What makes an ad a winner
"Winner" only means something against a rule you set before launch. For most online stores, that rule is about sales:
- The metric: cost per purchase (CPA) at or under target, or ROAS at or above target.
- The volume: enough purchases that it isn't luck. One sale proves very little.
- The time: long enough for delivery to settle, so a lucky first morning doesn't count.
BulkAds' Autopilot uses a clear version of this rule on the tests it launches: after 48 hours, an ad with three or more sales at or under your target CPA is a winner. You set the target, the rule does the rest.
What a winner isn't
- The ad with the most likes. Engagement is nice, but purchases pay the bills.
- The ad with the best CTR. Lots of clicks and no sales usually means the ad attracts curious people rather than buyers, or promises something the landing page doesn't deliver.
- The ad Meta spent the most on. Delivery leans towards ads the system expects to perform, but spend alone isn't proof that they did.
A worked example
Say Brewly, a fictional coffee brand, sets a target cost per purchase of $5.15 and tests eight ads. After 48 hours:
| Ad | Spend | Purchases | CPA | Verdict |
|---|---|---|---|---|
| Morning ritual, UGC-style | $25 | 6 | $4.10 | Winner |
| Café price comparison | $15 | 3 | $5.15 | Winner |
| Bean origin story | $10 | 0 | — | Stopped at 2× target |
| Gift box, bold claim | $9.30 | 1 | $9.30 | Needs more time |
The other four ads have spent too little to judge, so they get more time too.
What to do with a winner
- Scale in steps. Raise the budget gradually. Big jumps can make results less stable and may count as a significant edit that sends the ad set back into the learning phase.
- Make variations before it fatigues. Same angle and offer, new hooks, scenes and formats.
- Learn from it. Note its angle, hook style, format and size, and feed them into the next batch.
- Keep testing. Every winner fades eventually.
On those tests, Autopilot scales a winner one step at a time — +25%, +50% or +100%, depending on the pace you choose, and never above your daily limit — and seeds four fresh variants of it. In the default ask-first mode, it proposes the budget increase rather than applying it.
Questions people ask
How many sales does an ad need to count as a winner?
Decide before you launch. More sales make the verdict more trustworthy. For the tests it runs, Autopilot's default rule is three or more sales at or under your target cost per purchase after 48 hours.
Should I turn off my other ads once I find a winner?
Not necessarily. Stop the clear losers, but keep testing new ideas alongside the winner — it will fatigue eventually, and you'll need replacements ready.
Why did my winning ad stop performing?
Usually creative fatigue — the audience has seen it too often. Before replacing it, also check for seasonal auction pressure, landing page changes and tracking problems.
Related terms
Put it to work
BulkAds AI is an independent product and is not affiliated with, endorsed or sponsored by Meta Platforms, Inc. Facebook and Instagram are trademarks of Meta Platforms, Inc.