The agency-design decision framework sorts every service you sell into one of three buckets — keep, change, or stop selling — using two questions: where the value sits (assembly or judgement) and what AI does to the delivery economics. Services that are mostly judgement and relationship you keep. Services that are assembly with a thin layer of judgement you change. Services that exist mainly to bill for assembly you stop selling.
I have run agencies and client teams, and I have watched the last two years turn "what should an agency sell in the AI era?" from a conference question into a margin question. Most agency leaders I talk to know something has to change in their product mix. Fewer have a way to decide what, service by service, without gut feel. This is the framework I would use, and the reason it is deliberately blunt.
Why the product mix is now the strategy
For years, an agency's mix was stable. You sold strategy, planning, creative, media, and reporting, and the economics were set by how many people each service needed. The work was bundled, the hours were billable, and the retainer was really a headcount plan with a margin on top.
AI breaks the link between a service and the hours it takes. A reporting line that needed two analysts can run on a workflow and a reviewer. A creative service that needed a studio can produce variants from a brief. The service still has a name and a price, but the cost to deliver it has changed underneath, and the old pricing has not noticed. That gap — between what a service costs to deliver and what you charge for it — is where agencies will win or lose the next few years. The mix is the strategy because the mix is where that gap lives.
The three buckets
Keep. Services whose value is judgement, relationship, and accountability, where AI assists the delivery but does not change who is essential. Strategic counsel, senior media judgement, client leadership, the final creative call. These are the services you protect and invest in.
Change. Services whose value is real but whose delivery economics AI has rewritten. Planning, creative production, reporting, research, localization. You keep selling them, but you redesign how they are delivered and reprice them for the new cost base, or you watch a competitor do it and undercut you.
Stop selling. Services that existed mainly to bill for assembly — the parts of the bundle that were always mechanical and were priced as if they were skilled. Volume-based production line items, manual reporting packages, routine trafficking and QA as a standalone service. These are the ones AI makes unprofitable, and continuing to sell them under the old model is a slow leak.
The framework at a glance
| Bucket | Value sits in | What AI does | What to do |
|---|---|---|---|
| Keep | Judgement, relationships, accountability | Assists delivery, does not replace the essential human | Invest, protect, put your best people here |
| Change | Real value, rewritten economics | Removes most of the labour cost | Redesign delivery, reprice, retrain the team |
| Stop selling | Assembly priced as skill | Makes the price indefensible | Exit deliberately, before a competitor does it for you |
The mistake most agencies make is treating this as one decision — "we need an AI strategy" — when it is really a portfolio decision made service by service. Some things you keep, some you rebuild, and some you walk away from. A single answer for the whole agency is how you end up protecting a service you should have exited and cutting one you should have invested in.
How to classify a service
Take each service and ask four questions. The answers put it in a bucket without much debate.
- What is the client actually paying for? If it is the output — the report, the assets, the plan document — the value is assembly. If it is the judgement, the relationship, or the accountability behind the output, the value is human.
- What share of the delivery is mechanical? Pulling, joining, formatting, generating, adapting. The higher the share, the more AI changes the economics.
- Would the client notice if a machine did the first pass? If not, the service is a "change" candidate at least. If they would notice because the value is the judgement, it is a "keep".
- Could a competitor now deliver this at a fraction of the price? If yes, you are in "change" or "stop" territory, and the only question is whether you move first.
Ask these honestly and most agencies find the same shape: a handful of "keep" services where their people are genuinely scarce, a large "change" middle where the rework is real, and a few "stop selling" lines they have kept out of habit or fear of an awkward client conversation.
The people implication
The framework is not only about services. It is about which people you need, and this is the part agencies find hardest to act on.
"Keep" services need senior judgement and relationships, so the roles there look like the ones I described in what roles an AI-native marketing team actually needs — fewer, more senior, more accountable. "Change" services need people who can design and run the new workflow, which is a different skill from doing the old assembly, and most teams do not have enough of them yet. "Stop selling" services need a plan for the people whose work is disappearing, and pretending otherwise is how an agency loses its best juniors to a competitor with a clearer story.
The talent answer and the product answer are the same decision seen from two sides. If you reprice a service but keep the old staffing, the margin you thought you were capturing goes back out as idle hours. I made the case in AI raises the floor, depth is how you win that the differentiator moves to depth, and the framework is how that belief shows up in the P&L: you invest in the services where depth is the product and exit the ones where assembly was.
What to do with the "change" bucket
"Change" is where most agencies live and where most of the margin opportunity sits. The moves are specific.
- Redesign the delivery first. Rebuild the workflow around AI doing the first pass and a senior person reviewing, before you touch the price. Pricing a service you have not redesigned just destroys the margin you were trying to protect.
- Reprice to the new cost base. Once the delivery is genuinely cheaper, share some of that with the client and keep some. Agencies that reprice without redesigning lose both the margin and the client.
- Productise rather than customise. A service delivered as a repeatable workflow with a clear scope is easier to price, easier to staff, and harder for a client to shop around on hourly rates.
- Retrain, do not just redeploy. The person who did the assembly is not automatically the person who reviews the output. Give them the system-design and judgement skills the new workflow needs, or be honest that the role is changing.
Do this service by service and the mix shifts from a bundle of hours to a portfolio of outcomes, which is the only version that survives a client who now has access to the same tools.
FAQ
Should an agency stop selling creative and planning because AI can help?
No. Those services have real value; the question is how they are delivered and priced, not whether they exist. They are "change" services — keep selling them, redesign the delivery, reprice for the new cost base, and protect the senior judgement that clients actually buy.
How do we know what to stop selling?
Look for services priced as if the labour were skilled when most of it is mechanical, and services a competitor could deliver at a fraction of the price with AI. If you cannot name what the client is paying for beyond the output, you are in "stop" territory — and moving first beats being undercut.
Does the framework mean fewer people?
It means different people, concentrated toward judgement and system design. The assembly roles shrink; the senior and the builder roles grow. Pretending the mix does not change staffing is how an agency loses its margin and its best people at the same time.
Where should an agency start?
With an honest inventory of what you sell and what each service is really made of. Classify every service into the three buckets before you buy anything or reprice anything. The classification is uncomfortable because it forces decisions you have been deferring, which is exactly why it is useful.
The short version
The agency-design decision framework sorts every service into keep, change, or stop selling, based on where the value sits and what AI does to the delivery cost. Keep the judgement-and-relationship services, redesign and reprice the ones with rewritten economics, and exit the ones that were assembly priced as skill. The people decision and the pricing decision follow the classification. The Agency-model guide covers the rest of that lane, and the for-agencies track works the framework through pricing, headcount, and new business.
I have been on both sides of this — buying agency services and selling them — and the agencies that adapt are never the ones with the best tools. They are the ones willing to stop selling something.
If you have run this exercise on your own mix, I would like to hear which service was hardest to put in the "stop" bucket. That is usually where the strategy is hiding.
Cheers, Chandler