AI is up and running. Trust is not.

Two-thirds of the compliance departments we surveyed already use AI in their day-to-day work. For many teams, though, it hasn't brought noticeable relief yet: they can't rely on the results and end up checking much of the output themselves. That is the finding of a survey of 85 senior compliance leaders at large European enterprises.

Compliance under constant pressure

Half of respondents face more regulatory obligations than two years ago. At the same time, headcount isn't keeping up: in 40% of departments, the team has not grown with the workload. Only 1% say their team has grown faster than the work.

By industry, the pressure is strongest in industrial companies. Obligations are rising there at a similar pace as in banking (65% in industry vs. 69% in banking). The difference lies in team size: in industry, 68% say headcount hasn't kept pace with the workload, compared with just under a quarter in banking and pharma.

It is also striking how evenly the effort is spread across very different areas of work. Risk assessment is named by 75% as one of their three biggest risk areas, yet it gets no more than 17% of their time.

AI isn't (yet) taking much off their plate

AI promises relief. But when it comes to compliance decisions, only 39% of respondents trust its output, almost half are undecided, and 13% distrust it.

So a lot gets re-checked. Three in four AI users manually verify at least a quarter of the results, and one in three verifies half or more. If AI prepares a third-party due diligence check and you then verify every other data point yourself, you save very little time. Paradoxically, the teams that re-check the most are the ones that are already short on people.

What teams require of AI

We asked compliance leaders what AI has to deliver for them to trust it and use it effectively. Out of many options, four requirements stood out.

Human-in-the-loop is essential: 68% want a human to give the final sign-off on decisions. Almost as important are decisions that can be traced back to the underlying rule, and a complete audit trail. Respondents also expect the AI to produce the same result for the same input. That last point carries real weight: if an AI gives different answers, 64% would rely on it much less.

For compliance teams, these are basic prerequisites. Yet the tools in use often lack them: only a quarter of users say their tools provide a complete audit trail.

Everyone is investing

Virtually everyone is investing in AI for compliance. 94% of respondents have already received or requested a budget for compliance automation or AI solutions. Over the next twelve months, 76% plan to introduce or expand AI, yet only 43% of them trust the results today. Investing now without settling the question of trust risks that AI mainly adds to the checking workload.

Our view: Without trust, these investments won't pay off. What matters when introducing AI

The budgets are signed off and the rollouts are about to begin. Anyone who doesn't first establish whether the results are reliable is mainly paying for one thing: additional checking work. Three things make the difference:

1. Make trust a product requirement. Answers that trace back to a rule, a complete audit trail, the same result for the same input, sign-off by a human. These belong in the requirements catalog, not on the wish list.

2. Build trust before the rollout. The trust question belongs in every test phase. Power users test the system under load and help improve it, so that trust is established before the broad rollout rather than assumed afterwards.

3. Put AI governance in place before you scale. Define which systems are in use, what they decide, and how their results are documented.

The next twelve months will decide whether AI relieves compliance teams or becomes a burden.

About this study

Commissioned by Bayshore, the independent research partner Potloc surveyed 85 compliance, legal and risk leaders at European companies with more than 1,000 employees in September and October 2026. Industry comparisons are based on small groups and should be read as directional.

The authors: Philipp Wiegand is founder and CEO of Bayshore. Leonard Smits works in GTM & Ops at Bayshore.