WPP ended 2025 with 9,389 fewer people than it started with, and told the market it expects to be growing again in 2027. Both of those things are true. The profit forecast is getting all the coverage and the headcount number is getting almost none, so this piece is about the headcount number.

I was reading Bloomberg, a piece by Chloe Meley published on 8 September, on the state of the ad industry. It is careful, well sourced reporting. Bloomberg Intelligence figures reported by Bloomberg staff show the media sector heading for the worst earnings growth in the S&P 500 this year. In Europe, the Stoxx Europe 600 media and entertainment benchmark is forecast to show no earnings growth at all while the broader market runs about 15% ahead. WPP, Publicis and Omnicom are all pointing investors at a better 2027.

I think the recovery forecast is right. My problem is with how it is being read. Most of the commentary treats a return to profit as a return to the way things were, and it will not be that. The profit comes back because the cost base was cut, and cost cuts of this size do not get reversed when demand improves.

Profit returns in 2027. The jobs do not return with it.

The number that nobody put in a headline

WPP reported full year 2025 in late February 2026. Revenue less pass-through costs fell 5.4% to 10.2 billion pounds, the worst result since the pandemic. Pre-tax profit fell 87% to 131 million pounds, after a 641 million pound goodwill impairment written mostly against Ogilvy, Grey and AKQA. Those are the numbers that got reported.

Here is the one that got a single line. Headcount went from 108,044 at the end of December 2024 to 98,655 at the end of December 2025. That is 9,389 people, in twelve months, from one company.

Then look at what WPP announced on 26 February 2026. Elevate28 collapses the group into four units, WPP Media, WPP Creative, WPP Production and WPP Enterprise Solutions, across four regions, all connected through an AI platform called WPP Open. The target is 500 million pounds of annualised savings by 2028, at a restructuring cost of around 400 million pounds. Cindy Rose, who joined as chief executive in September 2025 from Microsoft UK, described the shift as moving "from a holding company structure to a single company".

That sentence is a staffing plan. Here is what it means in practice. A holding company owns a set of agencies, and each one carries its own account teams, its own studio, its own producers and its own back office. A single company running one platform needs one of each. The duplicated layers between those two structures are where the 9,389 came from, and where the next 500 million pounds of savings has to come from as well.

Return to growth is a cost story

Every recovery being promised for 2027 is built on cost cuts first. The earnings coverage usually stops short of saying that, so here it is plainly.

Omnicom closed its acquisition of Interpublic on 26 November 2025. At its fourth quarter results in February 2026 it doubled the cost synergy target to 1.5 billion US dollars a year, of which roughly 1 billion comes from labour, through redundancies, offshoring and back office outsourcing. About 4,000 redundancies were announced in December 2025. On 28 July 2026 the company raised its full year organic growth guidance for core operations to 5.0%, with second quarter core revenue of 6.0 billion US dollars and an adjusted EBITA margin of 17.8%.

Put those events in order. Cut about 4,000 roles in December 2025, book roughly a billion dollars a year in labour savings, raise growth guidance seven months later. The better outlook and the redundancies are the same story. Guidance went up because the cost base came down. Omnicom cannot hire those roles back without giving up the synergy number it has promised investors, so the profit line and the headcount line move in opposite directions.

Dentsu is doing the same thing with less room to move. It cut 2,100 roles across 2025, has 1,300 more planned, and is targeting an 8% reduction in international headcount, roughly 3,400 people, to reach about 52 billion yen of annual savings by 2027. It suspended its dividend.

What actually got automated

This next part I can describe from doing it rather than from reading about it.

I write a brief in Claude. Claude talks directly to Adobe to cut and grade the assets, to Gamma to build the deck, to Google Drive to file it, to Runway to generate the video. I am not exporting files between six tools and re-uploading them. The tools talk to each other, and I supervise the judgement calls.

The plumbing that makes this work is called the Model Context Protocol. Anthropic published it as an open standard on 25 November 2024. OpenAI adopted it across ChatGPT and its agent tooling in March 2025. Google confirmed Gemini support in April 2025. It was donated to a Linux Foundation body in December 2025, and by early 2026 the public registry listed roughly 17,500 servers. In plain terms, it is a standard plug. Any AI system can now reach into any tool that ships the plug, and do the work inside it.

Now ask what that plug replaces. For as long as digital media has existed, the connective tissue between a brief and a live campaign has been people. Somebody resized the asset for nine placements. Somebody built the campaign in the platform. Somebody trafficked the tags, ran the QA, versioned the copy for three markets, pulled the numbers into a weekly deck, and sat on a status call explaining the weekly deck. Every one of those steps is an API call now.

The platforms have been building toward this from the other direction. Meta has disclosed that more than one million advertisers used its generative AI tools to create more than 15 million ads in a single month, and estimated a 7% lift in conversions for businesses using its image generation. L'Oreal's chief executive Nicolas Hieronimus told analysts that the company's in-house generative AI content lab had cut production costs by 40%, with staff producing 50,000 marketing assets through it. Vaibhav Agarwal, chief financial officer at RingCentral, put it to investors on a June call, reported by Bloomberg staff: "we don't need external agencies".

Take away the file moving and most account management has nothing left to manage.

Why an Australian bank still rings Publicis

Take the question to the local market, and I will flag this as my inference rather than reported fact, because the Australian holding company arrangements are commercially confidential and I have verified figures for the market rather than for individual banks.

An Australian bank spends the overwhelming majority of its media budget inside two or three self-serve platforms. The ACCC's Digital Advertising Services Inquiry, final report August 2021, found Google held a share exceeding 70% at every stage of the ad tech supply chain in Australia, and 90 to 100% of impressions for publisher ad server services in 2020. IAB Australia data puts search advertising at around 8 billion Australian dollars in 2025, with digital taking roughly three quarters of the total ad market.

The bank already has direct accounts on those platforms. It has its own first party data, which is the entire basis of its targeting. It has a compliance team that has to sign off every claim regardless of who wrote it. So the honest question a chief marketing officer should be asking is narrow and specific. Which line on the agency invoice buys something the platform does not already give me, and which line buys a person moving a file?

There are real answers to that question, and I will get to them. Campaign setup is not one of them. Asset versioning is not one of them. Weekly reporting is not one of them.

The in-housing number is the wrong number

Everyone reaches for in-housing as the counter-story, so look at what the Association of National Advertisers, the ANA, actually found.

The curve is flattening. The ANA itself predicted penetration will top out somewhere between 85 and 90%. So in-housing on its own does not explain the agency headcount numbers. Most commentary stops at this figure and misses that.

The work is going into tooling, and the same tooling is available to the agency and the client. A brand does not need to build a large internal studio to replace an agency studio. From my own production work, a well designed workflow across Adobe, Claude and Runway, run by a small team, covers what a much bigger team covered three years ago. So the brand keeps a small team and adds tooling. Anyone assuming the agency roles will simply reappear on the client side should test that assumption, because the same workflow removes the need for most of them on both sides.

The trading floor already ran this experiment

This has already happened in another industry, and the outcome is documented, so it is worth reading across.

Share trading used to be men in coloured jackets shouting across a pit. The New York Stock Exchange rolled its hybrid electronic market out across 2006 and 2007, and the number of people on the floor fell to around 2,100 from about 3,000 within months. CME Group announced in February 2015 that it would close most of its Chicago and New York futures pits by 2 July that year, because open outcry had fallen to "just one percent" of its total futures volume. The remaining pits closed permanently in 2021.

Trading volume kept rising through all of that, and the exchanges became larger and more profitable than they had been in the pit era. What disappeared was a category of job. Those people stood between an order and the market and passed it along. The exchange still does that job every second of the day. Software does it now.

Travel agents tell the same story with a different ending. US employment in the occupation peaked at about 339,000 in 2000 on Decennial Census data, then fell roughly 60% as airlines cut commissions and online booking arrived. The agents who survived moved upmarket. They stopped selling tickets and started selling judgement, itineraries and access, and they charge more for it than they ever charged for a booking.

Map that onto an agency and the equivalents are specific. The floor broker is the campaign manager who builds the buy inside the platform. The ticketing clerk is the studio operator who resizes the asset for nine placements. The travel adviser who survived is the strategist a client rings before deciding what to do at all. The first two roles are the ones under pressure. The third is the one to move towards.

What the counter-evidence actually shows

Three facts get raised against this argument. Take each one properly and two of them support it.

Publicis is growing. It reported net revenue of 14.5 billion euros for 2025. Organic growth was 5.6%. Operating margin was 18.2%, a record for the group. It added about 5,800 people and now has roughly 114,000. It has guided to 4 to 5% organic growth for 2026.

That reads as a counter until you ask why. Publicis rebuilt itself around data and platform capability years before it had to. It is winning because it moved first, which is the argument in this piece rather than an objection to it. And here is my prediction, marked as a prediction. Publicis cuts roles in 2027. Once the platform is embedded, the same automation that removes work at a client removes work inside the agency running it. Revenue up and headcount down is the destination for all of them. Publicis is arriving at it from the profitable direction.

In-housing does not always stick. Intel built an internal agency and won In-House Agency of the Year in 2017. It shut the agency at the end of 2018 and went back to outside partners. Keurig Dr Pepper closed its internal agency Liquid Sunshine in May 2025, about 80 people, months after winning the same award for 2024. PepsiCo moved work to VaynerMedia in June 2025.

Read those as failures of in-housing and you miss what they are. They are procurement decisions about where the work sits. Intel chose to buy it outside because buying it outside suited Intel. The part that matters for this argument is what happened to the people. Liquid Sunshine closed with about 80 roles in it. The work moved. The jobs did not move with it.

Scale buys things a workflow cannot. Holding companies pool client spend and negotiate on the total. They take principal positions in media, which means buying inventory and reselling it. They indemnify clients. They carry the legal and brand safety risk when an ad goes wrong. My workflow does none of that, and this one is a genuine advantage.

It protects the company though, and not the jobs inside it. WPP holds every one of those advantages and still removed 9,389 people in 2025. Omnicom holds them and is taking out about 4,000. Scale is why these firms survive. It is also being funded by the layer underneath it.

So the split is between companies, and the direction on jobs is the same everywhere. Publicis and Omnicom are pulling ahead. WPP and Dentsu are cutting hard to catch up. All four are removing the same layer, whether the work ends up inside a client or outside it.

How this hits your business

Sort the invoice by function. Take your last twelve months of agency billing and sort every line into two buckets: judgement, and throughput. Strategy, creative concepting, negotiation and accountability go in the first. Campaign build, resizing, trafficking, versioning, reporting and status meetings go in the second. The second bucket is your exposure, and you should expect its market price to keep falling. Renegotiate it before your agency reprices it for you.

Learn the tools before you decide what to insource. The most expensive mistake available right now is hiring an internal team to do work a connected workflow already does. Spend a quarter with Adobe, Claude and Runway wired together on one real campaign, with your own people. Then size the team against what is genuinely left. Doing it in that order saves you the headcount you would otherwise have to remove in 2028.

Pay for judgement and stop buying throughput. If you run an agency, run the same sort exercise on what you sell. Every hour you bill for work a connected workflow can do is an hour whose price will keep falling. Reprice towards the work where a person makes a decision and carries the consequence: the strategy, the negotiation, the creative call, the compliance sign-off. Those hours hold their price because the client needs a name attached to them.

One more point, aimed at anyone currently running a campaign or studio team. The 2027 recovery will be real, and the company it arrives at will have a different structure. The skill that transfers is being good at directing these systems: writing a brief a model can act on, spotting when the output is wrong, and owning the result. Very few people have properly learned that yet. That is the gap worth filling.

Run one real campaign through a connected workflow this quarter. That will tell you where you need to invest, and what your actual agency budget is.