If you run more than a handful of ad variants a week, you already know the real problem isn't collecting data — it's deciding what the data means. Learning how to score creative test results turns a messy spreadsheet of clicks, views, and cost numbers into a single, defensible answer: which ad wins, which gets iterated, and which gets killed.
Why a single metric can't score a creative test properly
It's tempting to just look at ROAS or CPA and call it done. The issue is that one metric hides the reason a creative worked or failed. A video can have a strong hook rate but a weak offer, or a brilliant hold rate but a call to action that's too soft. If you only score on the bottom-line number, you learn whether an ad performed — but not why — and you lose the ability to reuse what worked in the next batch.
Scoring creative tests properly means tracking the ad across the funnel it's actually trying to move someone through: attention, interest, action, and cost efficiency. Each stage has its own metric, and each metric tells you something different about the creative, not just the targeting or the offer.
The core metrics that belong in a creative scorecard
You don't need twenty data points. A clean scorecard usually needs four to six metrics, each mapped to a stage of attention.
Hook rate (first 3 seconds)
The share of people who keep watching past the opening seconds. This is the purest signal of whether your opening line, visual, or pattern interrupt is doing its job, independent of everything else in the ad.
Hold rate (video retention through the middle and end)
Measures whether the body of the ad — problem, benefit, proof — keeps people watching once the hook has already worked. A good hook with a bad hold rate usually points to a weak middle section, not a weak opening.
Click-through rate
Tells you whether the message and call to action are compelling enough to move someone off the platform. A high hold rate with a low CTR often means the offer or CTA needs sharpening, not the creative's attention-grabbing elements.
Cost per acquisition or ROAS
The business-outcome metric. It matters most, but it's also the slowest and noisiest signal, since it's affected by audience, bidding, and landing page as much as by the creative itself.
Secondary engagement signals
Comments, shares, saves, and watch-to-completion can be useful tiebreakers, especially on TikTok and Reels where organic-style engagement often predicts how well an ad will scale once it leaves the test budget.
How to build a weighted scoring formula, step by step
Here's a repeatable way to turn those metrics into one comparable score per creative.
- Decide what 'winning' means for this specific test. A cold-audience awareness test should weight hook and hold rate heavily; a retargeting test should weight CTR and CPA heavily.
- Pick four to six metrics maximum. More than that and the score becomes noise instead of a decision tool.
- Normalize every metric against your own account average or median, not against an outside benchmark. For example: creative hook rate ÷ account average hook rate × 100. This gives you an index score, so a hook rate of 28% means something different on a page where the average is 20% versus 35%.
- Assign weights that sum to 100%, matched to the funnel stage from step 1.
- Multiply each normalized metric by its weight and add them up to get a composite score per creative.
- Rank creatives by composite score and sort into decision buckets: scale, iterate, or kill.
- Log the composite score and the raw metrics behind it, so next test you can spot patterns in what made top scorers score high.
A simple top-of-funnel weighting to start testing with: Hook rate 30%, Hold rate 25%, CTR 20%, CPA/ROAS 20%, Engagement 5%. For a bottom-funnel retargeting test, flip it: CPA/ROAS 40%, CTR 30%, Hook rate 15%, Hold rate 10%, Engagement 5%. Treat both as starting points to test against your own account, not fixed rules.
Worked example
| Creative | Hook rate index | Hold rate index | CTR index | CPA index | Composite score |
|---|---|---|---|---|---|
| Ad A | 118 | 95 | 80 | 70 | 93 |
| Ad B | 90 | 110 | 105 | 120 | 108 |
| Ad C | 140 | 60 | 70 | 55 | 87 |
In this example, Ad C has the strongest hook by far but a weak hold rate and weak downstream numbers — a sign the opening is strong but the middle of the ad loses people. Ad B isn't the flashiest opener but scores highest overall because it holds attention and converts efficiently. Ad A sits in between. On raw CPA alone you might have crowned the wrong winner or missed why Ad C's hook is worth reusing even though the full creative underperforms.
Setting benchmarks and thresholds before you score
A score is only useful if it's compared against something stable. Before you start scoring, fix two things:
- Minimum sample size. Don't score a creative off a handful of impressions or clicks — give every variant enough delivery to reach a reasonably stable rate before its score counts, and treat early scores as provisional.
- A rolling baseline. Recalculate your account averages every few weeks so new tests are compared against current performance, not a baseline from months ago that no longer reflects your audience or offer.
If you're still deciding which variables are worth testing in the first place, it helps to lock that down before you build a scoring system — see Creative Testing Variables to Test First for a priority order that keeps your tests focused enough to score cleanly.
Common scoring mistakes to avoid
- Scoring too early. A creative that looks like a loser on day one can catch up once delivery stabilizes — wait for your minimum sample before calling it.
- Using the same weights for every test. A launch test and a scaling test are measuring different things; the weights should move with the goal.
- Ignoring the reason behind the score. A low composite score with a strong hook rate is a different problem than a low score with a weak hook rate — the fix is not the same.
- Comparing across different placements or formats without normalizing. A score from a Reels placement and a Stories placement can look similar but represent very different attention conditions.
- Not writing the score down anywhere. If scores live only in your head, you can't spot which elements (hook style, proof format, offer framing) keep scoring well across multiple tests.
If you're not sure your raw data is even clean enough to score reliably, it's worth checking your setup against How to Organize a Creative Testing Spreadsheet first — a scoring formula is only as good as the columns feeding it.
Turning scores into decisions
Once every creative in a test has a composite score, sort them into three buckets rather than a long ranked list. A ranked list invites overthinking; buckets force a decision.
| Bucket | Typical signal | Action |
|---|---|---|
| Scale | Composite score well above account average, stable across enough delivery | Increase budget, move to broader audiences, keep as-is |
| Iterate | Strong on one or two metrics (often hook rate) but weak downstream | Keep the winning element, rebuild the rest — new middle, new CTA, or new proof |
| Kill | Below average across most metrics with enough delivery to be confident | Pause and reallocate budget; archive the learning, not just the ad |
The 'iterate' bucket is where most of the long-term value sits. If a creative's hook rate is far above average but everything after it drops off, that opening line or visual is worth reusing in a new ad with a different body. This is also where it helps to have already flagged the underperforming parts of a creative correctly — 9 Signs an Ad Creative Is Underperforming walks through how to confirm a weak signal before you act on it.
Whatever you decide, write it down next to the score. A scoring system that isn't paired with a documentation habit loses most of its value after a few weeks — see How to Document Creative Testing Learnings for a format that keeps scores and the reasoning behind them in one place you can search later.
Scoring results when you need to show them to someone else
If you manage creative testing for a client or a stakeholder who isn't in the data daily, a composite score is also the easiest thing to present — it's one number they can track over time, with the underlying metrics available if they want to go deeper. For a format that works well in that setting, see How to Present Creative Testing Results to Clients.
Where FrameNotion fits into a scoring workflow
A scoring system is only as useful as the number of fresh creatives you can feed into it. FrameNotion speeds up the production side: paste a product link and FrameNotion AI writes and renders a custom 30-second vertical ad — hook, problem, benefit, proof, offer, and call to action — in about 10 to 20 minutes, with no templates. That makes it practical to generate several hook variations for the same offer and let your scorecard tell you which opening actually earns the highest hook-rate index before you commit media budget to it.
Because every ad also renders as 4:5, 1:1, and 16:9 alongside the 9:16 version, you can run the same creative across placements without re-scoring from scratch each time, and because change requests let you adjust text and color without starting over, the 'iterate' bucket from your scorecard becomes a quick edit rather than a full rebuild. FrameNotion doesn't publish ads or report on performance — scoring still happens in your ad platform or spreadsheet — but it removes the production bottleneck that usually limits how many variants you can actually test.
You can see finished examples at examples, check how the process works at features, or start with the first ad free at FrameNotion.
Frequently asked questions
Should I score creatives against each other or against fixed benchmarks?+
Score against your own account's recent average or median, recalculated every few weeks. Fixed outside benchmarks don't account for your audience, offer, or placement mix, so an index against your own baseline gives a more honest comparison between creatives.
What sample size do I need before a creative's score counts?+
There's no universal number, but treat any score built on very few impressions or clicks as provisional. Wait until delivery is high enough that the rate (hook rate, CTR, etc.) has stopped swinging significantly before you let it influence a scale or kill decision.
What do I do when metrics disagree, like a high CTR but a high CPA?+
That pattern usually points away from the creative and toward the landing page, offer, or audience, since the ad is clearly earning clicks. A weighted scorecard still ranks the creative fairly on what it controls, while the gap between CTR and CPA becomes a separate flag to investigate outside the ad itself.
Should hook rate matter more than CPA for a brand-new product launch?+
Generally yes, as a starting point. Early in a launch you're still learning which openings earn attention, and CPA is noisy until targeting and landing pages settle. Weight hook and hold rate heavier at launch, then shift weight toward CPA and ROAS as the campaign matures.
Can I use the same scoring formula across TikTok and Meta?+
You can use the same structure, but recalculate the index baselines separately for each platform, since attention and click behavior differ by placement. Keep the metrics and weights consistent for comparability, but never compare raw or indexed numbers across platforms directly.