Four Agencies for One Set of Login Credentials
Take a look at the supplier bill of a mid-sized company – say £70 million in revenue, a marketing team of six. There's a media agency steering the Google and Meta budgets. Next to it an SEO agency with a monthly report. A web agency looking after the shop and the landing pages. And for the last two years, a performance agency testing creatives and optimising audiences.
What these four actually sell only becomes visible when you ignore the org chart and look at the systems. The media agency operates the client's Google Ads and Meta Business accounts. The SEO agency pulls data out of the client's Search Console and drops it into a deck. The web agency works inside the client's Shopify back end, wiring up HubSpot, Stripe and an email tool. The performance agency uploads 40 ad variants into the same Meta account the media agency is already operating.
Every one of those platforms belongs to the client. Every one has a public API. What the agencies sell is the operation of the access – plus the labour hours that sit between access and result.
Four months ago, in “Cutting Out the Middlemen”, we argued that many SaaS tools are just a surface over an API the client already pays for, and that AI is collapsing the glue-code cost that held the model up. This piece applies the same logic to agencies. That's uncomfortable, because nh labs is an agency. We're writing about our own business model here.
What an Agency Actually Sells
Historically an agency delivered three layers of value, and it's worth separating them cleanly:
- Access and expertise. Someone who knows how Meta Ads works, how to read a Search Console property, how to hook a shop up to the inventory system. The knowledge to handle the tools and their APIs.
- Execution. The hours in which that knowledge gets applied: setting up campaigns, building reports, writing integrations, varying ad copy. Billed on time and materials.
- Judgement and accountability. The decision about which campaign runs at all, which budget goes where, what to tell the board at quarter end. And someone whose neck is on the line when it doesn't work.
For decades all three layers were sold at the same hourly rate. That was convenient for both sides, because in practice the layers couldn't be separated: whoever could read the Search Console also had to put in the hours to read it.
AI is separating them now. The first layer – access to the tools – has largely collapsed. A marketing manager with no coding background can have a model write a script today that pulls the Search Console data every week, writes it into a dashboard and flags the biggest ranking losses. The same model generates 40 ad variants from a brief, creates them through the Meta Marketing API and pulls the results. The Shopify-to-HubSpot integration a web agency used to bill two weeks for gets built in an afternoon.
The APIs were always public. What collapsed is the cost of using them yourself. And the second layer – execution – gets compressed in the same move, because nobody can bill 20 hours for what a model finishes in minutes.
What's left is the third layer. We'll come back to it.
The Hour Is Collapsing
Let's run the numbers soberly. An agency delivers something that used to take 20 hours – a reporting set-up, a campaign structure, an integration sprint. With AI assistance the team needs five. On time and materials, revenue for the same output drops by 75%.
The agency has three options. Keep billing 20 hours and hope the client doesn't notice. Bill five hours and lose three quarters of the revenue. Or change the pricing model.
Option one no longer works, for a simple reason: clients know AI is involved. They use it themselves. According to industry surveys, around a third of agencies have already received explicit requests for an “AI discount” – clients looking at the invoice and asking why copy a model drafted is still being charged at the old rate.
The big networks have understood this, and they now say so in public. Cindy Rose, CEO of WPP – the holding company behind Ogilvy, VML and GroupM – has said the time-and-materials model is “probably not sustainable in the long term because AI ultimately will enable us to do our work faster with fewer people”. She wants to serve clients “with a hybrid workforce of humans and agents”. WPP is in exclusive talks with Jaguar Land Rover on a contract where compensation is linked to sales and brand outcomes rather than hours. In August 2026 Rose conceded the move to outcome-based pay “will take a few years”.
The numbers behind it are less diplomatic. WPP cut headcount by about 8% in a year – from roughly 105,900 to 97,400 people – and is targeting £500 million in cost cuts over three years. Internally, WPP reports roughly a tripling of creative output with close to 50% efficiency gains from AI. You don't have to take those figures at face value to see the direction: three times the output with fewer people is the exact opposite of what an hourly model needs.
And it isn't only advertising. McKinsey disclosed in late 2025 that roughly a quarter of its global fees are tied to measurable client outcomes rather than time. When the most expensive consultancy in the world gives up the hour, it's no longer an experiment.
Why the Middle Gets Hit Hardest
The agency industry isn't being hit evenly. It's being squeezed from both ends.
The large networks have what you can't produce with a prompt: media-buying volume that negotiates rates, proprietary datasets across hundreds of clients, enterprise relationships that reach the boardroom. They can cut headcount, deploy agents and still gain share. That's exactly what's happening: Deutsche Telekom moved its agency set-up to a bespoke unit with Omnicom. BSH, the home appliances group, handed core marketing tasks to Accenture Song. Large clients are consolidating onto a few large partners that deliver data, scale and accountability from a single source.
The freelancers and two-person studios at the other end have almost no overhead. One experienced senior with good agents delivers today what needed a team of five three years ago, and can offer it at a price no agency with an office, account management and project leads can undercut.
In between sits the middle: agencies of 20 to 200 people whose revenue consists mostly of execution hours sold. Too small for proprietary data and buying power, too big for a freelancer's cost structure. German trade press described digital agencies at a “historic turning point” at the end of 2025: insolvencies at a high level, declining headcount, revenue that fell back again in November after a spring recovery – while network agencies gain share and mid-sized providers get squeezed. There's little reason to think the UK market, with its heavy exposure to retainer-based performance and digital work, is on a different trajectory.
And then there's the movement that hurts the industry most: insourcing. In “The Two-Person Team That Replaces a Department” we described how small teams with AI tooling deliver output that used to need whole departments. Clients are now running exactly that calculation against their agencies. A company paying £350,000 a year to four agencies hires two experienced people, gives them access to its own platforms and a budget for models and tooling – and takes back direct access to its own data. Not everything, but a large part of what used to be billed by the hour.
Where the Agency Still Earns Its Keep
This is where we have to be honest – against our own interest, if necessary. Because just as you have to separate pure wrappers from real platforms in SaaS, agencies split into middlemen and genuine value too.
Accountability and operation. Writing a script that pulls the campaign data is trivial today. Making sure it runs every day, that someone notices at 3 a.m. when the Meta API renames a field and the campaign keeps running on the wrong budget, that the report the board sees is correct – that's work nobody can prompt. This was already the core argument in the middlemen piece: building is not operating. Most clients don't want to run their own glue code, any more than they want to run their own mail server.
Judgement from many accounts. Someone who has looked after 30 client accounts for years spots patterns no single company can see in its own data. That a particular campaign structure has been underperforming across a sector for three weeks, that a Google update is hitting one category of pages, that a creative format has worn out – that's knowledge that comes from breadth. A model knows the training data. A good agency knows the last 90 days across thirty accounts.
Proprietary data and buying power. Media buying at scale, benchmarks across sectors, access to inventory and rates a single mid-sized company would never get. That's not a wrapper over an API – it's a product in its own right, exactly like an SEO suite's crawl index.
Creative work that isn't a derivative. Generating 40 variants of an ad is API work. Having the one idea that carries a brand for three years is not. The share of such work on a typical agency invoice is smaller than the industry likes to believe – but it isn't zero, and it's getting more expensive, not cheaper.
The wish not to do it yourself. A managing director with six marketing people doesn't want to orchestrate agents, manage credentials and maintain scripts. They want the numbers to be right on Monday and someone to be responsible. That wish isn't going away. It only shifts what gets paid for.
The rule of thumb is the same as in SaaS: the more an agency brings proprietary data, carries accountability or delivers judgement born of breadth, the less of a middleman it is. The more it sells hours in which someone operates a public platform, the more exposed it is.
What the New Agency Model Looks Like
We don't claim to have the answer. But we've placed a bet, and we're laying it out.
Price outcomes, not hours. The hourly model rewards slowness and punishes every efficiency AI brings. If you price outputs or outcomes – the ongoing operation of a reporting pipeline, a campaign with a defined target corridor, a finished tool – you keep the efficiency gain instead of handing it over as a discount. WPP needs a few years for this because its contracts, its clients' procurement departments and its internal systems are all built on hours. A small agency needs a quarter.
Build and operate the client's own tooling. This is our bet. Instead of operating the client's SaaS stack – five subscriptions whose interfaces we've learnt – we build the client small, AI-assisted software that solves exactly their problem and belongs to them: the reporting that pulls Search Console, the ad accounts and the shop into one dashboard; the integration that connects HubSpot and Stripe; the workflow that batch-processes product images. And then we run it. The client owns the code and the data. We're paid to make sure it works. That's the operational accountability from the middlemen piece – as a business model.
Be the team that uses the API on the client's behalf – and is transparent about it. The alternative to the disguised hourly rate is to say openly: yes, the model wrote the first draft, which is why this costs a fifth. What you're paying for is that we know which parts of it are wrong, that we build it into your systems, and that we pick up the phone when it breaks. Clients reward that honesty, because they already know what's going on.
Small senior teams with agents. Not 30 juniors producing hours, but five experienced people who can exercise judgement and leave execution to agents. That's the “hybrid workforce” Rose describes – except a small agency doesn't have to push it through against 97,000 existing jobs.
The uncomfortable question every agency should ask itself is this: if your client could prompt it, why are they paying you? If the answer is “because they don't know they could”, the business model is a deadline running out. If the answer is “because they don't want to operate it themselves, because we see patterns they don't, and because someone has to be accountable”, there's a future. It just looks different from today's invoice.
Conclusion
For decades the agency was the middleman between companies and the platforms their customers live on. It sold access, execution and judgement at the same hourly rate because the three couldn't be separated. AI is separating them. Access has collapsed, execution is being compressed, and the market is reacting exactly as you'd expect: AI discounts, consolidation onto big networks, insourcing into small teams, insolvencies in the middle.
What remains is the third layer – accountability, operation, judgement from breadth, proprietary data, genuine creativity. That's less than the industry bills for today. But it's what clients actually wanted to buy when they hired four agencies.
We think the agency of the next few years is small, senior, works with agents, prices outcomes instead of hours, and builds its clients their own tools instead of operating someone else's. That's our bet. It might be wrong. But the alternative – billing 20 hours for five and hoping nobody asks – certainly is.