Agencies are entering a period of existential change. The disruption isn’t about whether creativity or talent still matter. They do. But as AI reshapes the speed and structure of delivery, many agencies are finding that their commercial models no longer match the work they’re producing. Let’s look at the dynamics at play.
Billable hours are becoming liability hours
For decades, the dominant model was simple: agencies sold time, layered margin over overhead, and scaled by headcount. This worked as long as clients accepted that labour was the essential input and that every hour had a defensible cost.
That assumption no longer holds when AI can produce in minutes what once took teams days. Research, design, scoping, copywriting, even strategic outputs, were once premium, protectable work. Now, many of them are at least partly automatable. The result is awkward: many agencies still charge by the hour when the underlying model no longer applies.
Those with large retained teams face an uncomfortable trade-off: protect utilisation or pass on savings. So far, most have opted for the former.
Value becomes the commodity
Clients are beginning to expect pricing that reflects outcomes, not inputs. A project that once cost $400,000 might now be deliverable for a quarter of that, with AI augmenting the end-to-end process. Agencies that charge by output or impact are better placed to survive this shift.
Value-based pricing, performance-linked retainers, and AI-informed deliverables (e.g. “$X per personalised experience generated”) are all emerging. But pricing is only part of the shift. So is cost structure.
Meanwhile, legacy offerings are looking increasingly out of sync. Slide-heavy strategy presentations, five-month CRM roadmap programmes, “design and build” website projects with 80-page Discovery decks — all still priced for a world where human labour is the bottleneck. Clients are noticing.
Charging for compute, not just people
If AI is a new kind of colleague, then it comes with its own line items — and not just for compute.
Yes, token costs matter. LLM inference, embedding models, and orchestration layers now form a new kind of infrastructure overhead — and must be priced in, just as agencies once did with studio hours or freelance day rates.
But increasingly, data is the bigger cost. Access to high-quality, structured, and up-to-date sources often exceeds token spend in real-world applications. Answers are only as good as the information behind them — and that information is rarely free.
Then comes the hidden cost of reliability. AI systems demand carefully managed test sets, token-intensive evaluation runs, and ongoing human-in-the-loop refinement. As models evolve and content shifts, workflows must be revalidated — consuming time, attention, and budget.
Clients may ask why they’re being charged when “the AI did the work.” Agencies must be ready with a clear answer — not just in people-hours, but in the total effort required to make an AI workflow accurate, compliant, and production-ready.
It changes the conversation. Agencies are no longer charging for labour alone. They’re charging for orchestration, governance, and results.
Fast, Good, Cheap? Pick all three
The old iron triangle of project management held that you could have two of three: fast, good, or cheap.
AI has started to break this model. A carefully semi-automated workflow can deliver all three. Quality still needs human oversight, of course, but the economics are different enough to matter.
The impact on legacy service lines is not subtle. Large-scale customer research programmes — once staffed by teams running fieldwork, transcriptions, slide packs — are being outpaced by AI platforms, which can conduct intent analysis, identify gaps, and synthesise at scale. Often in hours rather than weeks.
This creates uncomfortable optics for agencies still selling time-intensive work while quietly automating under the surface.
Independent agencies have an edge
Holding companies were designed to scale through headcount and margin stacking. Some are slow to adapt. Their legal, procurement, and compliance machinery is optimised for a different century.
By contrast, independent agencies are quietly building custom models, automating internal workflows, and experimenting with lean AI pipelines. Some are adopting hybrid models, part AI and part human, with traceable provenance and token-level cost attribution. Others are building their own internal copilots.
These firms can price more flexibly, pivot faster, and experiment without the quarterly pressure of public markets.
Billable, but not believable
Agencies face a reckoning, not with creativity, but with economics and credibility.
The old levers — scope, time, headcount — are giving way to new ones:
- Compute and orchestration
- Data access and reliability overhead
- AI-powered velocity
- Value of outcomes, not effort
- Transparent blend of human and machine input
To stay relevant, agencies must define new models, not just defend old ones.
Because once clients can see the difference, will they still pay for the past?
