Adoption is high and trust is low

All 270 CFOs in the Datarails survey use AI somewhere in finance. Very few let it near the outputs that matter. Only 4% trust it with the month-end close, and 5% trust it to produce board-ready reports without human review. Asked what holds them back, 75% named lack of auditability, ahead of accuracy.

A caveat on that source: Datarails sells finance software, and its sample is US companies with more than 1,000 employees.

The UK points the same way. In research Odoxa ran for Sixthfin, a close-software vendor, 67% of 303 UK CFOs said more reliable accounts was their top priority for the close, ahead of closing faster. Only 42% called their close process very satisfactory.

Finance leaders are using AI. What they won't do is sign under a number they can't trace back to its source.

Verification is now a cost line

The clearest European figure comes from IDC, in research commissioned by Sage. Across 2,275 finance leaders, respondents in Europe, the Middle East and Africa put verification at about 12.8 hours a week of senior finance leadership time. North America came in close, at roughly 13.1. IDC notes some respondents may have counted their own time and others their team's, so treat the figure as an order of magnitude.

In interviews, IDC found teams building templates, exception reviews and repeatability checks around their AI tools. The brief calls verification "a new category of finance work." It adds a second cost: about a quarter of the time AI saves goes into working out how the AI reached its answer. And 71% of leaders said they would veto an AI system that was 99% accurate if it couldn't show its reasoning.

Put plainly: if senior people check every output by hand, much of the productivity gain goes straight back into review.

The bottleneck is trust

What follows is our argument. The surveys describe the problem; they don't prescribe the fix.

The models are already good enough for a lot of finance work. What most teams lack is a trust layer around them. IDC's own conclusion leans the same way: it sees adoption held back more by trust architecture than by what the models can do. This challenge is part of the broader 'agentic gap' in AI governance, where accountability for AI agents remains undefined. For more on this, see The Agentic Gap.

A trust layer has four parts:

  • Controls built into each AI-assisted step, so a reconciliation or a variance comment is checked as it is produced.
  • Governed source data, so the AI works from the same numbers the ledger holds.
  • Exception rules that decide which outputs a person reviews and which pass through.
  • An audit trail an auditor or a board member can follow from the reported number back to its source.

Each process needs this designed once: the close, reconciliations, forecasting, the board pack. After that, the team runs it. That makes the work a bounded project with an end date, and it doesn't call for a permanent new role.

Why a fractional CFO should build it

The fractional CFO's job here is to define the rules and build the system that applies them. The checking runs inside that system, and people step in only on the exceptions the rules flag. That is how we set up Bridge AI Finance: the CFO writes the rules, the AI does the routine checks, and a person signs off on what gets escalated. Four reasons a fractional CFO is the right person to build it.

They have seen the failures before. Someone who has run closes at several companies knows where reconciliations break, which data feeds drift and which board numbers get questioned first.

They carry the authority. At some point someone has to tell a board that a number produced with AI is now reliable. That sentence carries more weight from a CFO than from a vendor or an analyst.

They build the system once and leave it running. The company gets senior judgment for the design and avoids a permanent salary for checking that a well-built system should do on its own.

The engagement has a clear finish. The process runs, the checks hold and the in-house team owns it. A finish line like that is easier to buy than open-ended advisory.

How to package the offer

Start with a promise a CFO can check. For example: "In 90 days I'll make your close and board reporting AI-assisted and audit-ready."

Scope it per process and price it on the result. Hours are an awkward unit for work whose whole point is to cut hours.

What the client receives:

  • A map of the controls in each AI-assisted step
  • A review policy that sets when a person checks AI output
  • The audit trail design
  • A playbook the team keeps running after you leave
  • A before-and-after baseline of verification time

How success is measured:

  • Hours spent verifying AI output each week
  • The number of exceptions that reach a person
  • Audit findings on the processes in scope

Measure the baseline before you change anything. Outcome pricing only holds up when both sides agree on the starting point, and a verification-hours number taken in week one is the easiest way to settle that.

Guardrails first

AI gives operators leverage once someone has designed the guardrails. Until then, it mostly hands senior people a longer review queue. The fix is better rules, built into the process, so fewer outputs need a human at all.

The fractional CFOs who build these layers and write down what happened will have a record few advisers can show today: verification hours before and after, exceptions that stopped reaching a person, audits that came back clean. Buyers will look for that record. So will the systems buyers are starting to use to find experts.