Stop selling assembly priced as skill. The services an agency should exit in the AI era are the ones whose price reflects the labour they used to take rather than the value they deliver now: manual reporting packages, volume production line items, trafficking and QA sold as a standalone, formatted decks that a model can draft, and generic content at scale. The decision is usually obvious. The hard part is exiting without losing the client.
The agency-design framework sorts every service into keep, change, or stop selling. This piece is about the "stop" bucket specifically — which services land there, how to tell, and how to walk away from them cleanly rather than carrying them until they leak margin.
The test for a stop-sell service
Ask one question: if a competitor used AI, could they deliver this at a fraction of the price and still be credible? If yes, the service is priced against labour that no longer exists, and you are one comparison away from losing it anyway. The price is a liability, not an asset.
The second test is what the client is actually paying for. If they are paying for the output and nothing behind it — the report, the assets, the formatted plan — then the value is assembly, and assembly is what AI made cheap. If they are paying for the judgement, the relationship, or the accountability, the service is defensible even if it involves production. Stop-sell services fail both tests.
The usual stop list
The services that consistently land in "stop" share a shape. Manual reporting packages are the clearest: the work is pulling, joining, and formatting, and it was priced as if the analysis were the product when the assembly was most of the hours. Volume production line items are next — a bundle of assets priced per unit, where the model has collapsed the marginal cost. Routine trafficking and QA sold as a standalone service is a third.
Generic content at scale is the fourth, and the one agencies resist most, because it feels like the core business. But content whose only differentiator is volume was never the differentiator; the judgement and the brand fit were. And a fifth: planning documents sold by the hour, where the client is effectively buying the planner's production time rather than the plan's quality.
Why it is a client conversation
The most common mistake is treating "stop selling" as an internal decision. It is not. The client bought the service, it is on the scope, and quietly withdrawing it reads as a cutback. The exit has to be framed as an upgrade: we are no longer going to bill you for assembly, because we can now do that faster and cheaper and put the money into the judgement that actually moves your results.
That reframe is not spin; it is the truth about the new economics. The agency that leads with it looks ahead. The agency that keeps selling the old line item looks like it is either unaware of the cost shift or hoping the client does not notice. The second position is the one that loses the relationship.
How to exit cleanly
Give the client a bridge, not a cliff. Move the service from a priced line item to an included capability, so the outcome improves while the invoice is restructured rather than cut. Renegotiate the scope around the judgement that replaces the assembly. And be explicit about what changes: what the client gets more of, what they get less of, and why the trade is better for them.
Sequence the exits. Do not stop five services in one review; the client will hear "the agency is shrinking." Pick the one with the worst price-to-value ratio, exit it well, show the improvement, and use that as the pattern for the next. Each clean exit builds the case for the next, and none of them looks like a retreat.
The reporting package, in detail
The reporting package deserves its own note because it is the service most agencies defend and least defensible. The old package bundled data pulls, chart-building, formatting, and a few observations into one monthly line item, and most of the hours were assembly. AI now does the pulls and the formatting, and often a competent first pass at the observations. What is left worth paying for is the interpretation: what the numbers mean, what to do next, and who is accountable for the call.
So the exit is to split the service. Stop selling the assembly and start selling the intelligence. The client still gets the report — in fact they get it faster and more often — but they are paying for the read, not the build. That is the same reframe I made in reporting as intelligence, applied to the commercial side. An agency that keeps selling the assembly is charging for the part the client can now do themselves.
Sequence the exits over a year
The exit plan should be a rolling one, not an annual purge. In the first quarter, take the single service with the clearest assembly-and-price mismatch and restructure it. Show the client the improved outcome. In the second quarter, use that proof to move the next one. By the end of the year the menu has changed without any single conversation feeling like a cutback.
This pacing matters because trust is the currency. A client who sees four services dropped at once reads it as an agency in trouble, whatever the logic. A client who sees one service upgraded, then another, reads it as an agency getting ahead of the market. The economics are identical; the story is not, and in agency work the story is half the relationship.
The services that only look safe
A few services sit just outside the obvious stop list and are worth re-examining, because they look like judgement but are mostly assembly wearing a senior label. "Strategic" audits that follow a fixed template. "Bespoke" research that is a search and a summary. "Senior oversight" that is really a status update. These pass the eye test and fail the value test.
The tell is whether the service changes shape with the client. Real judgement produces a different answer for a different context. Template work produces the same answer with different numbers. If you can hand the service to a junior and get the same output, or now to a model and get the same output, it is assembly — and it belongs in the stop bucket even if the label says otherwise.
What to sell instead
The replacements are the keep and change services: senior judgement, measurement design, brand stewardship, the operating model itself. These are the things a model cannot do alone and a client cannot easily shop around on price, because their value is specific to the client's context and the agency's accountability.
The move is not "sell AI." It is "sell the outcome the old service was supposed to produce, without the assembly that used to sit in front of it." An agency that frames its menu this way turns the cost shift into a stronger pitch, because it is selling what the client actually wanted all along.
FAQ
Will stopping services shrink our revenue?
In the short term, sometimes. In the medium term, no — you are swapping revenue for a fee base that AI cannot undercut, and a client who trusts the new pitch. The agencies that lose are the ones that protect the old line items until a competitor makes the comparison for them.
How do we tell the client without looking weak?
Frame it as an upgrade, because it is one. You are removing a charge for work that is now cheap and reinvesting in judgement. Lead with what the client gets, not what you are dropping, and be specific about the improved output.
Should we ever keep a stop-sell service?
Only as a loss leader or a bridge for a strategic client, and only with an expiry date. Left open-ended, it becomes a permanent margin leak. If you keep it, price it at cost and say so, so the client understands it is a transition, not the new normal.
What if the client only wants the assembly?
Then you have a client who is buying a commodity, and a competitor with AI will serve them cheaper. Either move them to the judgement service or accept that this is not the relationship you want to build the agency around.
The short version
Stop selling assembly priced as skill — reporting packages, volume production, standalone QA, generic content, and hours-based plan documents. The test is whether a competitor with AI could deliver it far cheaper and stay credible. Exit cleanly, one service at a time, framed as an upgrade, and replace it with the judgement the client actually wanted. The framework post covers the full keep/change/stop sort, and the for-agencies track works through the transition.
If you have exited a service, I would like to hear how you framed it — that conversation is where the strategy becomes real.
Cheers, Chandler