How to Use Pencil AI Scores to Decide Ad Budget

November 17, 2026

Predicted scores from Pencil AI are most useful as a budget allocator: put more money behind high-ranked creatives, less behind the rest, and set a kill rule before launch. Used that way, teams commonly cut wasted test spend without losing winning creatives. The scores are relative rankings, not forecasts of your conversion rate.

Most teams use the score only to pick which creative to launch. The larger gain comes from how you split spend across the ones you do launch.

What does a score actually tell you?

It tells you one creative is predicted to beat another against the model’s benchmark of similar ads on the same placement. That is a relative statement. A score of 82 does not mean 8% CTR.

Because it is relative, scores from different briefs and different placements are not comparable. Rank within a brief. Do not build a leaderboard across campaigns.

How do you split budget from the ranking?

A workable three-tier pattern:

  • Top two by score: 50% of the test budget, run as the primary challengers.
  • Next three: 35%, as supporting variants with the same audience.
  • Bottom of the selected set: 15%, held for fresh audiences where the model has less signal.

Re-rank after the first week of real data. Real performance rarely matches prediction exactly, and the point is to let reality override the model, not the other way around.

What kill rule stops budget bleeding?

Decide before launch: if a creative is below your account’s baseline CTR by 30% after 1,000 impressions, pause it. Write that number down. In-platform, teams otherwise let a losing creative run for a week because pausing feels like admitting the prediction was wrong.

It was a prediction. Being wrong is part of the design.

How do predictions perform against a plain A/B test?

A plain A/B test is more accurate for one specific creative pair and costs more to run. Prediction is cheaper per creative and covers many at once. The practical combination: predict broadly, A/B test narrowly on the finalists.

If you already run a disciplined test schedule with a strong creative pipeline, the marginal gain from prediction drops sharply. Pencil AI pays off when you produce more creative than you can test.

Where the budget logic breaks down

Margin. Score says nothing about profitability. A high-score creative that sells a loss-leader is a loss.

Also audience saturation: the model does not know that you already hammered this segment for six weeks. If frequency is above 3 on a segment, no creative score rescues the campaign.

Quick Summary

  • Scores are relative rankings within a brief, not absolute forecasts.
  • Split test budget 50/35/15 across top two, next three and the rest.
  • Write a kill rule before launch (e.g. pause at 30% below baseline CTR after 1,000 impressions).
  • Predict broadly, A/B test the finalists.
  • Scores ignore margin and audience saturation.

Platform coverage and setup requirements are on the Pencil AI tool page.