Is the Marketing Retainer Dying? How AI Changes Agency Pricing

The short answer: the retainer lives, the hours logic does not

The original title of this article declared the retainer dead. That overstated it. The IPA's agency remuneration best practice guide describes the retainer, ideally tied to an agreed scope of work, as the most common way agencies are paid, and ISBA's guidance for advertisers sets out alternatives such as payment by results. What AI is breaking is narrower and more useful to understand: the retainer that is really an hours contract, where the fee is justified by the number of people-hours the agency expects to spend.

When a first draft, a set of ad variations or a monthly report took a day, those hours were real. When the same drafts take minutes and the day becomes an hour of checking, a fee built on the old hours no longer describes the work. Something has to give: the price, the scope, or the honesty of the timesheet.

What a retainer actually buys

Strip the jargon away and a monthly retainer buys three things.

1. • Capacity. Somebody will do the work this month without a fresh negotiation.
2. • Continuity of context. The people (and increasingly the systems) remember your business, so you do not re-brief every time.
3. • Judgement. Someone decides what to do next and why.

AI changes the cost of the first item dramatically, strengthens the second if the agency stores context well, and barely touches the third. A sensible retainer in 2026 should be priced mostly on the second and third, not the first.

An illustrative retainer, before and after

Illustrative example. A retainer scoped at 40 hours a month.

• • 25 hours of production: drafting copy, resizing creative, building reports, researching keywords.
• • 15 hours of judgement: planning, analysing results, deciding what to test, client calls.

Suppose AI assistance cuts the production work to 10 hours, including the time to check it. The agency now delivers the same outputs in 25 hours. Three honest outcomes are possible:

• • More work for the same fee. The client gets extra tests, pages or campaigns each month.
• • Lower fee for the same work. The saving is shared.
• • More judgement for the same fee. The freed hours go into analysis and strategy, and the client can see it in the reporting.

The dishonest outcome is the fourth: the same outputs, the same fee, the same 40 hours on the timesheet. That is the version buyers should be alert to. The numbers here are round and hypothetical; the point is the shape, not the size of the saving, which varies by task. A Harvard Business School working paper (Dell'Acqua and colleagues), based on a field experiment with consultants, reported that AI helped markedly on some tasks and hurt accuracy on a task outside its capabilities, so the saving is uneven and has to be checked task by task.

Pricing models and how AI affects each

Agency pricing models under AI

Model	How it works	Effect of AI	Suits
Hours-based retainer	Fixed monthly fee for an estimated number of hours	Weakest: the hours estimate stops describing the work	Very few buyers now, unless hours are audited
Scope-based retainer	Fixed monthly fee for defined outputs, decisions and reporting	Holds up well: price follows what you receive	Most ongoing marketing relationships
Project fee	Fixed price for a defined deliverable	Holds up; AI may lower competitive prices over time	Launches, rebuilds, one-off audits
Performance-linked	Part of the fee depends on agreed results	Neutral; still needs agreed measurement and attribution	Businesses with clean tracking and patient timelines
Hybrid	Base scope fee plus a smaller performance element	Holds up; the base covers judgement, the bonus rewards outcomes	Mature relationships with trusted data

What to negotiate at your next renewal

• • Replace hours with a written scope. List the outputs, the recurring decisions (for example, a weekly review of paid campaigns) and the reports you will receive.
• • Ask where saved time goes. More tests, a lower fee or deeper analysis are all acceptable, but you should know which.
• • Ask who reviews AI output. A named person, and what they check.
• • Keep your assets. Ad accounts, analytics, content and data should stay in your name. GOV.UK guidance on copyright ownership says commissioned copyright stays with the creator unless it is assigned in writing, so check that the contract assigns it to you, and take legal advice if it is unclear.
• • Agree the measure before the work. If a performance element is on the table, agree the source of truth for the number first, or the bonus becomes an argument.
• • Read the notice terms. A scope-based retainer should be easy to review at set intervals.

Where this leaves agencies

Good agencies are not threatened by this. The ones that already sell judgement and context find AI makes them faster and more valuable per pound. The pressure lands on agencies whose margin came from production hours billed at senior rates. Buyers who ask the questions above will find out quickly which kind they are paying.

AfM runs as one managed service rather than a traditional agency: our AI operating system does much of the production and analysis, and the founder supplies the judgement and review. We do not publish prices here because scope varies; if you want to compare, ask for a written breakdown of your current marketing first. For a broader look at the choices, read agency vs AI marketing system vs DIY.

FAQ

Should I ask my agency to cut its fee because it uses AI?

Ask what has changed first. If the agency now delivers more tests, faster turnaround or deeper analysis for the same fee, that may be better value than a discount. If nothing you receive has changed, a fee review is a fair request.

Is payment by results the fairest model now?

It can be, but only with clean tracking and an agreed attribution method. Without those, both sides end up disputing what caused the result. Many buyers do better with a scope-based fee and a small performance element once measurement is trusted.

How do I check whether my agency's hours are real?

Ask for a breakdown of time by task for one recent month and compare it with the outputs you received. Large blocks of production time on work that AI tools now draft quickly are the place to start the conversation.