The first thing AI did for agencies was make the work faster. The second thing it did was make the old pricing logic harder to defend.
That tension is now moving from internal team meetings into client negotiations. WPP CEO Cindy Rose recently said AI-driven efficiency will put pressure on agency pricing because clients will expect to share in the savings. She also pointed to a harder question hiding underneath: if production takes less time, what exactly is the agency still charging for?
The answer can't be “hours saved.” That math is heading for the door.
The price of faster work
For decades, agencies have sold a bundle of labor, expertise, access and creative judgment. The labor was the easiest part to count, so it became the part clients saw most clearly. Retainers, billable hours, headcount and scopes gave both sides a shared language, even when the real value came from something less measurable.
Generative AI breaks that language. A strategist can now produce ten starting points where she used to produce three. A media team can test more variations before lunch. A small agency can take on work that once required a much larger department.
That is good news for delivery. It is awkward news for pricing.
Rose's comments, reported by Storyboard18, are useful because they cut through the sales language. AI doesn't automatically create a new premium. In many client conversations, it creates a new demand for a discount.
The old defense was that expert work takes time. The new client response is obvious: if the machine takes care of the first 70 percent, why am I paying for the old staffing model?
Production is becoming a commodity
The uncomfortable part is that much of agency output was already drifting toward commodity territory. Weekly social calendars, resized creative, basic landing-page variants, standard reports and first-pass research were never the parts clients wanted to protect. They were simply the parts that were easiest to include in a retainer.
AI makes the distinction visible. It separates what can be generated from what has to be decided.
That distinction matters more than whether a team uses Claude, Gemini, ChatGPT or an internal model. The tool choice is a delivery detail. The commercial question is whether the agency can make better decisions with the same budget, not whether it can create the same deliverable with fewer keystrokes.
The agencies at risk are the ones that describe their value in production nouns: posts, concepts, edits, campaigns and reports. Those nouns will get cheaper. Some will get close to free.
The safer offer is built around decisions. Which audience deserves another dollar? Which creative signal is real and which is noise? Which brand compromise will damage memory six months from now? Those questions aren't made less valuable because an AI system can draft a response.
I've written before about the zero-cost AI marketing threat. The point wasn't that every marketing service becomes worthless. It was that the cost floor falls faster than most firms are changing their offer. That gap is where margin disappears.
Value is not the same as output
A client doesn't really want 40 ad variations. They want to know which two deserve money behind them.
They don't want an impressive dashboard. They want fewer arguments about whether the campaign did anything.
They don't want a faster content calendar. They want a brand that still sounds like itself when every competitor can produce a similar volume of content.
That is the commercial reset. Agencies have to move up the chain from output to interpretation, then from interpretation to accountable choices.
Google's own guidance on new ads for the AI era of Search points in the same direction. The platform is absorbing more of the execution, from creative generation to how commercial prompts are matched with offers. Marketers still need to decide what the brand should say, where it should show up and what a useful outcome looks like.
That makes strategy sound important, which is where many agency decks go to die. “Strategy” is not a premium simply because it appears in a proposal. It becomes valuable when it changes a decision and the agency can show the chain from that decision to the result.
A sharper offer might include:
- A fixed fee for a decision system, not a pile of deliverables
- A clear experimentation budget with agreed rules for killing weak ideas
- A measurement layer that separates platform activity from business movement
- A senior review cadence where the agency takes a position instead of presenting options forever
None of that means agencies should pretend AI has no effect on their costs. Clients are right to ask where the efficiency goes. The agency's job is to show that the saved time is being reinvested in better judgment, not quietly converted into more margin.
The retainer is under examination
The monthly retainer won't vanish overnight. It is still useful for access, continuity and planning. But the vague retainer is getting harder to defend because AI makes vague capacity look like unused capacity.
A client will tolerate paying for availability when the team is scarce and the work is visibly manual. Once a lean team can produce more with less, the client starts asking whether the retainer is buying access or merely subsidizing an operating model they no longer understand.
That is why outcome-based pricing keeps appearing in agency conversations, even though it remains difficult to implement. Outcomes are noisy. Media platforms change. Sales teams miss targets. Brand effects take time. A pure percentage-of-revenue model can punish an agency for decisions it doesn't control.
The practical answer is probably not a heroic leap into full performance pricing. It is a hybrid model with fewer hidden assumptions. A base fee pays for senior access, planning and infrastructure. A variable component rewards a small number of outcomes both sides can actually influence. The agreement says what happens when the model is wrong.
That last sentence matters. AI increases the number of things teams can try. It does not guarantee that the underlying signal gets cleaner. As I argued in the search visibility measurement problem, more machine-generated activity can create less confidence if nobody agrees on what counts as evidence.
What agencies should sell next
The next agency advantage won't be access to a model. Clients can buy that access themselves, often for less than the cost of a lunch.
The advantage will be a point of view that survives contact with the data. That sounds less scalable than generating content, but it is the part of the work clients can't easily automate without accepting the risk of being wrong in public. That is why the adoption gap inside companies matters commercially, not just operationally.
Three changes follow.
First, agencies need to expose the production layer. If AI reduces the cost of making a deliverable, say so. Hiding the tool does not preserve value. It just makes the client suspicious when the output starts looking interchangeable.
Second, agencies need to make senior judgment visible. Not through more thought-leadership posts or another strategy framework. Put the actual call in the work: spend here, stop that test, reject this audience, protect that brand asset. A recommendation that never forces a tradeoff is just decoration.
Third, agencies need to price the feedback loop. The valuable system isn't one that creates a perfect campaign on the first attempt. It is one that learns quickly, records what changed and prevents the team from paying for the same mistake twice.
That is also where internal marketing teams have an opening. If agencies don't define the value of judgment, procurement will define the value of everything else. And procurement is very good at counting hours.
The part nobody can automate
AI will make agency work cheaper in some places. Clients should receive part of that benefit. The mistake is assuming the entire benefit belongs in the invoice.
The real prize is not a smaller production bill. It's a better decision made before the money is spent, repeated often enough to matter and explained clearly enough that the organization can learn from it.
Agencies that sell the old labor bundle will feel constant price pressure. Agencies that sell accountable judgment will still have to prove their value, but at least they'll be defending something scarce.
The machine can make the work faster. It can't decide what the work is for.