Creative Research
Work

Paying to Find the 5%

Most ads never win, and the usual way to find out is to pay for the test. We built a way to rank new concepts against an account's own winners before any money goes in.

Case study · 2026 · 2 min read

Ask a media buyer how they know a creative will work and the honest answer is that they launch it and wait. That made sense when targeting did most of the work. It makes much less sense now.

Why guessing got expensive

Winners are rare. In Motion's 2026 benchmarks, covering more than 550,000 Meta ads and about $1.3 billion of spend, roughly 5% of ads spent at least ten times their account's median. About half got little or no spend at all.1

Creative is the biggest lever. NCSolutions' 2023 analysis of about 450 campaigns credited creative with 49% of the sales lift from advertising, ahead of brand, reach and targeting.2

Creative now picks the audience. Since Meta rebuilt ad retrieval with Andromeda, its own guidance says the focus "has shifted from niche targeting to creative diversification."3 A weak creative costs you reach as well as spend.

Every test has a floor. An ad set needs around 50 results in a week to leave Meta's learning phase.4 Testing ten concepts to find one winner is a real line in the budget.

What we built

  • The account's memory. We pull 12 to 24 months of ads with their spend and ROAS.
  • Every ad tagged the same way. Hook, first three seconds, format, persona, offer, angle, creator and text overlay.
  • A winner defined by the data. An ad that spent ten times the account's median, not one somebody liked.
  • Two scores for every new concept. How close it sits to past winners, and how different it is from what's already live. Meta groups look-alike ads together,5 so a near-copy of a winner ends up competing with itself.
  • A shorter test list. The bottom third is cut before production. Test budget goes to the top band plus one wildcard.
  • A check on itself. Every round logs predicted against actual, so the scores have to keep earning trust.

What it can't do

  • It ranks. It doesn't promise. The auction is still the final test.
  • New accounts start thin. With little history, and most of it losers, early scores are rough.
  • Every account gets its own model. What wins for one brand says little about another.

Sources

  1. Motion, Creative Benchmarks 2026 (vendor research), on hit rates and spend concentration.
  2. Inside Radio, What drives advertising effectiveness, reporting NCSolutions' 2023 study.
  3. Meta, The creative advantage: diversification with Meta Andromeda.
  4. Ignite Social Media, Exiting the Facebook learning phase, on the 50 events in 7 days guideline.
  5. Meta, Demystifying creative diversification, on grouping ads with shared visual and thematic attributes.