How to find your best-performing Meta Ads
The hard part is not sorting by ROAS; it is making sure the ad at the top is genuinely good rather than lucky. A real winner clears a volume floor, holds up against ads that had a comparable chance to spend, and wins on the metric that matches its objective. Once you have found it, the useful output is not the ad itself but the pattern behind it — the hook, the angle or the format you can apply to the next batch.
To find genuine winners, check:
- Conversion volume first — below the floor, the ranking is noise
- Like-for-like comparison: same objective, comparable spend opportunity
- The metric that matches the objective, not ROAS on everything
- Hook rate and hold rate separately for video
- Whether the result survives a longer window
- Whether reported performance is inflated by view-through
- The pattern behind the winner — angle, hook, format
What separates a winner from a fluke
| Signal | Looks like a winner | Actually a winner |
|---|---|---|
| Conversion volume | High ROAS on 2 conversions | Clears 3–5 conversions in the period |
| Spend opportunity | Great rate on $200 of spend | Held up while carrying meaningful budget |
| Comparison set | Beat every ad in the account | Beat ads with a comparable chance to spend |
| Objective match | High ROAS on a traffic campaign | Won on the metric its objective optimises for |
| Attribution | Retargeting ad at 8× ROAS | Checked for view-through inflation |
| Durability | Best week it ever had | Still winning across a longer window |
The process
Step 1
Set the comparison set before you rank anything
Restrict to the ads making up the first 90% of cumulative spend, filtered to active entities. Ranking across every ad the account ever ran puts tiny-sample flukes at the top and buries the ads that actually carried the account. The question is not "which ad had the best rate" but "which ad performed best among those that had a real chance to spend".
Step 2
Apply the volume floor
Discard anything below 3–5 conversions in the period before you rank. A 12× ROAS on two purchases tells you almost nothing — it is within the range of chance for any ad. This single filter removes most false winners.
The same applies to segment-level claims. Before declaring an ad the winner for a particular placement or age bracket, require at least 100 impressions and 3 conversions in that segment.
Step 3
Rank on the metric that matches the objective
Evaluate each ad against what its campaign optimises for. ROAS on a sales campaign, cost per lead on a lead campaign, cost per ThruPlay and hook rate on a video campaign. Applying ROAS to everything makes upper-funnel ads look like failures and makes the account's best awareness creative invisible.
Step 4
Split the video funnel
For video, a single CTR number hides which part worked. Hook rate is 3-second views divided by video plays; hold rate is thruplays divided by 3-second views. An ad with a strong hook and weak hold has a different lesson than one with a weak hook and strong hold — and only the first tells you the opening is worth reusing.
- Good hook, poor hold — the opening works, the middle loses people. Reuse the hook, rebuild from three seconds on.
- Poor hook, good hold — the content is strong but not enough people stay to see it. Rebuild the opening.
- Good hook and hold, poor CTR — people watch and do not click. The CTA is the problem, not the creative.
The thresholds we use: hook rate at or above 40% is good and under 25% poor; hold rate is judged against the account's own 90-day average rather than a fixed number. These are our working figures, not an industry standard.
Step 5
Check durability and attribution
Re-run the ranking over a longer window. An ad that wins in one week and disappears from the top in a four-week view had a good week, not a winning format. And check the attribution profile before celebrating: retargeting ads and Advantage+ Shopping campaigns aimed at existing customers often post high reported ROAS that is largely view-through or would have converted anyway.
Step 6
Extract the pattern, not just the ad
The point of finding the winner is deciding what to make next. Look across the top performers for what repeats — the angle, the format, the opening, the tone of the copy. One winning ad is a result; a repeated pattern across three winners is a brief.
- Which selling point recurs across the winners?
- Image, video or carousel — is one format carrying them?
- For video, what kind of opening keeps appearing in the strong hooks?
- Copy structure — length, tone, whether it leads with problem or benefit
Step 7
Scale carefully
Having found a winner, resist scaling it hard. Increase budget by no more than 20% at a time, at the campaign or ad set level, and give it time to restabilise before the next step. Large jumps restart learning and frequently destroy the performance you were trying to buy more of.
You cannot allocate budget to an individual ad — budgets exist at campaign and ad set level. Scaling a winning ad means increasing its ad set's budget, or pausing the weaker ads around it so more of the existing budget flows to it.
A worked example
Illustrative example. The numbers below are made up to show the method. They are not a client account and not a benchmark.
Five ads, sorted by ROAS. The top of the list is not the winner.
| Ad | Spend | Conversions | ROAS | Read |
|---|---|---|---|---|
| Ad A | $310 | 2 | 7.4 | Below the volume floor — discard |
| Ad B | $9,800 | 118 | 3.6 | The real winner |
| Ad C — retargeting | $2,400 | 41 | 6.1 | Check view-through before believing it |
| Ad D | $8,100 | 76 | 2.9 | Solid, second place |
| Ad E | $640 | 4 | 4.2 | Clears the floor but little spend opportunity — watch |
Ad A tops the sort and is the least interesting line in the table. Two conversions is inside the range of chance; its 7.4 ROAS carries almost no information.
Ad B is the genuine winner — it held 3.6 ROAS while carrying nearly $10,000 of spend, which is far harder than posting a high rate on a few hundred dollars.
Ad C needs a second look rather than a promotion. Retargeting at 6.1 is exactly the profile that turns out to be view-through heavy, and those conversions may have happened anyway.
Ad E is a candidate, not a conclusion. It clears the floor but has not had the chance to spend. The right action is to give it more opportunity and re-check, not to crown it.
What not to conclude too quickly
Nearly every false winner comes from one of these:
- Ranking by rate without a volume floor. Small samples produce spectacular, meaningless rates.
- Comparing ads that never had a comparable chance to spend.
- Applying ROAS to campaigns not optimising for purchases.
- Believing retargeting ROAS without checking view-through share.
- Crowning a one-week result. Re-run it over a longer window.
- Scaling the winner too hard. Above 20% in a step, you risk resetting learning and losing the thing you found.
- Copying the ad instead of the pattern. A duplicate of a winner usually underperforms the original; the reusable asset is the angle or hook.
How to automate finding winners
Done properly this means pulling ad-level data with video metrics, computing the spend-weighted comparison set, applying volume floors, splitting hook from hold against a 90-day account baseline, and checking attribution profiles — repeatedly, as new creative enters rotation.
GoMarble runs that continuously and connects the creative attributes back to performance, so the output is the pattern worth repeating rather than a leaderboard.
The shape of it
- Connect the ad account
- Ask which creative is working, or schedule a creative review
- Ads are compared within the spend set, against the account's own baseline
- Hook, hold and CTR are separated so the lesson is specific
- Recommended next creative directions come attached
The useful output is a brief for the next batch, not a ranking of the last one.
Finding winners checklist
- ☐ Restrict to the first 90% of spend, active entities only
- ☐ Apply a floor of 3–5 conversions before ranking
- ☐ Rank on the metric matching each campaign's objective
- ☐ Compare only ads with comparable spend opportunity
- ☐ Split hook rate and hold rate for video
- ☐ Judge hold rate against the account's 90-day average
- ☐ Re-run the ranking over a longer window
- ☐ Check view-through share on high-ROAS retargeting
- ☐ Require 100 impressions and 3 conversions for segment claims
- ☐ Identify the repeating angle, format and hook across winners
- ☐ Scale at no more than 20% per step, at ad set or campaign level
Questions people actually ask
How many conversions does an ad need before I trust its ROAS?
Is the ad with the highest ROAS my best ad?
How do I know if a winning ad will keep working?
Should I just duplicate my winning ad?
How fast can I scale a winning ad?
Where this method comes from
The comparison method, the volume floors, the hook and hold formulas and the scaling constraint on this page come from the methodology GoMarble's agents run when they review creative on a live Meta account — the 90%-of-spend comparison set, the statistical significance floors, the video funnel split, the attribution caveats and the guardrail that budgets exist at ad set and campaign level only. Benchmark figures are ours rather than industry standards.
Related
Find the pattern, not just the leaderboard.
Connect Meta and GoMarble connects creative attributes to performance — so you get a brief for the next batch.