Finance leaders lose AI gains to verification work

Sage has published research suggesting finance leaders are losing expected AI productivity gains to the work of explaining and checking AI outputs. The study surveyed more than 2,000 senior finance decision-makers and influencers across North America and EMEA.

According to the findings, 26% of finance leaders said time saved by AI is being absorbed by the effort required to explain results to stakeholders. Among organisations seeing the sharpest impact, 22% said more than half of all time saved through AI is being consumed by verification work.

The study points to a growing problem for finance teams that have adopted AI tools but still need to validate what those systems produce. Rather than removing manual effort, AI can add new layers of checking, reconstruction and internal justification when outputs cannot be easily traced back to their underlying reasoning or source material.

Researchers found that finance professionals spend nearly 13 hours a week on average reconstructing, validating and defending AI outputs. Globally, 48% said they spend 15 hours or more a week on verification, while 19% said they spend 30 hours or more.

Trust gap

The figures suggest trust, rather than model performance alone, is becoming a central issue in finance software buying decisions. More than half of organisations, 54%, said they would pay a premium for AI tools that offer greater visibility into how outputs are generated.

The research also found that 71% of finance leaders said a supplier's move towards what it described as Glass Box design principles would strongly or critically improve that supplier's standing as a preferred partner. The term refers to systems that let users inspect the logic, sources and reasoning behind AI-generated recommendations, in contrast to black-box systems that are harder to interrogate.

That matters in a function where audit trails, accountability and defensible judgments carry more weight than speed alone. The findings indicate that many finance teams remain cautious about handing over work to fully autonomous systems without clear mechanisms for review.

Only 3.8% of organisations surveyed were described as largely autonomous in their finance operations. Meanwhile, 62% remain manual or rules-based, indicating that widespread end-to-end AI autonomy in finance is still rare.

Finance caution

The report suggests this caution is not simply resistance to new technology. Instead, it reflects how finance departments are judged: a result that cannot be explained may create operational and governance risks, even if it appears accurate at first glance.

"In finance, almost right has always been wrong. As AI takes on more complex financial workflows, the cost of uncertainty is simply too high," said Aaron Harris, chief technology officer at Sage.

"This research shows that the next era of AI won't be won on raw model intelligence alone; it will be won on trust infrastructure. Finance teams cannot afford to spend hours playing detective with black box AI outputs. They need solutions that bring transparency, control, and traceability into the systems behind those outputs, so they can execute with absolute confidence," Harris added.

The survey was conducted by IDC and covered 2,275 senior finance decision-makers and influencers. Respondents came from companies with 20 to 1,999 employees across 17 industry verticals. Of those surveyed, 63% were based in North America and 37% in EMEA.

Buying criteria

The findings add to a wider debate over how businesses measure returns from AI investments. While many AI projects are assessed by headline gains in speed or automation, the data suggests those gains can be diluted if staff must spend significant time reviewing, explaining and defending machine-generated outputs before they can be used.

For software suppliers, that may shift the sales conversation away from automation alone and towards evidence, traceability and user oversight. In finance, those factors can determine whether a tool reduces work or simply moves it to a different stage of the process.

IDC said organisations that address those issues early may be better placed to scale their use of AI. "The organizations that will achieve the most durable AI advantage are those that reframe trust infrastructure not as a constraint on AI deployment, but as the foundation on which scalable AI is built. Organisations have a choice, act early to operationalise trust or risk becoming overwhelmed by verification overhead," said Kevin Permenter, research director, financial applications at IDC.