Which paid YouTube placements should we keep and which should we cut?
Built the classification and efficiency-tiering workflow, and refreshed the exclusion lists
April to October 2026
Ad platform exports, scripted classification, spreadsheet
The question
A video-ad programme bought views on YouTube across a very large universe of videos. Which placements sat close enough to a buying decision to deserve spend, and which were only cheap attention?
The method
I worked in two passes. First, I classified every video by funnel stage and by intent: business, website or ecommerce. I used title rules plus a transcript check where a transcript existed, because titles alone misread intent. Only bottom-funnel placements stayed on the keep list. Ambiguous ones went to manual review instead of being forced into a bucket.
Second, I tiered the keepers by average cost per view behind a minimum-sample gate, so a placement with a handful of views was never labelled inefficient on noise. The worst tier became the candidate list for negative placement exclusions.
I ranked on cost per view only after the funnel filter, because cheap views on irrelevant content look efficient and would have rewarded exactly the wrong placements.
The evidence
| Step | What it does |
|---|---|
| 1. Funnel filter | Only bottom-funnel placements stay. Ambiguous ones go to manual review. |
| 2. Sample-size gate | Placements with too few views are never judged. |
| 3. Cost-per-view tiers | The worst tier becomes the candidate list for exclusions. |
The result
Placements went from working about half the time to working nearly all the time. The share judged effective almost doubled, and click-through rate more than doubled.
The programme also moved from ad hoc judgement to a repeatable keep, remove or review list. I refreshed it at least three times between April and October 2026. I then reused the same classified universe to size how many qualifying bottom-funnel videos exist.