The Agency Business Model When Clients Do The Work Themselves

The agency business model has always assumed the client could not easily do the work themselves. That assumption is being withdrawn, and one of the largest agency groups in the world just said so in public.

On 1 September 2026, the Financial Times reported that WPP plans to cut up to 1,000 more roles by the end of 2026. The reason given contains two different problems joined by an "and," and the join is doing a lot of work, the same kind of blurred framing we unpacked when the Coinbase CEO gave every other CEO permission to cut headcount.

What WPP announced

• Up to 1,000 additional roles going by the end of 2026
• On top of roughly 11,000 already cut since the start of 2025
• Headcount of 97,388 as of 30 June 2026
• London offices consolidating from three buildings into two
• Non-core businesses being sold

The stated reason: reduced client spending, and increased use of artificial intelligence tools for certain tasks.

Why those two halves are not the same problem

Reduced client spending is a cycle. Budgets contract, budgets return, and everyone who has spent a decade in services has survived several rounds of it. It is painful and it is temporary.

Clients using AI tools for certain tasks is not a cycle. It is a transfer. Work that used to leave the building now stays inside it, and transfers do not reverse when conditions improve. Nobody rebuilds an internal capability and then dismantles it because the market picked up.

Putting both in one sentence makes the second sound as temporary as the first. It is not.

Which tasks are "certain tasks"?

Nobody has said, and the vagueness is the only part worth your attention, because in your business it is not an abstraction. It is specific people.

The vague category resolves to whoever produces the first draft, the resize, the variant, the localization, the recap, the version of the deck for a different audience.

The tasks above were never the reason a client hired you. All of them were frequently the reason the invoice was that size.

Creative tasks by whether clients move them in-house first or keep them outside longer

Moves in-house first	Stays outside longer
First drafts and variants	Deciding what to make in the first place
Resizing and reformatting	Being accountable for the result
Localization of existing assets	Judgment about what will not work
Status recaps and reporting	Telling a client something they do not want to hear
Producing the deck	Reframing the question the deck answers

Why in-housing is different this time

Clients have been threatening to bring work in-house for as long as agencies have existed, and mostly they have not, or they have done it and quietly come back. It is fair to ask why this round is different.

The old barrier to in-housing was never desire. It was fixed cost. Bringing production in-house meant hiring people you had to keep busy fifty-two weeks a year for work that arrived in bursts. Agencies existed partly to absorb that variability, and that is a genuine service.

What has changed is that a chunk of the production work no longer requires a person to be kept busy. A client can now handle the burst without carrying the headcount between bursts, which removes the specific economic reason they kept outsourcing it.

The shift is why this round sticks in a way earlier ones did not. It is not that clients suddenly became more capable. It is that the capacity they needed stopped being a person.

What clients still cannot do without you

Writing this as pure loss is easy, and that is not what the evidence shows. Publicis grew headcount from 108,179 at the end of 2024 to 114,079 at the end of 2025 while its peers were cutting, which means the pressure is not uniform and not fatal.

The parts that hold up are consistent across the firms doing well.

• Deciding what to make. A tool that produces content faster does not tell you which content is worth producing. That question gets harder as production gets cheaper, not easier.
• Being outside the politics. An internal team cannot tell a CMO their favorite campaign is not working. That is frequently the most valuable thing an external partner does, and it has no software equivalent.
• Pattern recognition across clients. You have seen twenty businesses try the thing your client is about to try. They have seen one. That asymmetry is not reproducible internally at any price.
• Accountability. Someone has to be answerable when the decision is wrong. Internal teams are answerable to the person who made the decision, which is not the same thing.

The four qualities above are not production. All of them were historically bundled into a price justified by production hours, which is exactly the bundling that is now coming apart.

How the conversation usually arrives

The conversation is worth rehearsing, because it rarely announces itself and it is easy to mishandle in the moment.

The conversation does not start with "we are cutting you." It starts with a question that sounds administrative. Could you break the retainer down by activity. Could we see what the design hours cover. We are reviewing supplier spend across the board, nothing specific to you.

By the time that question is asked, someone internally has already tested whether they can do part of it themselves, and formed a view. The question is not information gathering. It is confirmation.

The firms that come through this well are the ones that opened the conversation first, on their own terms, with their own restructured proposal already on the table. The firms that struggle are the ones that answer the breakdown question honestly, line by line, and hand over a document that makes their own commodity tier easy to identify.

What this looks like at your scale

Look at the shape of what WPP did. It removed roughly 11,000 people since the start of 2025 and still employs 97,388. Nobody deleted the agency.

They deleted a slice of it, quietly, over eighteen months, while continuing to serve the same clients. That is what this looks like at scale, and it is what it will look like at yours.

The billable hours that leave first are not the ones anyone valued. They are the ones nobody ever defended, which is exactly why they go quietly and why the loss shows up as margin rather than as a lost account.

You do not get fired. The work just gets smaller, and it keeps getting smaller, and no single conversation ever happens about it.

WPP has 97,000 people and a public share price, so its version becomes a headline. Yours does not. It arrives as a scope conversation where the client already knows what they can now do without you, and has not mentioned it yet.

The window for doing this on your own terms is open while the client is still experimenting rather than deciding. Most firms notice it only after it has closed, which is why the conversation feels like an ambush when it finally happens.

What to do before that conversation

1. Audit the invoice line by line. Mark each line "hard" or "tedious." Be honest, because the client is already doing this exercise and they are not grading generously.
2. Reprice around the hard lines. If the tedious lines are subsidizing the hard ones, that subsidy is ending. Better to restructure the pricing yourself than to have it restructured for you.
3. Automate your own tedious tier. If work is genuinely specifiable, do it with tools and keep the margin. Defending the price of something your client can generate in a browser is not a winnable position.
4. Get in front of the scope conversation. Raising it first turns a demotion into a repositioning. Waiting means responding to a decision that has already been made.

Repricing without shrinking

The instinct when tedious work leaves is to replace the volume, taking on more accounts at the same rates to fill the gap. That usually makes things worse, because it adds coordination cost to a business that just lost the margin funding it.

The alternative is to accept a smaller volume of work at a higher rate, which sounds obvious and is difficult in practice, because it means turning down revenue while you still need it, the same repricing discipline covered in why the market just repriced the business of selling human hours.

Two things make the transition survivable. The first is doing it before the client forces it, so you set the terms. The second is being genuinely cheaper on the tedious tier by automating it yourself, so you are not asking a client to pay agency rates for something they know costs almost nothing to produce.

The second point is the one most firms skip, and it is what separates repricing from simply charging more for less.

Is the agency model finished?

No. Publicis grew headcount over the same period that others were cutting, which is a reminder that the pressure is not evenly distributed and not fully deterministic.

What is finished is the version of the model that billed for effort. When production and formatting were expensive, selling hours was a reasonable proxy for selling value. Now that they are not, the proxy has broken, and anything priced on it is exposed, the same repositioning we cover for agencies rebuilding their model around AI.

Go through your last invoice line by line. Which lines exist because the work is genuinely hard, and which exist because it was tedious and nobody on the client side wanted to do it?