Stop pricing by hour the moment production stops taking hours. If a deliverable that took two days now takes two hours and you bill by time, you have just cut your own rate by 90% for getting better at your job. The pricing models that survive are per-deliverable, per-outcome, and retainer, and the conversation that decides whether you keep your margin is the one where a client says "but you used AI, so it should be cheaper." This guide is about having that conversation well.
Why hourly collapses first
Hourly billing assumes time and value move together. Generation broke that assumption in one direction only: the production hours fell, the judgment did not. Deciding which of forty variants ships takes exactly as long as it always did, and it is the part the client cannot do.
Bill by hour in that world and you are paid least for the work you are best at. Worse, you are incentivised against efficiency, which clients eventually notice and resent.
What to price instead
Per deliverable. A price for the thing: a campaign, a set of fifteen tested variants, a brand board, a 30-second spot with cutdowns. The client buys an outcome and does not audit your method. This is the default replacement and it is what most agencies moved to.
Per outcome, where it can be measured. Pricing tied to a result the client cares about. Rare, hard to negotiate, and the only model where AI efficiency makes you more money rather than less.
Retainer for creative volume. Increasingly the natural fit: the client needs a steady supply of tested creative, and the supply is the product. Price the throughput, not the hours behind it.
Hourly only for genuinely open-ended work. Discovery, consulting, and anything where the scope cannot be written down. Say so explicitly, so hourly reads as a deliberate exception rather than your default.
The "you used AI, so it's cheaper" conversation
It is coming, so decide your answer in advance rather than in the meeting.
The honest position, and the one that holds: the client is paying for the decision, not the render. Anyone can generate forty variants. Knowing which one to ship, and why the other thirty-nine fail against this brief and this audience, is the service. The generation is a tool the way a camera was a tool, and nobody itemised the shutter.
Where you should concede: if a project genuinely got cheaper for you and the relationship is long-term, passing some of that back buys loyalty that outlasts the project. Concede deliberately and name it as a decision, not as a discount you were argued into.
Where you should not concede: the concept, the direction, and the judgment. If a client wants to pay only for output, they can generate output themselves, and the ones who try usually come back.
Building a per-deliverable price that holds
Work from three numbers, in this order.
1. Your cost floor. Time (yours and the team's), plus the actual production spend. That second number is now small and knowable. Using canon 8frame prices, a credit being $0.01 in a top-up pack: fifteen tested video variants drafted on Veo 3.1 Lite is about 810 credits, roughly $8; three client-facing style frames on Nano Banana Pro is 63 credits. A campaign's generation line is single or double-digit dollars, so it is no longer a meaningful part of the floor. The floor is time.
2. The alternative cost to the client. What producing this the traditional way would cost them: shoot day, crew, talent usage, edit. That number is the ceiling of what the deliverable is worth to them, and it did not fall just because your method changed.
3. The revision boundary. Two rounds included, further rounds priced. This single line protects the margin on more projects than the headline price does, because unbounded revision is where per-deliverable pricing dies.
Set the price between the floor and the alternative, closer to the alternative than instinct suggests.
The scope lines that protect the price
Rounds. Two included, priced after. See how to run a creative review meeting for making those two count.
Deliverables enumerated. "One video" is nine files once formats and cutdowns arrive. Write the deliverables list into the quote.
Usage. Where and how long the work runs, priced separately from making it. See what is a usage licence.
Disclosure. Whether AI generation is used, and whether the client needs it disclosed downstream. Better agreed at quote stage than discovered at delivery.
What to do with the efficiency
You now have a real choice, and it is a positioning decision rather than a pricing one. Take the margin, and be a smaller, more profitable operation. Or pass it through as more creative volume for the same fee, and be the supplier who ships fifteen tested variants where competitors ship three. Both work. Doing neither, and quietly billing the old number for the old volume, is the option that gets found out.
FAQ
Should I tell clients I use AI? Yes, at quote stage. Discovery later reads as concealment and costs you more than the conversation would have. Many clients now expect it, and some require disclosure for their own compliance.
How do I stop clients demanding AI discounts? Price per deliverable rather than per hour, so the method is not the unit being purchased. Then hold the line that the judgment is the service.
What if a competitor undercuts me with generated work? Some will, and some clients will take it. Those clients were buying output, and output is now cheap for everyone including them. Compete on the part that is not cheap.
Production got cheap and judgment did not. Price the judgment. The 8frame canvas is free and unlimited, and generation is paid from $19/month.