How to Structure an Agency Retainer That Doesn't Erode
Short answer: There are three common retainer models — hours-based, deliverable-based, and access-based — and each fails in a predictable way. Hours-based turns into clients auditing your timesheet. Deliverable-based turns into arguments about what counts as a deliverable. Access-based turns into unlimited demand. You're not choosing the model with no downside; you're choosing which conversation you'd rather have, and then writing the clauses that keep it manageable.
The three models
Hours-based
A pool of hours per month at an implied or explicit rate.
Works when: the work genuinely varies month to month and the client values flexibility.
Fails as: the client starts managing your hours instead of your outcomes. Every task becomes a negotiation about how long it should take, and efficiency gains get taken from you rather than shared.
Deliverable-based
A defined set of outputs per month — four articles, two campaigns, one report.
Works when: output is genuinely standardisable and the client wants predictability.
Fails as: arguments about what counts. Is a rewrite a new article? Is a variant a new campaign? Every ambiguous case becomes a conversation, and you'll lose most of them to keep the peace.
Access-based
A monthly fee for ongoing access to your team — a defined scope of responsibility rather than a defined output.
Works when: the relationship is genuinely ongoing and trust is high.
Fails as: unlimited demand. Without a boundary, "access" expands to fill whatever the client needs, and the price was set against a much smaller assumption.
Every model leaks somewhere. Pick the leak you can manage and write the clause that caps it.
The clauses that stop erosion
A named scope of responsibility, not just a list of outputs
"We are responsible for X" is more durable than "we will produce four of Y," because it survives the work changing shape — which it always does.
A defined revision allowance
Two rounds included, further rounds billable. Most retainer margin loss is revisions, not scope. Without a number, there's no natural stopping point and asking for one mid-project feels like a complaint.
Explicit exclusions
What the retainer does not cover, named. This is more useful than an inclusion list because it's the ambiguous cases that erode you, and inclusion lists can't anticipate them.
A rollover rule
Decide whether unused capacity carries forward, and cap it if so. Uncapped rollover produces a client who "has 40 hours banked" and expects them in one month.
A review point
Every six or twelve months, a scheduled conversation about whether the scope still matches the price. Building this in makes repricing routine rather than confrontational.
What actually keeps a retainer healthy
The structure matters less than whether both sides can see what's being consumed.
Retainers erode gradually — a little extra each month, none of it worth a conversation on its own. By the time it's obviously wrong, twelve months of precedent say it's normal. A client who can see consumption raises the question themselves at month three, when it's a scoping conversation rather than a renegotiation.
Frequently asked questions
What's the best retainer model for an agency?
None is best in general. Hours-based invites clients to manage your time; deliverable-based invites arguments about what counts; access-based invites unlimited demand. Choose based on which failure you can manage, then write the clause that caps it.
How do I stop a retainer from eroding?
A named scope of responsibility, a defined revision allowance, explicit exclusions, a capped rollover rule, and a scheduled review point. Revisions are the largest single source of erosion in most retainers.
Should unused retainer hours roll over?
Only with a cap. Uncapped rollover creates a client with a large banked balance who wants to spend it in one month, which breaks your capacity planning.
When should I reprice a retainer?
At a scheduled review point, with consumption data behind you. Repricing mid-term without evidence is a difficult conversation; repricing at a built-in review with numbers is a routine one.
The short version
Pick the failure mode you can live with. Cap revisions, name exclusions, cap rollover, schedule a review.
Then make consumption visible to both sides — it's the difference between a client raising a scope question at month three and you raising a price question at month twelve. See what visible consumption looks like — 7 days for $1.