Your agency delivers a batch of product videos. Automated delivery puts most of the budget behind a few versions. The rest move into a “failed ads” folder, which becomes negative reference material for the next AI-generated batch. The workflow is fast. The labels may be doing more work than the evidence allows.
Low spend describes what happened in delivery. By itself, it does not explain whether a creative was eligible, received enough observation or produced an interpretable customer response. Brands working with China-based product and production teams should settle that distinction before turning a report into the next shoot brief.
Performance data is entering the creative workflow
Runway's October 27, 2026 announcement opened Runway Ads to paid plans and described generating new variants with performance context from connected accounts. Rechecked November 4, it establishes the company's stated workflow direction, not a transferable customer result or causal proof about a particular scene.
TikTok's October 2026 Product GMV Max exploration documentation distinguishes availability for testing, exploration and signal collection from ineligibility caused by authorization, rejection, exclusion, unavailability or inactivity. Those are product-specific descriptions, not universal statuses for every ad system. Their practical implication is that superficially similar delivery outcomes can require different actions.
The ONCE creative feedback card
This is a proposed review method, not a platform feature. Keep a stable creative ID and record the conditions behind the result before assigning a learning label.
| Check | Record | Decision it informs |
|---|---|---|
| Eligibility | Account, intended use, permissions and actual approval or blocking reason | Fix a setting, asset issue or cooperation condition first |
| Observation opportunity | Live interval, delivery setup and available spend and impression records | Evaluate whether a judgment is supported or remains unresolved |
| Observed outcome | Agreed objective, metric, cutoff date and relevant audience | Form a next hypothesis without inventing a cause |
| Production action | Retain, investigate, test, correct or remake, with a reason | Turn the review into an executable assignment |
If information is unavailable, say so. Do not infer an unreported eligibility status from low spend. “The platform allocated little budget” and “customers dislike this story” are different claims.
Three low-spend assets, three possible responses
Imagine a fictional portable-cleaning-tool campaign, not an ONCE client case. An installation demonstration cannot enter normal delivery because permission for the intended use is unresolved. A maintenance tutorial has only just started and has little observation. A use-case video has received more observation under documented conditions, but its relevant outcome falls short of the team's predefined objective.
The installation asset needs the permission problem addressed. It supplies no customer response that would justify labeling installation demonstrations a failed direction. The tutorial remains unresolved; the team can decide whether further observation is worthwhile. The use-case asset deserves investigation, including audience, product, destination and setup. An unfavorable aggregate result still does not identify the story as its sole cause.
All three may receive no additional production immediately. That can be a reasonable resource decision for different reasons. Preserve those reasons so a future product or market team knows which ideas were actually evaluated and which were merely unavailable or insufficiently observed.
Unresolved does not mean unlimited testing
Every asset does not need equal budget. Nor should a team keep spending indefinitely to vindicate an idea. Media and analysis owners should define a suitable observation plan using the objective, costs and available traffic. This article provides no universal impression threshold, spend amount or number of days.
If the budget ends before the question is answered, record the limitation. When a specific creative comparison matters, design an appropriate comparison separately. TikTok's current Split Testing documentation describes controlling other variables and separating audience groups. Two videos running together in ordinary automated delivery are not automatically that experiment.
Give the next AI brief the uncertainty as well
Replace the simple winner-versus-loser archive with explained references: assets that performed well under specified conditions, assets needing improvement under observed conditions, and assets for which the evidence is insufficient. The last group need not be used as generation input. It should not silently become a negative lesson.
A brief might say: “Use A's clear product action as a reference for a new maintenance opening. B has insufficient observation, so do not exclude the maintenance theme because of B. Explain which record supports each proposal and what remains to be tested.” This is a suggested briefing practice, not a claim that a particular tool implements it reliably.
The concern here is reference selection, prompts and a brand's internal knowledge base. It is not a claim that these records train a vendor's foundation model. Share client information with outside services only within the actual permitted scope, without unnecessary personal or transaction data.
Buy the feedback work as part of content operations
A recurring creative scope can separately identify asset organization, mapping to delivery records, review classification and the next production recommendation. Receiving a dashboard does not entitle a production partner to promise sales lift. Giving an editor a single “bad” label does not specify what to change.
Keep a reviewable action from each cycle: which asset was fixed, tested further or rewritten, why, and when it will be reconsidered. For ONCE's longer-term creative-data model, conditional learning is more useful than an expanding blacklist. Any cost or margin benefit would still require actual labor, rework and business records.
Further planning topics are available in ONCE's video marketing articles. For the next production brief, identify what the low-spend record actually establishes before using it to rule out a creative direction.
Sources and scope
Checked November 4, 2026: Runway Ads Is Generally Available (October 27); TikTok's Product GMV Max creative exploration guidance (updated October 2026) and About Split Testing (updated January 2026). No ad account was connected and no product was tested. Vendor-reported improvements are not used as forecasts. The feedback card, examples and cover are original analysis or fictional illustration.
