What the report actually provides
Runway's July AI Media Report says it examined hundreds of enterprise customers and measured time and cost changes through customer-reported comparisons with similar work in the previous year. It describes adoption inside large organizations, including a group platform serving thousands of daily users and an independent film workflow using extensive generative video for full-film animatics before live action.
The cases establish that AI video is entering enterprise media workflows rather than remaining an individual experiment. The report does not disclose client identities, shot samples, contract prices or controls that readers could reproduce. Claims of two-to-three orders of magnitude in savings remain customer-reported findings in vendor research.
Why one generation price is not a budget
Production cost is incurred at approved delivery, not when a model returns a file. Rejected candidates, queueing, prompt and reference preparation, review, identity-drift repair, audio fixes, upscaling, editing and compliance records all belong in the cost.
The same model therefore has different economics in animatics, ad variants and serialized drama. Animatics can tolerate rough detail while valuing speed; serial drama needs character, world and narrative continuity. An average saving that ignores shot class hides production risk.
A ledger that can be reproduced
Every shot record should retain target duration, model and version, input hashes, run count, model spend, queue and operator time, failure reason, post repair and accepted seconds. Weekly summaries should include first-pass rate, average retries, cost per approved shot and cycle time.
Routes become comparable only with aligned assets, versions, periods and acceptance rules. Vendor studies can help a board understand adoption direction; project budgets have to come from the production's own ledger.