the first quarter, then monthly. A rising override rate is an early warning that trust is slipping, and it is far cheaper to fix in week three than in month six.
The first quarter decides whether a system sticks. In the first 30 days, run the training on the live tool, appoint champions inside each user team, and publish the usage dashboard so the target is visible. In days 30 to 60, hold weekly reviews of override and usage data, and fix the workflow frictions the data exposes. By day 90, the system should have a stable base of weekly active users and a falling override rate, or the owner escalates with a specific remediation plan. Adoption that is not visible by day 90 rarely recovers on its own.
Ableneo treats adoption as part of delivery, not an afterthought handed to a training vendor. The team designs the workflow, the human-oversight model, and the staff-literacy plan alongside the system itself, which is why around 4 of 5 Ableneo projects reach production and why clients come back for an average of 1.7 projects each. The work spans regulated banks and insurers in Slovakia, Czechia, and Austria, where the EU AI Act and DORA make staff enablement a matter of record, not goodwill. You can see how this fits the wider delivery approach in the Ableneo AI Transformation FAQ.
Key takeaways
Planning AI in a regulated business? Ableneo takes systems from classification to governed production.