The Big AI Secret · ongoing research

What we got wrong

Every piece of research has a shape it cannot see past. Here is ours, before anyone else points it out.

9 September 2026 · 5 minute read

We haven't spoken to everyone we need to

Edition 1 draws on over 1,500 in-depth interviews with business leaders across sectors. But more often than not, the person we spoke to was already leading on AI or marketing inside their organisation.

That is a specific vantage point. It is the view from the person most likely to know what is happening and most likely to be enthusiastic about it. We did not, in Edition 1, systematically speak to the people underneath them — the ones doing the work the AI is meant to be helping with.

We think that matters more than we realised at the time. Everything we have learned since suggests the real finding sits in the distance between those two views.

We measured intent more than outcome

Interviews are good at capturing what people believe, plan and intend. They are much weaker at capturing what actually happens.

When a leader told us AI was saving their team time, we recorded that. We rarely had a before-and-after number to put next to it, because in most cases one did not exist. Measurement was the weakest of the six dimensions across the entire study, and that is a finding about the research as much as about the businesses in it.

Marketing-led recruitment shaped what we heard

Our interviewees came predominantly through marketing and agency networks. That gave us unusual depth in one sector and a thinner picture in others.

When we report that agencies lead on adoption, part of that is a real finding and part of it is that we asked more agencies. We have separated those two things where we can. We cannot always separate them cleanly, and where we cannot, we say so.

Self-reported adoption runs optimistic

People describing their own organisation's use of AI tend to describe the best version of it. Not dishonestly — the enthusiast genuinely sees the enthusiast's view.

We have no way to correct for that in Edition 1. In Edition 2, we built a way to measure it by scoring interviews against evidence rather than assertion and comparing the two. That comparison does not exist yet. When it does, we will publish it whichever way it falls.

A snapshot cannot show movement

Edition 1 is one moment in time. It can tell you where 1,500 businesses stood in 2025. It cannot tell you whether any of them moved, in which direction, or why.

Nobody has measured the same businesses twice on AI integration. That is the gap Edition 2 exists to close, and it is why the Integration Index is now open to anyone rather than being a study we run and publish at people.

Why publish this

Because you would find it anyway, and finding it yourself is worse than being told.

More usefully: research you can see the edges of is research you can act on. A study that presents itself as complete is asking to be believed. One that shows you where it stops is asking to be checked, which is a better relationship to have with the people whose businesses depend on getting this right.

The full method is published at thebigaisecret.com/method. If you see a hole we haven't mentioned, we would rather hear about it.