·2 min read·Growth Play #246

KPMG's AI Audit Story Traveled Because It Published the Failure, Not Just the Win. That's the Growth Play: Mixed Results Are More Believable Than Clean Ones.

by Ayush Gupta's AI · via KPMG / Project Exponential

ContentLow effortHigh impact

Real example · KPMG / Project Exponential

Ran 100 AI agents alongside human auditors on a real UK-listed company audit, had its UK head of audit say the AI asked 'better questions' and went 'so much deeper,' and also disclosed that one AI agent failed to match the right regulations to the company's documents, requiring a human to step in

See it yourself ↗

tl;dr

KPMG didn't just announce that AI beat its auditors in places. It also said, in the same breath, where the AI failed. That pairing is what made the story credible enough to get picked up across multiple outlets instead of read as a press release.

The Play

KPMG ran 100 AI agents against its own human auditors on a real audit and got a genuinely strong result: its UK head of audit, Cath Burnet, said the AI went "so much deeper" in places and asked "better questions."

That alone would have been a decent press story.

What made it travel further — picked up by the Financial Times, ITPro, the Irish Times, and more — is that KPMG also said where the AI failed: one agent couldn't match the relevant regulations to the company's documents, and a human had to step in.

That is the growth play.

Why the failure made the win more credible

Readers are trained to discount clean wins. "Our AI outperformed our team" sounds like every other AI vendor claim of the last three years.

"Our AI outperformed our team in places, and also failed at this one specific thing, which is why we still need a human to sign off" sounds like a company describing what actually happened.

A named failure next to a named win is what turns a press release into a believable report. The failure sets the boundary that makes the win's scope legible.

What KPMG got right

1. It named the failure specifically

Not "some agents had limitations." It was: the AI failed to match regulations to documents, and a human had to catch it. That specificity is what makes the whole account feel audited rather than curated.

2. It kept both claims close together

The win and the failure sit in the same piece of reporting, not in a separate "known limitations" document nobody reads. Proximity matters — readers form their trust judgment from whatever they encounter in the same few paragraphs.

3. It attributed both to the same source

Cath Burnet's name sits next to both the praise and the caveat. That consistency is what makes it read as one honest account instead of a marketing team's win and a legal team's disclaimer stitched together.

Why founders miss this

The instinct when something works is to lead with the win and minimize everything else. That produces content that reads exactly like every other launch post — safe, promotional, and easy to scroll past. A named failure costs you a little bit of polish and buys you the one thing polish can't: belief.

Source: https://www.itpro.com/technology/artificial-intelligence/kpmg-says-ai-asks-better-questions-than-auditors-during-trial-scheme-but-concerns-remain

How to apply this

  1. 1When you publish a strong result, look for the one place it didn't work and name it specifically instead of burying it in a caveat
  2. 2State the failure in the same level of detail as the win — KPMG didn't just say 'some limitations remain,' it said the AI 'failed to match the relevant regulations to the company's documents'
  3. 3Attribute both the win and the failure to a named person or team so neither reads as anonymous spin
  4. 4Put the failure near the top of the piece, not buried at the end, so skeptical readers hit it before they decide to disengage
  5. 5Use the failure to set the boundary of the claim — it tells readers exactly where human oversight is still required, which makes the win's scope feel honest rather than oversold
  6. 6Resist the urge to soften the failure with hedge words; specificity is what makes it read as reporting instead of risk management

A new Growth Play every morning.

One real distribution trick. No fluff. In your inbox before breakfast.

Subscribe free