What belongs in the number
The temptation is to count the generations that made it into the cut. The honest version counts everything the accepted minute consumed.
- Every generation attempt, not only the accepted one
- Repair attempts, and repairs that failed with collateral change
- Human review time per accepted minute - operators are not free
- Provider failure and retry cost
- Localization work per language track
- Storage, delivery and support
- Variance between the estimate, the authorization and the final cost
Reviewer time is the line most often omitted and frequently the largest. A workflow that halves generation cost while doubling the number of results a person has to look at has usually made the production more expensive, not less.
The pricing positions that follow
- Sell accepted packages before seats
- A seat licence prices access. During validation what a customer is buying is an accepted production, so that is what gets priced.
- Show estimate and ceiling before authorization
- Two numbers, both visible, before anything is dispatched. A budget discovered afterwards is not a budget.
- Separate planning from rendering
- Review and planning are labour; rendering is variable cost. Bundling them hides which one is actually expensive.
- Do not charge for provider failures
- The party choosing the router should carry the cost of choosing badly. That is an incentive, not a courtesy.
- No unlimited generation
- Unlimited only works by discouraging use or degrading quality. Generation is metered or authorized against a ceiling.
- Target 70% contribution margin before scaling
- A production that performs but loses money is a failure. So is a consistent one nobody accepts.
The number we do not have yet
Frequently asked questions
Why not just compare price per second of generated video?
Because a second you reject cost the same as a second you keep. Per-second pricing compares suppliers on the one dimension that does not determine your cost, which is why it is the number most often quoted.
What is an accepted minute?
Sixty seconds of publish-ready output that an authorized human has accepted for the intended delivery. Not generated, not rendered, not internally liked - accepted, by someone with the role to accept it.
How do you keep the estimate honest?
By tracking estimate-to-actual variance as a metric in its own right and reporting it. An estimate nobody checks afterwards is a sales number.
Does a cheaper provider lower cost per accepted minute?
Only if its acceptance rate holds. That is why acceptance performance is a routing input: a provider that is cheap per attempt and needs many attempts is the expensive choice, in money and in reviewer hours.
Will Tosheo ever publish a price list?
Platform fees or seats are introduced only after repeated customer behaviour reveals stable roles and usage, and generation stays metered or budget-authorized regardless. Until then, pricing is scoped per production and we would rather say so than post a number we would have to renegotiate.
