Agency scopes for next year are being drafted right now. In most B2B organizations the sequence runs from late September through November: marketing submits its plan, procurement opens renewal conversations with incumbent partners, and somewhere in that process a statement of work gets attached that looks substantially like last year’s. Twelve blog posts a month. Four campaign concepts a quarter. A named account team at a blended rate, with a deliverable schedule in an appendix that nobody reads after signature.
This year there is a new question in the room, and it is being asked by finance rather than marketing. If AI has made content production dramatically cheaper, why is the retainer the same?
It is a fair question with a bad obvious answer. The obvious answer is to demand a fifteen or twenty percent efficiency discount against the same deliverable schedule, book the saving, and move on. A meaningful number of organizations will do exactly that in the next two months. They will get the discount, because the agency wants the renewal. They will also get a materially worse relationship, for reasons that are predictable if you understand what the retainer was actually buying.
What the Retainer Was Paying For
Strip a traditional agency retainer down and it was purchasing three distinct things, bundled because they arrived together in the form of people.
The first was production capacity: the hands to make the assets. Writers, designers, editors, trafficking, versioning, localization. This was the bulk of the hours and, in most scopes, the bulk of the cost.
The second was access to scarce craft: a strategist who had seen forty positioning problems, a copywriter with genuine range, an art director with taste. Hard to hire for a single company, economical to share across a portfolio of clients.
The third was throughput insurance: the ability to absorb a surge without your headcount plan absorbing it. Product launch moves up a quarter, the agency flexes, your org chart does not.
The first and third have collapsed in cost. Production capacity is no longer scarce in the way it was even two years ago, and surge absorption is much less dependent on having warm bodies available. The second has not collapsed at all. If anything it has appreciated, because the volume of work now competing for attention has gone up while the supply of people who can tell good work from plausible work has stayed flat.
So the honest read is that roughly two-thirds of the historical cost basis of an agency relationship has deflated, and the remaining third is worth more than it used to be. That is not an argument for paying the same amount for the same scope. It is an argument for buying something different.
Why the Efficiency Discount Backfires
Here is the mechanism that plays out when a client holds deliverable volume constant and cuts the fee.
The agency has to protect gross margin, which means reducing the cost of servicing the account. The cheapest hours to remove are not the production hours — those are already automated or offshored in most shops. The expensive hours are senior: the creative director’s review, the strategist’s time in your quarterly planning, the editor who kills a draft rather than polishing it. Those get thinned first, quietly, over a couple of months. Nobody sends a notice.
What you end up with is a scope that still delivers twelve posts a month, produced substantially by the same tooling you have access to in-house, with less senior scrutiny than before. You have paid an agency margin for production you could have run yourself, and you have defunded the only part of the arrangement you could not replicate.
There is a second-order problem underneath this. The traditional apprenticeship model in agencies ran on production work: juniors learned judgment by making things badly under supervision and being corrected. When that production work is automated away, the training ground goes with it, and the pipeline of people who will be able to exercise judgment in five years narrows. This is the same structural issue in-house teams are wrestling with in their own review cycles, and it is worse on the agency side because the economics of a shared services business punish carrying unproductive juniors. When you squeeze an agency’s senior time, you are also squeezing the mechanism that produces senior people. Both of you are consuming the same finite resource.
What Is Actually Scarce Now
If production is not the constraint, a scope built around production volume is measuring the wrong thing. Five capabilities are genuinely hard to source, and they are what a 2027 scope should be organized around.
Deciding what not to make. The binding constraint in most content operations is no longer throughput, it is editorial judgment under abundance. An agency earning its fee should be reducing your output while improving your results, and should be able to defend a specific deletion — this campaign, this asset, this channel — with a reason. If your partner’s recommendations only ever add, they are still selling production.
Original inputs. Everything a model generates is derivative of a corpus your competitors can also reach. The scarce inputs are primary: customer interviews, proprietary data cuts, field observation, a genuine point of view held by a named person who will defend it. Buying primary research and original thinking is a different line item than buying assets, and it should be priced and scheduled separately.
Accountability and risk transfer. An agency that puts its name on the work, carries insurance, indemnifies you against IP problems in what it delivers, and can be fired is absorbing risk your internal team cannot. This has real value and it is rising as generated assets introduce provenance questions that did not exist in a stock-photo world.
Cross-portfolio pattern recognition. A good agency sees twenty companies’ worth of channel behaviour. In a year when discovery paths, deliverability economics, and bidding mechanics have all shifted underneath everyone, a partner who can tell you what is working across their book is providing something you cannot buy from a research subscription. Ask for it explicitly and put it in the scope as a recurring obligation, not a nice-to-have in a QBR deck.
Systems literacy over asset literacy. The technically consequential work has moved to how content is structured for retrieval, what your crawler access policy permits, how conversion signal is defined and fed back to platforms, how a message survives being summarized by a machine before a buyer sees it. Plenty of agencies are still organized around campaigns and assets and have not rebuilt this muscle. Some have. The difference will not show up in the credentials deck.
The Contract Terms Nobody Updated
Most agency master services agreements in force today were written before generative tooling was central to how the work gets made. Six clauses are worth opening before you sign a 2027 scope.
AI usage and disclosure. Not a prohibition — a prohibition is unenforceable and you do not actually want it. What you want is a stated policy: where generative tools are used in the workflow, what human review applies before delivery, and what gets disclosed to you. If the answer is vague, that is your finding.
Ownership of derived assets. Prompt libraries, fine-tuned models, custom agents, and workflow automations built on your brand corpus, your customer data, and your performance history. Under most standard agreements the agency owns these as methodology. That is defensible for genuinely generic tooling and unacceptable for anything trained on your data, and it is the single most common gap in a current MSA. Decide deliberately rather than by default, because this is the switching cost that will determine whether you can ever leave.
Data rights and training. Whether your data, briefs, and results can be used to improve tooling that serves other clients, including competitors. Get the subprocessor list while you are at it — your security team will ask eventually, and it is cheaper to ask now.
Consumption cost pass-through. Agency tool bills are increasingly metered rather than seat-based, and that cost is going to reach you one way or another. Better to see it as a transparent pass-through with a cap than as opaque margin, and better to agree the mechanism before it becomes a mid-year renegotiation.
Regulatory flow-down. Transparency and disclosure obligations that attach to systems your marketing function operates do not stop attaching because a vendor built them. If an agency is running conversational agents or producing synthetic media on your behalf, the compliance obligations need to be allocated in writing.
Indemnity scope. Confirm that IP indemnification actually covers generated material. Several standard forms carve it out, which moves the risk to you at precisely the moment the risk became harder to assess.
How to Run the Conversation
A few practical moves for the next six weeks.
Separate thinking from making in the scope, and price them differently. One line for strategy, research, and senior judgment, with named people and a minimum time commitment. Another for production, priced at market rates that reflect what production now costs. This makes the efficiency conversation tractable: you can take the deflation where it genuinely exists without defunding the part you need.
Ask incumbents a direct question: what did you stop charging for this year? A partner who has repriced honestly will have an answer, and will usually have redeployed that capacity into something more valuable. A partner whose rate card is unchanged since 2024 is either not using the tooling or not passing any of it back, and both are worth knowing.
Replace the credentials review with a working session on a live problem. Ninety minutes on something real tells you more about judgment than any case-study deck, and it is now the only reliable differentiator — anyone can produce a polished deck.
Name the team and commit the tenure. When production is commodity, the specific humans are the product. A scope that does not name them is not a scope.
And resist the instinct to insource everything on the theory that AI makes agencies redundant. Bringing production in-house is often correct on cost. What comes with it is the review burden, the recruitment problem for senior judgment in a tight market, and the loss of an outside view at exactly the moment everyone’s internal content is converging toward the same generated median. Insourcing production while retaining a smaller, more senior, more expensive strategic relationship is usually the better trade, and it is a harder sell internally because it does not look like a saving.
The Honest Summary
The efficiency question finance is asking is legitimate, and the deflation is real. Marketing leaders who defend an unchanged retainer against an unchanged deliverable schedule will lose that argument, and should.
But the saving is not a discount to be extracted, it is a reallocation to be designed. The same money buys either more production you no longer need or more judgment you cannot get anywhere else, and the scope you sign in the next six weeks determines which.
There is a pattern here that runs through everything the industry has repriced this year: martech contracts moving from seats to consumption, internal reviews moving from output to judgment, media buying converting into data operations. Effort is getting cheaper and discernment is getting more expensive. Agency relationships are simply the last major line item in the marketing budget to be marked to that market, and most of them are about to be marked wrong.