How should my agency be using AI?

Start where the time is, which for most agencies is the work around the work: proposals, meeting notes, research, reporting, internal admin. Those are low-risk, quick to learn, and where the savings have actually shown up in agencies I survey. Be far more careful with the craft you are paid for, and make sure a human who understands the work checks anything that leaves the building. Then decide, deliberately, what the saving is for. Most agencies skipped that last step.

Where do we start with AI when it moves so quickly?

With your own use. The strategy can wait a month. Pick two or three jobs in your week and use it properly on them for a month. You will learn more about what it is good at, in your business, than any amount of reading, and you will be able to tell a real change from a passing one. The speed of the industry is mostly noise. The things that actually change how an agency works move slowly enough to keep up with.

Should I charge less now that AI makes the work faster?

Not by default, and never by accident. If the work is faster but the result to the client is the same, the value has not dropped, only the hours it took. To date, most agencies have kept the price and done more for it: 29 of 73 in the latest Agency Loop, the largest group by some distance. That is a price cut nobody chose. The better questions are what you sell, how you package it, and whether some of the saving becomes margin. Decide it on purpose, whichever way you go.

Do I need an AI policy for my agency?

Yes, and possibly two, both short. One is for your clients: where you do and do not use AI and why, and how you protect their data and their IP along the way. Clients are starting to ask, and a clear answer is becoming part of looking competent. The other is for the team: what is off-limits, such as client data in tools that train on it or anything published without a human check, and who can use it for what. Most of the problems agencies report come from that second one, usually someone trusting the output when they should not, at every level of experience.

How do I get my team to actually use AI?

Use it yourself first, visibly, and share what works. Teams copy what the founder does far more than what they say. Then find the willing people and give them a specific, boring job to improve rather than a mandate to "use AI more". Make it safe to say what did not work, because plenty will not. The sceptic on the team is often right about something, and worth listening to rather than converting.

Will AI replace my agency?

Some of what you sell, probably. Clients are already doing pieces of the work themselves, and some are being sold "AI agencies" that do nothing you do not. The agencies that get replaced are the ones selling hours of a task that no longer takes hours. Judgement, taste, understanding the client's business and being accountable for the result are harder to replace, and worth more when everyone can produce the raw output. The job is to move what you sell toward that, at a pace you choose rather than one chosen for you.

Making AI work in your agency | Mat Bennett

You use AI every day. So does most of your team, in their own ways, with their own tools and their own habits, and none of it is written down. Some of it is genuinely brilliant. Some of it you would rather not look at too closely. Clients have started asking what you do with it, or telling you what they now do without you. Meanwhile your feed says the whole industry is about to be rewritten, and has been saying so every week for nearly four years. You know this matters more than most of the shiny things that cross your desk. Working out what it actually means for your agency, and finding the time to act on it, is the hard part. It is a growing share of what founders bring me.

Has AI landed in your agency, or just arrived?

It might have just arrived if some of these sound familiar:

Everyone is using it, and nobody could tell you how, or with what.
The work got faster and the price stayed the same, and nobody quite decided that.
You have tried a few automations. Some work. The rest took longer than the manual job.
Someone on the team trusts the output too much, and someone else refuses to touch it.
A client has asked for your AI policy, and you do not have one.
You have lost work, or a lead, to “we can do that ourselves with AI now.”
You are further along than most of your team, and you are not sure that is a good sign.

My take on AI in agencies

I have been through one of these before. In the late ’80s I was in a design studio when desktop publishing arrived. It did not kill design. It killed the way we had been paid for it, changed who did which jobs, and rewarded the studios that worked out what to charge for next. The tools were the easy bit, and the industry took years to work out what it was now for. That is roughly where agencies are with AI, and I say that as someone whose own feelings about it swing about more than my feed does. I am not going to pretend to a settled view on where it ends. I do have a fairly settled view on what to do in the meantime.

Start with your own use. You cannot lead an adoption you have not done yourself. I have watched founders commission AI strategies, buy tools for the team and write policies, whilst their own use stalled at conversations with their favourite agent. The team notices. More importantly, you cannot judge what is worth doing in the business until you have found out, personally, what the thing is good and bad at, in your work, this month. That means using it properly for the parts of your job you actually do: thinking through a decision, preparing for a difficult client conversation, reading your own numbers, writing the proposal. Once that is real, the rest of the agency has something to copy, and you have the judgement to know what to copy.

The wins are mostly in the work around the work. The impressive demos are about the craft. The savings that have actually turned up are in proposals, meeting notes, research, reporting, and the admin nobody wanted to do. That is where I would look first, because it is where the time is, and because it is the lowest-risk place to learn.

Decide what happens to the saving. When the work gets faster and the price does not move, the client keeps the saving, or the tool vendor does. That is a price cut, and it is one most agencies made without ever choosing to. Whatever you decide about pricing, packaging or what you sell next, decide it. I have written about this in more detail in AI made your agency faster. Who kept the saving?

Ignore most of the noise. Every month there is a new thing the industry must adopt or die. Most of it is a phase. The changes that matter move more slowly and are more boring than the noise, which is good news, because it means you have time to get them right.

What the data tells us

I run a quarterly survey of UK agency leaders, the Agency Loop, and the most recent round asked specifically about AI. It shows an industry that has settled, and is only beginning to think.

Of 73 agencies, 37 have settled on shared tools and ways of working bolted onto their existing process. 12 are still ad hoc. 16 have redesigned how they deliver work to assume AI, and 6 are selling things they could not have offered before. Most agencies, in other words, have adopted the tools without yet changing the business.

The pricing question is where that shows. Asked what happened to the price when AI sped up delivery, 29 of 73 hold the price and do more for it, 20 have not worked out an answer yet, and only 15 of 73 have consciously turned any of the saving into margin. Asked about margin directly, 35 of 72 report no real change. Faster, busier, and no better off.

It has made the job easier for most: 41 of 73 say leading the agency is easier with AI, whilst 21 of 73 say it has given them more to do. Where the extra work comes from is consistent: checking output that looks finished and is not, working out who can safely use it for what, and experiments that went nowhere. In May 2025, 28 of 42 agencies had no written AI policy of any kind. Reading this year’s answers, the policies are being written now, mostly the hard way.

For all that, 52 of 73 feel positive about what AI means for their agency. So do I, on most days. The point is that the thinking has not caught up with the using, and the agencies that close that gap first will be in a very different position to the ones that never do.

What working on this with me looks like

I have been researching and talking about generative AI in agencies since 2022, before ChatGPT launched and presented some of my research around the topic at conference back in 2022. Since then I have helped a number of agencies and consultancies with adoption. I rarely do the building. My part is a considered perspective, judgement, a bit of demonstrating, and steady pressure.

We usually start with you - the agency founder. What are you using, for what, and where has it stopped? I will show you how I use it in my own work, which is a lot less exotic than the feeds suggest and a lot more useful, and we find the two or three things in your week where AI could bring you benefit straight away.

From there the order follows what the agency needs, and we take it one thing at a time across the two calls a month, with light contact in between. For some founders that is workflows and automations: picking the repeatable, boring jobs where a saving is real, and getting them built and kept. For some it is the team: guardrails, what is off-limits, how to bring along the sceptic and rein in the enthusiast, and a policy that is short enough to be read. For some it is the bigger questions: what to charge now the work is faster, what to sell next, what your service looks like in three years, and what to tell clients who are asking. Usually it is a bit of all four, in an order that suits where you are.

It stays your call throughout. I will push back when something is a phase rather than a change, and push harder when a decision about pricing or positioning is being left to drift. The aim is calm and control: an agency where you know what AI is doing, why, and what it is worth.

If that sounds like where you are, get in touch for an initial conversation. No pitch, just a proper talk about where AI has landed in your agency and whether I can help.

What changes

You use AI well yourself, so you can tell the difference between a real change and a shiny one. The team has a shared way of working with it, with limits everyone understands, and the checking has stopped being a hidden second job. The boring, repeatable work is quicker and the saving has been priced, packaged or deliberately passed on, rather than leaking away. When a client asks what you do with AI, you have an answer you are pleased with. None of that requires being first, just having decided.

Making AI work in your agency