Price the output, not the hours. Hourly billing ties your revenue to production time, so every efficiency gain shrinks your own invoice. The four workable models are hourly, value-based, output-based, and a retainer with a volume commitment. The agencies holding margin moved to output-based or committed retainers, and they moved before the client asked why the bill got smaller.
A campaign set that took your team three weeks last year takes four days now. Same client, same scope, same quality bar, same approvals. Under an hourly contract you just invoiced a fraction of what you invoiced last year for identical work.
You got better and you got paid less. That is not a market correction. That is a billing model doing exactly what it was built to do.
The conversation is coming whether or not you have an answer ready. Around a third of agencies surveyed by Productive have already been asked for an AI discount, and nearly half expect to hear the question soon (productive.io, published November 2025, accessed July 2026).
This covers the four models, the transition mechanics, the margin math, and what to say when the client asks.
Key Takeaways
- Hourly billing is the only model that punishes you for getting faster. Every hour AI removes from production is an hour you no longer bill. The model converts your capability gain into a client discount automatically, with no negotiation required.
- Output-based pricing is the cleanest exit. Bill per asset, per variant, or per campaign. Production time becomes your problem and your margin, which is where it belongs.
- A retainer with a volume commitment is the strongest ongoing structure. The client buys a committed monthly volume at a fixed fee. Predictable revenue for you, predictable output for them, and speed accrues to you across the whole term.
- Move before the client asks. A repricing conversation you open is a strategy discussion. The same conversation the client opens is a discount negotiation.
- Reprice at a natural boundary, never mid-engagement. Renewal, a scope change, or a new brand are the moments where a new structure reads as normal rather than opportunistic.
- When the client finds out you use AI, do not deny it and do not apologise. Answer with what you carry: the tool cost, the art direction, the QA pass, the rights position, and the liability.
- Judgment does not compress. AI speeds up generation, not art direction, selection, or QA. Those become a bigger share of your cost base, so price for them.
| Model | How you bill | What happens when production speeds up | Best for |
|---|---|---|---|
| Hourly | Your rate times hours worked | Revenue falls automatically. Your gain becomes the client’s discount | Genuinely unscoped discovery, short exploratory engagements |
| Value-based | A share of the commercial outcome the work produces | Revenue holds, but only if the outcome is measurable and attributable to you | Work with a direct, traceable commercial result |
| Output-based | A fixed price per asset, variant, or campaign | Revenue holds and margin widens. The speed is yours to keep | Repeatable, well-specified production |
| Retainer plus volume commitment | Fixed monthly fee for a committed monthly volume | Revenue holds and stays predictable. Speed compounds across the term | Ongoing relationships with steady, forecastable demand |
Why does hourly billing break when production gets faster?
Because it sells time, and time is the thing that just got cheaper. An hourly contract makes your invoice a function of how long the work takes. Improve the process and the invoice shrinks on its own. No other model does this to you. You earn less for identical output, and the client never has to ask for the discount.
The trap is that most agency retainers are hourly contracts wearing a different name. If your retainer is a bank of hours, a monthly allocation of days, or a fee the client reconciles against a timesheet, it inherits every defect above. The unit is still time. Renaming it does nothing.
Check yours with one question: if your team halved its production time next month, would the client owe you the same amount? If the answer is no, you are billing hourly.
What are the four pricing models, and which one fits?
Four models are worth running. Hourly sells time. Value-based sells a share of a commercial result. Output-based sells a defined deliverable at a fixed price. A retainer with a volume commitment sells a guaranteed monthly volume for a fixed monthly fee. Most agencies need two of them, not one: something for unscoped work, and something for production.
Hourly still has a narrow home. Genuine discovery, exploratory work, and engagements where the scope truly cannot be defined in advance are all fair. Keep it there. Do not run repeatable production on it, because production is the part that is specifiable and the part that compresses fastest.
Value-based pricing is clean in theory and hard in practice. You need a result that is measurable, attributable to your work, and agreed in writing before you start. Creative production rarely clears all three. A campaign’s revenue depends on the media buy, the offer, the product, the pricing, and the season. Arguing that the creative caused the lift is a negotiation you will reopen every quarter. Use it where attribution is genuinely clean, which is rarer than the pricing literature suggests.
Output-based pricing is where most creative production belongs. A product shot is one price. A campaign set of twenty variants is another. The client knows what they are buying before they buy it, and every minute you save stays with you.
It demands one discipline: define the unit precisely. “One product shot” should mean one product, one background treatment, two delivered formats, and one revision round. Vague units get eaten by revision cycles, and the model fails for reasons that have nothing to do with the model. This is the same unit logic that governs what brands pay per image on a real photoshoot, and the discipline transfers directly.
A retainer with a volume commitment is the strongest structure for an ongoing client. The client commits to a monthly volume at a fixed monthly fee. Forty assets a month, one number. They get budget certainty, you get revenue certainty, and your efficiency gains compound across the full term instead of being handed back one project at a time.
How do you move an existing client off hourly without a fight?
You change the model at a boundary, not mid-stream. Renewal, a scope change, a new product line, or the annual planning cycle are the moments where a new structure reads as normal. Then you present it as something the client receives, which is price certainty and more output, rather than as a repricing. Never open with your own economics.
The mechanics, in order:
- Pick the boundary. Repricing mid-engagement invites the question “what changed?” and the honest answer is one you do not want to give first. Wait for the renewal. It is usually closer than it feels.
- Reprice the new work, not the old. Leave the existing engagement alone and put the next scope on the new model. The client loses nothing, and you get a live side-by-side comparison to point at.
- Lead with certainty. “You will know your creative cost before the quarter starts, and it does not move when a revision round does.” A fixed price transfers overrun risk from the client to you. That is a real benefit and it is worth real money.
- Price the first engagement off your own history. Pull the last four comparable projects. Set the unit price so the engagement lands near your historical revenue, not near your new cost. You are changing the model, not cutting the price.
- Expect the anchor question. “Last time this was forty hours at your rate. Why is it the same money?” Because they are no longer buying forty hours. They are buying a defined set of assets at a fixed price, delivered to a date, with the overrun risk on your side.
Some clients will only ever pay for time. As production compresses, those are clients you cannot keep profitably. Better to find out at renewal than in year three.
What does the margin math actually look like?
Under hourly, both sides of your margin scale with hours, so speed does nothing for you. Under output-based, the price is contracted and only your cost moves, so every improvement widens the gap. Same work, same client, opposite direction of travel. Write the two formulas down and the decision makes itself.
Hourly contribution per project:
(billed hours x your rate) minus (billed hours x your fully loaded cost per hour)
Both terms scale with hours. Cut the hours and both shrink together. Your margin percentage may hold perfectly well while your absolute contribution falls, which means you have to win more projects just to stand still. This is the number that gets missed, because the percentage on the dashboard looks fine.
Output-based contribution per project:
(unit price x units) minus (production cost + tool cost + review and QA time)
Only the right side moves. The gap is yours to keep.
Use your own figures, not a benchmark. Pull your last four comparable projects and work out three numbers: what you billed, what those hours actually cost you fully loaded, and what the same scope would cost to produce today. The gap between the second and third is the money your current model hands to clients for free every month.
One honest correction to the optimism. Generation compresses. Judgment does not. Art direction, model selection, culling, QA, and client review are still human hours, and they become a larger share of your cost base as the production half shrinks. Price the review, staff the review, and do not model a cost curve that assumes it disappears. Getting more creative output without more headcount is a real gain, and it is a smaller gain than a spreadsheet with no review line will tell you.
What do you do when the client discovers you use AI?
Answer plainly and do not apologise. You sell a finished, on-brand, rights-cleared asset. How it was produced is a method question, not a value question. The agencies that get hurt here are the ones who hid it and got caught, not the ones who put it in the contract. Disclose first and the discovery moment never happens.
- Put a method clause in the SOW. It costs nothing to write and a great deal to be caught without. It turns a potential breach-of-trust moment into a documented, agreed way of working.
- Answer the discount question with what you carry. The tool cost, the model selection, the art direction, the QA pass, the revision rounds, the rights position, and the liability if the asset is wrong. The client carries none of it. That is what the fee buys.
- Refuse the hourly-equivalence trap. “It only took you two hours” is a valid argument only if you sold hours. If you sold a defined asset at a fixed price, your production time is not the client’s business, in the same way a photographer’s shutter count is not.
- Have the rights answer ready. Where assets originate, what the licensing position is, and whether the client’s own product photography is the source. Vague answers here cost more than the discount would have.
The strongest version of this answer is a portfolio, not a paragraph. Assets built from the client’s actual product photo rather than from a text prompt do not come out looking generic AI, and that is a claim you settle by showing the work.
How do you structure a retainer with a volume commitment?
Four components: a defined asset unit, a committed monthly volume, a fixed monthly fee, and an overage rate for volume above the commitment. Add a use-it-or-lose-it clause so unused volume does not accumulate into a liability you owe forever. Get the unit definition right and the rest follows. Get it vague and revision rounds will eat the retainer.
- Define the unit. What counts as one asset, how many revision rounds are included, which formats ship, and what counts as a new asset rather than a revision. This one definition is where retainers succeed or fail.
- Set the volume from real demand. Use what the client actually consumed over the last two quarters, not what they aspire to.
- Price the fee against your historical revenue for that volume, then keep your cost improvements.
- Set an overage rate slightly below your standard unit price, so volume above the commitment is rewarded, and specify how it is invoiced.
- Use it or lose it. Allow one month of carry if you must, never a permanent bank. Banked volume is a liability that arrives all at once in Q4.
- Review quarterly. Demand moves, and an annual review is too slow to catch it.
Give video its own line. It is by far the most expensive operation on any AI creative stack, and it should never sit inside a general asset commitment priced like a still image.
What does the tooling cost against a retainer?
Less than most agencies assume. Tool cost belongs in your cost of goods next to production and review time. It is an input to your margin math, not a number the client should be pricing your work from. State it honestly in your own model and it strengthens the value argument rather than undermining it.
For a working number, DesignerBox prices by credits. Generating or editing an image is 5 credits. Ultra is $200 a month for 8,000 credits, which is 1,600 images at that rate, and it is the tier built for agency use: team collaboration, shared brand kits, white label, API access, a priority queue, an SLA, and a dedicated account manager. Five seats are included, extra seats are $19 each, and storage is unlimited.
State the gating honestly, because it matters here. Everything below Ultra is single seat. Free, Basic, Pro, and Premium are one user each at $0, $15, $35, and $75 a month, for 112, 500, 1,000, and 2,500 credits. An agency running several people across several clients is on Ultra, or it is not really running a team.
Credit packs top up a paid plan without a tier change: 100 credits for $5, 500 for $20, and 1,500 for $50. They are one-time, they need an active paid subscription, and they expire after 12 months. That is the mechanism for an unusual month, not for a permanent step up in volume.
Video breaks every calculation above. It is priced per second of output. A Veo 3 clip with audio at eight seconds costs 6,400 credits, which is more than Premium’s entire monthly allocation of 2,500. Price video as its own line in any retainer that includes it, and see the full model catalog before you commit to a per-clip number.
Two mechanics matter for multi-client work. Brand profiles keep each client’s look separate, which is the first thing that fails when one team runs six brands through one tool. Workflows save a campaign setup so the team reruns it for the next product, drop, or client, which is what turns a good one-off result into a repeatable unit you can actually put a price on.
The MCP server exposes 43 tools, so an agency can drive generation from Claude, ChatGPT, or Cursor inside its own pipeline instead of through the interface.
One bill across every model is also the point. The six-tool stack charges you six subscriptions, and the real cost is not the subscriptions, it is the seams between them.
For agency and studio use specifically, team pricing and shared credit pools are a conversation rather than a published number. Start at the pricing page.
FAQ
Should an agency ever bill hourly now?
Yes, narrowly. Hourly still fits genuine discovery, exploratory work, and engagements where scope cannot be defined in advance. It does not fit repeatable production, because production is the work that compresses fastest and the work that is easiest to specify. Most agencies should run two models: hourly for the unscoped edge, output-based or a committed retainer for everything else.
What is the difference between value-based and output-based pricing?
Value-based pricing charges a share of the commercial result your work produces, so it needs an outcome that is measurable, attributable to you, and agreed in advance. Output-based pricing charges a fixed price for a defined deliverable regardless of the result. Output-based is far easier to sell, defend, and administer for creative production, because the deliverable is specifiable and the outcome usually is not.
How do I tell a client I use AI?
Before they ask, and in writing. Put a method clause in the SOW describing how work is produced and what the rights position is. When the discount question follows, answer with what you carry rather than what you saved: tool cost, model selection, art direction, QA, revision rounds, and liability for the asset. Disclosing first turns a trust problem into an agreed way of working.
How do I stop revisions eating an output-based price?
Define the unit before you price it. Specify what counts as one asset, how many revision rounds are included, which formats ship, and what constitutes a new asset rather than a revision. Price additional rounds separately. Most output-based pricing fails on a vague unit definition rather than a wrong number, and the fix is contractual, not commercial.
Will clients ask for an AI discount?
Many already have. Around a third of agencies surveyed by Productive had already faced the request, and nearly half expected it soon (productive.io, published November 2025, accessed July 2026). The request is much easier to answer from an output-based contract, where production time was never what the client was buying, than from an hourly one, where it explicitly was.
Does AI actually reduce the headcount an agency needs?
It compresses generation, not judgment. Art direction, model selection, culling, QA, and client review stay human, and they grow as a share of your cost base once the production half shrinks. Plan for a shift in the mix of roles rather than a straight reduction, and keep the review line in your cost model.
What should go into a retainer with a volume commitment?
A defined asset unit, a committed monthly volume drawn from real consumption over the last two quarters, a fixed monthly fee, an overage rate slightly below your standard unit price, and a use-it-or-lose-it clause capped at one month of carry. Review the commitment quarterly. Price video separately, because it costs far more per unit than still imagery.
Agency AI-discount survey data verified from productive.io as of July 2026. DesignerBox credit costs, plan allocations, seat limits, and feature gating verified against live product configuration, July 2026. Individual results vary.
