The Big AI Secret · ongoing research

The gap is not adoption. It is integration.

Ask a room of business leaders whether they use AI and almost every hand goes up. Ask whether AI completes work in their business without someone starting it, and the room changes.

9 September 2026 · 9 minute read

That distance is the whole finding of The Big AI Secret, across more than 1,500 in-depth interviews with business leaders.

Adoption was never the hard part

AI has been adopted faster than any business technology in living memory. There was no procurement cycle, no implementation project, no training programme. Somebody opened a browser tab and started using it, and within eighteen months almost everybody had.

Which means adoption tells you nothing. It is a floor, not a signal. Every business you compete with crossed it at roughly the same moment you did, and none of you had to be good at anything to do it.

What separates businesses now is not whether they use AI. It is whether the work moves on its own.

The four stages

Which one are you? A short diagnostic

Answer honestly rather than aspirationally. Most people place themselves one stage above where the evidence puts them, which is itself one of our findings.

1 · Think of the last piece of AI-assisted work that mattered. Who started it?

A person opened a tool and asked — Explorer or Implementer. Something happened, a document arrived, a date passed, an enquiry landed, and the work began — Integrator or above.

2 · If the person who uses AI most left tomorrow, what would stop?

Quite a lot — Explorer. One workflow, and someone would pick it up — Implementer. Nothing, because it runs rather than being run — Integrator or above.

3 · Can anyone tell you what AI gave back last quarter, with a number?

No, but it feels faster — Explorer or Implementer. Yes, for at least one workflow — Integrator. Yes, by role and by workflow — Transformer.

4 · Somebody has a good idea for using AI on Tuesday. When could they try it?

They could not, or would not know who to ask — Explorer. They would wait for one particular person — Implementer. There is a route and it is slow — Implementer or Integrator. This week, safely, without asking — Integrator or above.

5 · If someone pasted a client document into a public chatbot this afternoon, would anyone know?

No — your stage is capped, whatever the other answers said. That last question matters more than it looks, and we come back to it.

The line that actually matters

That line — between pockets of use and work completing on its own — is where the difference sits. Everything before it is enthusiasm. Everything after it changes what the business can take on.

And it is not a technology line. The businesses either side of it are running much the same tools, often the same subscriptions. Nothing about the software explains which side you are on.

What separates them: three mechanisms

1 · The work starts itself

This is the single clearest difference, and it is smaller than it sounds. In an Implementer business, the sequence is: a person decides to do something, opens a tool, asks for it, takes the output, and puts it somewhere. AI is in the middle of the work, and a person is at both ends.

In an Integrator business, something arrives and the work begins. A payment date passes and the chase drafts itself. A document lands and the summary appears. An enquiry comes in and the record is created before anyone has read it.

The person moves from the start of the work to the end of it. They check rather than produce. That change is what makes AI compound. Work a person has to start is limited by how often they remember to start it. Work that starts itself runs at the frequency of the thing that triggers it.

2 · Somebody counted

Measurement was the weakest of the six dimensions across the whole study. Most businesses could tell us AI was helping. Very few could tell us by how much.

That matters more than it sounds, because you cannot expand something you have not measured. A leader who believes AI is helping will let it continue. A leader who knows a workflow returned a quantified amount of capacity will fund the next one, and the one after that.

Unmeasured adoption is the most common state in the entire research: businesses using AI widely and having no idea what they are getting back. It is also the most expensive, because nothing justifies the next step.

The fix is smaller than a measurement programme. Pick one workflow. Write down what it costs now — people, hours, frequency. Change it. Write down what it costs after. That is one afternoon, and it is the difference between Measurement 0 and Measurement 1.

3 · The rules exist, and people can name them

Not a policy in a folder. Rules people can actually state. Businesses without them were not reckless. They were stuck — because in the absence of clear boundaries, nobody will let AI near anything that matters. The result is AI confined to the low-stakes edges of the business, which is precisely where it produces the least value.

This is why question five in the diagnostic caps your stage regardless of the others. Without governance, integration cannot happen — not because it is forbidden, but because nobody will risk it.

One global agency in our research discovered that two-thirds of its AI activity was happening outside approved systems. Their CIO explained it:

“It wasn't malicious. It was impatience. People were trying to connect what the stack didn't.”CIO, global agency network

Shadow AI is usually a plumbing problem wearing a discipline problem's clothes. The businesses that solved governance did not solve it by restricting more. They solved it by making the approved route faster than the workaround.

Why this is better news than it sounds

If the gap were about technology, it would be about budget, and most businesses would lose. Larger competitors would pull away permanently.

It is not. Everyone has the tools. What separates the top fifth is whether one workflow was connected properly rather than ten being experimented with — and that is a decision rather than a purchase. The businesses furthest ahead did not do more. They did less, and they finished it.

What to do on Monday

Not a programme. One workflow. Pick the thing that repeats. Not the most interesting problem — the most frequent one. Weekly beats quarterly. The thing somebody groans about is usually the right answer.

Write down what it costs today. Who touches it, how often, roughly how long. Estimates are fine. The point is having a before.

Find the trigger. A date, a document, an enquiry, a status change. If you cannot name the trigger, the workflow is not ready — that is the work to do first.

Decide what stays human. Judgement, price, relationship, anything a client would remember. Write it down before you automate anything, not after.

Connect the first step only. Not the whole workflow. The first step, running from the trigger, with a person checking the output. Then measure it, and use the number to fund the second one.

That sequence is what separates the businesses in the top fifth from the ones just below it. It is not a bigger version of what Implementers are doing. It is a different shape.

The Big AI Secret is an ongoing research programme into how AI is really being used inside British businesses. Edition 1 draws on over 1,500 in-depth interviews with business leaders across sectors. The method, including its limitations, is published in full at thebigaisecret.com/method.