Agencies Are Making the Same Mistake With AI They Made With Digital
In the early days of digital advertising, traditional agencies had a talent problem. Or at least that was how many of them described it.
They needed digital people. So they hired specialists, acquired digital shops or built new departments. Sometimes that worked. Sometimes the new people arrived inside an organisation that still thought, sold and managed work exactly as before.
The capability changed. The agency didn't.
AI is creating a familiar temptation. Hire an AI lead. Run workshops. Give everyone access to the latest models. Build a few agents. Announce an efficiency target.
Those things may be useful. They don't answer the harder question: what should an agency actually be paid for when clients can do more of the work themselves?
Clients are bringing more work inside
In the GWA KI-Whitepaper 2026 (page 30), 80% of the 170 surveyed company decision-makers said they intended to bring more marketing services in-house using AI. The study is not representative of all clients, and intention is not the same as an executed budget decision. Still, the direction deserves attention.
The first candidates will differ by company, industry and stage of development. A mature consumer brand, a regional utility and a growing B2B business do not have the same capabilities or reasons to outsource.
There is nevertheless a plausible pattern. Performance activation, content production, SEO, website work and repeatable digital execution can often be connected closely to internal teams, customer data and ongoing optimisation. For some clients, doing more of that work themselves will become increasingly attractive.
The harder-to-replace external contribution may be the work around long-running campaign platforms, distinctive creative ideas, brand positioning and the choices that hold a brand together over time.
That is a prediction, not a law. Clients will insource some strategic work too. And agencies that only produce generic strategic-looking documents have little reason to assume their work is protected.
Fewer hours are not automatically better margins
An agency paid for hours may initially see its strategy margins improve when AI helps its people complete work faster. But if the client eventually knows the work takes less time, the same hourly model becomes a problem.
The agency can sell fewer hours. It can lower prices. It can try to hide the productivity gain. None of those is a particularly attractive long-term strategy.
Moving away from hours is easier to recommend than to execute. A fixed fee still needs a sensible scope. Value-based pricing requires a credible understanding of the value created, and not every project produces an outcome that can be priced neatly. Retainers can work, but only if the client understands what continuing access and judgement are worth.
The pricing model still has to fit the assignment. Pricing based mainly on the time needed to produce a deliverable becomes increasingly fragile when the time needed changes dramatically. That belongs in the economics of strategic work, alongside delivery cost and the effect on the rest of the agency.
The mistake is treating AI as a department
A new technology can require specialist expertise. It can also change the work of everybody else.
Alongside the shortage of technical skills, many agencies struggled to rethink how strategy, creative, production and account management should work together once digital became central to the business.
AI poses a similar organisational question. If research becomes faster, when should strategy happen? If a client can generate plausible options, what is the agency's job in challenging them? If the first draft is cheap, who decides what is worth developing? If creative production becomes more scalable, how does the agency avoid producing more variations of a weak idea?
An AI team cannot answer those questions on behalf of the whole agency.
The quality problem hiding behind efficiency
Agencies have good reasons to care about costs. They also risk optimising the part of the business clients can increasingly reproduce.
Faster content production is valuable until everyone can produce content quickly. Faster research helps until every presentation contains the same readily available findings. Efficiency may improve the economics of today's service while doing little to defend its value tomorrow.
The more interesting question is whether the agency becomes better at identifying the problem, making strategic choices and turning those choices into creative work that matters.
That requires investment in how the agency thinks and works. Buying technology covers only part of that job.
Smaller agencies, different economics
Agencies may become smaller as AI changes the work required. That doesn't mean they will automatically become more profitable, or that every discipline will shrink at the same rate.
An independent agency that uses new capacity to give more accounts proper strategic attention may become more valuable. One that simply removes hours from existing deliverables may find itself negotiating away the saving.
On page 45, the same GWA survey asked what agencies would contribute as AI automates production. Surveyed clients selected creative leadership, insights and strategic advice alongside technological capability and production. The contribution of strategic judgement needs to be visible in the work that follows. That is encouraging, but it is not a guarantee of demand for every agency claiming to offer those things.
The real test will be whether an agency can demonstrate a contribution its clients recognise and are willing to pay for.
Hiring AI specialists may be necessary. Buying better tools may be sensible. Neither is a substitute for deciding what kind of agency should exist on the other side of the change.
For agency leads assessing Mirel, the commercial question is what the agency will do with the capacity recovered and how clients will recognise its value.