The Big AI Secret · ongoing research

The small agencies are ahead of the big ones

One of the more uncomfortable findings in our research, particularly if you run something large.

9 September 2026 · 6 minute read

An interviewee put it plainly:

“In fact the smaller agencies tend to be much more on it with AI than the bigger agencies.”The Big AI Secret interviews

It came up repeatedly, and it runs against what most people assume. Larger organisations have the budget, the specialists and the procurement muscle. They are behind anyway.

Why size works against you here

Four things showed up in the interviews, and none of them are about money.

The distance between the idea and the trying. In a small agency, someone has an idea on Tuesday and has tested it by Thursday. In a large one the idea meets a request process, a security review, a procurement cycle and a pilot framework. By the time it emerges the tool has been superseded twice.

Nothing to protect. Large organisations have working processes built over years, and a working process is something to be careful with. Small ones have less to lose and rebuild faster.

The person deciding is the person doing. A ten-person agency's decision-maker uses the tools themselves. They can tell whether something is working because they are the one it is or is not working for. In a large organisation that judgement is several layers from the desk.

Fewer handoffs to break. Integration is hardest where work crosses the most boundaries. A small team has fewer places for the work to stop and wait for someone.

Access was one of the six dimensions we scored, and it separated businesses more sharply than budget did. That is the finding underneath all four of those points.

What this does not mean

It does not mean small businesses are winning. Most of them are still Explorers, the same as everyone else. And the advantage has a ceiling, which is the part that matters most.

Speed gets a small agency to Implementer quickly, because Implementer is mostly about willingness to try things. Getting past it needs measurement and governance, and neither of those is a speed problem. They are a discipline problem, and discipline is where scale finally helps — larger organisations that do cross the line tend to cross it more thoroughly.

The pattern is a fast start and a hard stop, not a permanent lead.

If you run something large

The finding is not that you should be smaller. It is that the thing slowing you down is probably the distance between an idea and someone being allowed to try it. That is fixable, and it is cheaper than most AI budgets. A safe route from idea to trial, with rules people can name, would move more businesses across the line than another platform will.

What an access route actually looks like

Not a policy. A route — a defined path from somebody having an idea to that idea running, with the decisions made in advance rather than case by case. The organisations that had one shared five features.

A pre-approved tool list, and a real one. Three or four tools that anyone can use for defined purposes without asking. Not a list of everything under evaluation — a short list of what is already allowed. If the list is empty or theoretical, people will use whatever is open in their browser.

A data boundary anyone can state. Not a fourteen-page policy. One sentence people can repeat: what may go in, what may never, and who to ask about anything in between. If your team cannot say it out loud, it does not exist operationally.

A named person, not a committee. Someone who can say yes to a small experiment within days. Committees are how a two-week idea becomes a two-quarter one, and by then the person who had the idea has stopped having them.

A ceiling under which nobody needs permission. Below a defined threshold — no client data, no spend, no production output — people try things and report back. Above it, they ask. Most useful experiments sit below the line, and most organisations have not drawn one.

Somewhere the results go. A shared place where what was tried and what happened is written down. Without it, the same experiment is run four times by four teams who never learn of each other.

None of that requires budget. All of it requires somebody to decide, once, so that nobody has to decide again each time.

How to build one in a fortnight

Week one: write the data boundary sentence. One sentence, agreed by whoever owns risk. Test it by asking three people to repeat it back.

Week one: name the person. Someone senior enough to say yes, close enough to the work to know what is worth saying yes to.

Week two: publish the tool list and the ceiling. Three tools, one threshold, in writing, somewhere everyone can find it.

Week two: run one experiment through it. A real one, from someone who is not you. The route is only real once something has been through it.

That is the whole thing. It is not an AI programme, it is a permissions decision, and it will move your Access score more than any procurement exercise.

If you run something small

Your advantage is real and it is temporary. It buys you a fast start. What it does not buy you is the second half. Measurement and governance are what take a business past Implementer, and neither gets easier by being small. Most small agencies in our research had neither, and the ones that stalled had usually stalled for the same reason.

The ceiling arrives in a recognisable way. Things work, and then more things work, and then somebody asks whether any of it is actually making a difference and nobody can answer. Or a client asks what happens to their data and the honest answer is that nobody has decided. At that point the speed that got you here stops helping, because the next move is not another experiment. It is writing something down.

Measure one workflow before you need to. Not a measurement programme. One workflow, a before and an after, written down. It takes an afternoon and it is the difference between Measurement 0 and Measurement 1 — which is the gate on Transformer.

Write the data boundary before a client asks. The same one sentence a large organisation needs. It is easier to write when you have five people than when you have fifty, and having it early is a commercial advantage in any procurement conversation.

Keep the shared library from the start. Prompts, templates and what worked, in one place. At five people it feels unnecessary. At fifteen it is the thing that stops the same work being done twice.

None of that slows you down. It is what stops you stopping.

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.