Press
UnlikelyAI
AI was supposed to reduce burnout. For many, verifying AI outputs is causing it. AI The data from our AI Trust Ceiling research has been picked up by FT Adviser. It shows that the time saved by AI is being spent watching it.
The data
Across regulated industries, senior leaders spend an average of 2 hours 30 minutes every week manually verifying AI outputs, against 2 hours 41 minutes using it. The productivity gains that justified AI investment are, for most organisations, being almost entirely spent on the oversight that investment made necessary.
Moreover 77% of senior leaders manually check AI outputs before acting on them, which shows the scale of the verification burden. This is not the behaviour of the cautious few, it is what the overwhelming majority of senior decision-makers are doing right now, with tools their organisations have already invested heavily in.
Why verifying AI is not the same as checking someone’s work
When you review a colleague’s analysis, you can ask why they reached a conclusion. You can follow the logic, challenge an assumption, and find where the reasoning went wrong. When you verify an AI output, none of that is available. The model produces a result with no accessible reasoning behind it. Every check is a leap of faith dressed up as due diligence.
In regulated industries where decisions must be defensible to a board, a regulator, or a client that opacity is not a minor inconvenience. Large language models predict plausible outputs, they do not reason to valid ones. They cannot show their work, because they do not have work to show and it is the reason the ceiling exists.
“”Large Language Models have strengths in specific, limited areas, but there’s a huge lack of understanding about when to use them and when to look to other, less-fallible models. That’s where this trust gap is coming from.” — William Tunstall-Pedoe, CEO & Founder, UnlikelyAI”
What fixing this takes
We asked leaders directly: if AI were accurate and its reasoning fully auditable, what would change? Thirty per cent said they would increase their AI budget immediately. The appetite is not the constraint, the architecture is. Neurosymbolic AI combining neural language fluency with symbolic reasoning, produces outputs that are traceable and explainable at every step. It allows AI to be trusted to an extent when it does not need watching because it can show you why it is right.
The organisations that close this gap will not be those that adopt the most AI. They will be those that finally adopt AI they do not have to spend half their week second-guessing.
Not ready to speak to us? Start here.
See the sector by sector breakdown
