Your production partner has delivered two product ads. The media team gives them matching budget allocations, but one receives far more impressions. Should the team force exposure to match before evaluating the creative—or ask the supplier to replace the less-delivered video?
Neither follows from the impression count alone. ONCE recommends reviewing the intended settings, actual delivery and target outcome separately before commissioning the next batch.
Unequal impressions can be part of auction delivery
Google’s video experiment monitoring guide explicitly describes unequal impressions despite evenly allocated budgets: campaigns enter different auctions and win at different bids. It does not treat every such difference as experiment failure.
That is not proof that every imbalance is harmless. A disapproved ad, incorrect schedule or broken destination still needs diagnosis. Nor does it make a comparison between any two ordinary ad reports a causal experiment.
The official creation guide distinguishes basic A/B video asset testing from the advanced Custom path. The basic path makes creative the single variable and duplicates the control campaign to preserve other settings; availability is limited to specified video reach and video view campaign types. Have the media owner confirm the actual route rather than describing two uploaded ads as a completed platform experiment.
Use ONCE’s three-column review
| Column | Record | Question answered |
|---|---|---|
| Intended settings | Experiment route, asset IDs, tested variable, split rule, budget, dates and primary metric | What was the planned comparison? |
| Actual delivery | Spend, impressions, serving status, time coverage and available placement or audience information | How did the arms obtain exposure? |
| Target outcome | Preselected success metric, reported result status, business conversion definition and data completeness | What production decision is supported? |
This is ONCE’s proposed review method, not a Google report template. Equal settings do not promise equal delivered volume. More impressions do not establish superiority on the business goal. A metric lead does not establish that the evidence is sufficient.
A numerical example without an invented winner
The following figures are hypothetical arithmetic, not client results, industry costs or a forecast. Each arm spends $1,000 during the same observation window.
| Measure | Video A | Video B |
|---|---|---|
| Impressions | 100,000 | 50,000 |
| Clicks | 600 | 400 |
| CTR: clicks ÷ impressions | 0.6% | 0.8% |
| Average CPC: spend ÷ clicks | About $1.67 | $2.50 |
A delivers more impressions and clicks; B has the higher click-through rate. If the purchase decision depends on qualified inquiries, that outcome is missing. The table cannot tell the brand which opening deserves ten more versions.
Even if CTR was the primary metric, 0.8% exceeding 0.6% is not sufficient on its own to declare a verified winner. The design, tool result and data conditions still matter. The useful response is to identify the missing evidence, rather than automatically reshoot B or change the goal to whichever number looks strongest.
Do not change the comparison merely to equalize a chart
If both arms are serving normally, unequal exposure alone does not justify changing bids, audiences or budgets mid-test. Trying to catch one arm up can turn an opening-shot test into a combined creative-and-delivery test. Google’s creation guide instructs advertisers to end the video experiment before changing campaigns; the media owner should manage the appropriate action and any subsequent test.
Match the next purchase to the diagnosed problem
- Delivery is obstructed: investigate approval, schedules and technical status. Repair an asset if it is actually responsible; do not interpret a serving fault as audience dislike.
- Delivery is working but the result is unresolved: preserve the analysis plan and budget constraints, and record the uncertainty before ordering a large derivative batch.
- The planned analysis supports a decision: commission follow-up work for the tested conditions, without calling the result universal across markets or channels.
There is no universal impression-imbalance threshold or observation period offered here. Delivery needs diagnosis; an experimental conclusion needs its applicable method and conditions; the production order needs a clear scope for reuse.
Keep the reasoning with the creative
A test-ready handoff should identify the two revisions, the single intended change and the product claims and destination information held constant. Use asset IDs the media team can map to its setup. After the experiment, retain the settings, delivery, outcome and approved production action together.
For a global brand working with China-based production partners, that record can clarify whether the next job is a delivery repair, more analysis, a recut or a new shoot. For ONCE’s managed creative service, it is more reusable than a folder labelled Winner. No margin improvement or customer return is assumed.
The separate question of stopping after an early lead requires its own analysis. ONCE’s video marketing articles cover broader decision planning; this guide concerns the distinction between allocation and delivered exposure.
Evidence and limits
The two official Google Ads guides were checked on November 2, 2026. Precise publication dates were not displayed. This is a current-source evergreen guide, not a new-feature announcement. No customer account was accessed and no live experiment was conducted. The example is hypothetical, and the three-column review is ONCE’s proposed method.
