Price the delivery system from the end backwards
List the assets that may actually ship: hero film, cutdowns, product stills, local versions, captions, landing images, media exports and any reserve replacement. For each, write the approval standard and the owner. Only then estimate exploration. This makes it clear whether ten routes are being requested to choose one hero or whether the campaign genuinely needs ten cleared, localized assets.
Separate four buckets: decision work, making work, finishing work and release work. Decision work includes briefing, research and selection. Making work includes capture, generation, design and edit. Finishing work includes retouching, accessibility, localization and rights review. Release work includes trafficking, exports, provenance records, monitoring and correction paths. A tool invoice fits inside one bucket; it is not the whole estimate.
Use approval capacity as the main constraint
A team that can only review twenty assets thoroughly should not plan to produce two hundred near-finals. Set a review capacity per round: number of routes, variants, markets and approvers. If the desired volume exceeds it, reduce the option set, stagger decisions, simplify the version logic or add qualified reviewers with explicit scopes. Do not solve the gap by pretending unreviewed files are finished.
The existing Ad Field Notes budget note makes the same editorial distinction: cheap generation does not make an approved campaign cheap. Our operational extension is to make approval capacity a line item and a calendar commitment. That puts the real scarce resource—responsible decision time—where a producer can manage it.
A clearly hypothetical example
Hypothetical planning example only, not a market rate: a team plans one 20-second master, four cutdowns and six stills. It budgets three named concept routes, two selected previsualizations, one approved production route, two review rounds per deliverable and one final release check. The estimate is built from internal day allocations for strategy, art direction, production, editing, product review, accessibility, localization and delivery—not from a claim that a certain model costs a certain amount.
The decision worksheet asks: if one route fails product review, what work is repeated? If a local language version needs a new layout, which part is reused? If an approved asset is withdrawn, who makes the replacement? Those contingencies are scenario assumptions. Label them as such, then revise them using the team’s own completed-job records rather than borrowing a vendor benchmark.
Close the estimate with an approved-asset ledger
For each final asset, record planned status, actual approval date, number of material revisions, finishing owner, and whether it was released. At wrap, compare the original assumption with what happened: where did scope grow, which source assets created rework, what approvals blocked, and which versions were never used? This is not a performance ranking; it is a more honest input to the next estimate.
Do not use generated image count, prompt count or nominal platform access as a success measure. They may be useful production telemetry, but they do not show whether the work was accurate, on-brand, localized, accessible or accepted. The ledger’s unit is the asset that cleared the defined release gate.
Sources & evidence limits
Source-backed facts are distinguished from the editorial workflow proposed here. Brand and agency accounts document their own work, not independent proof of performance. Read the linked source for its scope.
- NIST — AI Risk Management Framework Core
Context for documenting oversight responsibilities. The budget method and hypothetical example are Ad Field Notes editorial proposals, not NIST cost estimates.
Checked 2026-09-19 · Source publication date not established