AI Video Pricing: How to Put Generated Video on the Rate Card

The AI Video Pricing Gap: Generated Video Is Still a Cost Line

AI video pricing is the unsolved half of the generative production story. The craft questions have answers now. The commercial question mostly does not, and the research is unambiguous about it. Forrester's 2026 study with the 4As found that nearly nine out of ten marketing agencies now use generative or agentic AI as part of marketing creation and delivery, and that for the third consecutive year AI has remained a cost of business for the majority of them.

The sharper number sits underneath that one: fewer than one in ten marketing agencies monetise generative AI investments as a line of business. Agencies are funding the models, the subscriptions and the failed generations out of their own margin, then handing the output over inside a fee that was scoped before any of it existed.

Client expectations are pulling in the same direction. The AI benefit corporate marketers most frequently ask their agency for is cost efficiency, cited by 71 percent, well ahead of marketing performance at 49 percent and revenue growth at 27 percent. Left alone, that becomes a one-way ratchet. Every efficiency gain gets handed back as a fee reduction, and the capability that took two years to build shows up nowhere on the invoice.

Why Hourly Billing Breaks on Generated Video

Start with what the raw material actually costs. Google Cloud lists Veo 3.1 with synchronised audio at 0.40 US dollars per second of output at 720p and 1080p, dropping to 0.10 dollars per second on the Fast tier. An eight-second hero clip is therefore about three dollars of compute. Burn five takes to get one you would show a client and the usable eight seconds costs roughly sixteen dollars.

Scale that to a finished thirty-second cut assembled from six usable shots at the same reject ratio and the generation bill lands somewhere near a hundred dollars. That figure is the whole problem in one number. Under a time-and-materials model you used to bill a crew, a location, a shoot day and a week of post against that same deliverable, and the client understood exactly what they were paying for because they could see it.

This is where a clear-eyed view of what AI video production actually costs matters more than any rate card, because you cannot price sensibly against a cost base you have not measured. But knowing your cost per usable clip only tells you the floor. Setting the price from that floor plus a markup is the trap: it anchors the entire conversation to compute, which is the cheapest and least defensible thing you contribute.

Hourly billing fails for a blunter reason. It penalises every efficiency you build. Spend three weeks templating a pipeline that halves your turnaround and an hourly model rewards you by halving your revenue. No other product business prices itself that way.

Decide What You Are Actually Selling

Before choosing a structure, define the unit. Most agencies discover they are selling three different things under one invoice, and that each behaves differently commercially. The hero cut is the anchor concept, and it is a one-off that wins the account. The variant set is hooks, endings, cutdowns and aspect ratios derived from it. The placement set is the same material conformed to each channel's delivery spec.

Only the first of those is a project. The second is inherently recurring, because paid social burns creative on a short clock and the queue has to stay full. Teams already running a structured UGC testing system know the variant set is not an add-on to the hero spot but the reason the relationship continues month after month. Giving it away free with the hero cut is the single most common pricing mistake in this category.

Write the deliverable spec down before quoting. Duration, number of distinct concepts, number of variants per concept, aspect ratios, language versions, delivery formats, and the definition of accepted. A deliverable nobody has defined cannot be priced, and it cannot be defended when the client asks for one more thing.

Four Structures, and the Job Each One Fits

A per-deliverable rate card is the simplest and works for campaign bursts and irregular clients. Price each named output as a line item: hero cut, additional concept, hook variant, cutdown, language version, static pulled from a frame. The discipline it enforces is useful in itself, because it makes you assign a value to work you were probably absorbing.

A capped retainer with metered overage suits ongoing programmes and is where most teams eventually land. The retainer covers strategy, an agreed monthly volume, review cycles and reporting. Above the cap, volume is billed at a stated per-unit rate. The cap is the whole point: quote a flat monthly fee with unlimited output and your compute climbs while your price does not.

Volume tiers with a declining unit price work when the client's own model is testing throughput. Twenty videos, forty, eighty, each with a lower per-unit price, mirrors how the buyer thinks and rewards them for consolidating spend with you rather than splitting it across three suppliers.

An outcome-linked layer belongs on top of a base fee, never instead of one. The 4As' own research on agency compensation describes growing momentum toward value-based and hybrid models, and warns that pulling a fee apart into its cost inputs reduces the agency to a supplier of inputs rather than a partner delivering growth. The workable version is a base that covers fixed costs plus a bonus tied to metrics you can genuinely influence, such as speed to market or tested-winner rate. Exposing the entire fee to a client's business results transfers risk you do not control.

Build the Price From the Work That Did Not Disappear

Generation collapsed the cost of pixels. It did not touch the work that determines whether those pixels are usable. Direction still takes a person: deciding what the shot is for, what the performance has to do, and which of forty near-identical takes actually plays. That judgment is now the scarce input, and it is what the client is buying whether or not your invoice says so.

Review is the second irreducible cost, and the one teams systematically under-price. Somebody watches every generated frame for artefacts, brand accuracy, continuity and legibility, and rejects a meaningful share of them. Running a disciplined pre-delivery QC gate takes real hours per batch, and those hours are the difference between a deliverable and a folder of clips. Price the gate, not just the generation.

Delivery conformance is the third. Encode specs, loudness targets, caption files, aspect-ratio safe areas and platform-specific supers all have to be right or the asset gets rejected downstream. None of that is glamorous and all of it is billable. Together these three lines are the honest basis for a rate card, and they happen to be the parts a client cannot replicate in-house with a one-click tool.

Guardrails That Stop the Margin Leaking

State a revision policy in the contract and price beyond it. Two rounds included, further rounds billed at a named rate, is standard and defensible. What makes this bite on generated work is being specific about what a revision means, because a revision handled in the edit costs almost nothing while a change that forces a fresh generation costs a full cycle. Separating those two cases in writing prevents most fee disputes before they start.

Cap included volume and publish the overage rate. A retainer without a ceiling is an open compute tab with your name on it. One and a half to two times your unit cost above the cap is a normal shape, and it keeps high-volume months profitable rather than punishing.

Define a variant precisely. A new first three seconds against the same body is one thing. A new concept, new script and new performer is another, and clients will happily describe both with the same word. Write the boundary into the scope, along with what happens to unused volume at month end, because rollover is a silent margin killer.

Put It in the Contract, Then Measure It

Three clauses do most of the commercial work. Usage rights and term, stated per deliverable, because generated assets have no talent agreement to inherit terms from and the vacuum defaults in the client's favour. Disclosure responsibility, naming which party labels the ad in market. And a model substitution clause reserving your right to change the underlying tool, since the model you quoted on may be deprecated inside the campaign window.

Underneath all of that sits the question of whether you can sell the output at all. Settling whether the work is commercially rights-safe before a rate card exists is not legal housekeeping, it is what makes the deliverable saleable, and the clearance work itself is a billable line rather than an overhead you absorb quietly.

Then measure three things monthly. Gross margin per deliverable type, which tells you which line items on the rate card are actually earning. Cost per accepted clip, which tracks whether your direction is improving or your reject ratio is quietly eating the retainer. And revision ratio by client, which surfaces the account that looks profitable on paper and is not.

The agencies that move generative video off the cost line will not be the ones with the cheapest compute. They will be the ones who wrote down what they sell, priced the judgment rather than the pixels, and put a ceiling on the volume before signing.