Pott Candles · testing review · 28 Oct 2026
Does a test winner win again in the main campaign?
We matched every ad created in 2026 across the testing campaigns and the main campaigns (same post, video or image), then compared each one's cost per sale before and after the move. Here is what the account actually says.
The short version
- A good cost per sale in testing mostly carried. Six of the nine moved winners held or improved at the spend they reached, and the ad set around them held too. The two that got worse both did so in the single bad week of mid-September, when the whole ad set was at £54.
- A bad cost per sale in testing has never turned into a good main ad. Not once in 2026.
- How much Meta spent on an ad in the test told us nothing. The ad that got the most test spend was rarely the one that went on to scale. Two April ads that got under £30 in the test became the main campaign's top spenders.
- The Fir winners have not failed in main. They have not been fed. The Christmas ad set went from £3,547 a day to £509 a day after the 26 Oct budget change, and the five Fir copies have had £152 between them.
1. The one time we moved everything over
April's spring test ran 14 ads. On 25 April we copied 11 of them into the Ocean and Sea Salt campaign as new posts. This is the only clean "move the lot" case we have, so it is the only place we can check whether the test ranking repeats.
- The ranking did not repeat. Meta's spend order in the test and its spend order in main are unrelated (rank correlation 0.06, where 1 would be identical).
- Cost per sale did carry. Ad 378 won the test at £28 and ran at £29 in main. Every ad that was £46 or worse in the test (379, 380, 386, 388) stayed bad or got nothing.
- The two unread ads became the winners. Ads 384 and 385 got £26 and £15 in the test. In main they spent £887 at £42 and £234 at £33, the most of the group. Not feeding an ad in a test is not the same as the ad losing.
2. Nine winners moved one at a time
Since July we have mostly moved single winners rather than whole tests. Nine of those have had long enough in main to read. Each row shows the cost per sale in the test, in the main campaign, and (grey tick) the blended cost per sale of the whole ad set over the same weeks, so a change can be read against what the account was doing at the time.
Became the biggest ad of the summer. Ad set held £41 while weekly spend went from £3.1k to £8.5k.
Only 3% of the ad set, so a modest read.
Citronella ad set went from £50 to £35 per sale in the weeks it ran.
Beat the rest of its ad set (£18 vs £36).
Better than the rest of the Seasonal set (£38 vs £45) on 7% of spend.
Good cost per sale but Meta gave it 1% of the set. A second copy in Seasonal did badly on £303.
Meta pushed £2,406 into it in the week the whole ad set hit £54. The next week it was back to £30.
£320 in the same bad week, £67 vs £54 for the set. Not enough to call.
£47 in three months. Not a loser, just unread. It was not a test winner anyway (£51).
3. What scaling does to the number
Ad 707 is the best example of why an ad's own cost per sale is not the whole read. It was only third by spend in its July test (£180 at £45). Moved into Evergreen in August, Meta gave it up to half the ad set and its cost per sale drifted from £29 to £56 as the weekly spend climbed. Yet the ad set it sat in went from £3.1k a week to £8.5k a week and stayed around £41 blended. That is a win for the campaign even in the weeks the top ad looked expensive on its own.
On a phone, swipe the charts sideways. Columns: weekly spend on Ad 707 (orange) and the rest of its ad set (grey), with Ad 707's share of the set above each column. Lines: cost per sale that week for the ad and for the rest of the set.
The flip side: Ad 855
Ad 855 tested at £26. In its first week in Evergreen Meta spent £2,406 on it at £78, while the rest of the ad set that week was at £48. That was the account's worst week of September across every ad set. The week after, on £420, it was back to £30. So it was fed too hard in a bad week rather than exposed as a bad ad, and we cannot fairly call it either way.
4. Tested first, or straight in?
Since June, 220 ads went into Evergreen, Gifting or Cook without a test. Only 11 came through a test first. The gap is too big for a fair comparison, but the pattern is clear: testing got ads funded. It did not obviously make them better once funded.
- Half of the straight-in ads never reached £100. Meta decided on day one and never came back to them.
- Once an ad reached £300, roughly the same share landed at £45 or under whichever route it took (43% straight in, 50% tested). Blended cost per sale: £45 straight in, £41 tested.
- The biggest ads of the year went in both ways. Ad 1000 and Ad 922 were straight in. Ad 707 was tested.
5. The Fir winners have not been read yet
The five Fir winners were copied into the Evergreen Christmas ad set on 27 October. Since then they have had £152 between them, so there is nothing to read. The reason is budget, not the ads: the campaign moved to £2,000 a day on 26 October, the three smaller ad sets kept their £300 to £400 floors, and the Christmas ad set absorbed the whole cut.
- Christmas ad set: £3,547 on 24 Oct, £701 on 27 Oct, £509 on 28 Oct.
- Ad 1000 still takes a quarter to a third of whatever is left, and the Icicle pair (1066, 1067) copied in on 26 Oct took most of the rest.
- The Fir originals are still live in both Fir test campaigns on the same posts, spending £311 on 27 and 28 Oct. So the creative is still being bought, just not in the main campaign.
6. The same ad in a different ad set is a different ad
Ailis's point on the call stands. The same post, in a different ad set, has given very different results. Ad 420 (the Flora subscription video) has run in five places:
Ad 852 (Autumn seasonal static): £21 per sale in Evergreen Ad Set 1, £101 in the Seasonal set a fortnight later. Blue bars are testing campaigns, orange are main campaigns.
What we suggest through to the end of the year
- Read a moved ad on two numbers, never one. Its own cost per sale at the spend it reached, and the ad set's blended cost per sale over the same weeks. A top ad drifting to £50 while the ad set scales at £41 is a success. A top ad at £78 in a week the set is at £48 is a question, and one week is not an answer.
- Treat an unfed test ad as unread, not as a loser. If it had under £100 in the test, it has not been judged. Ads 384, 385 and 707 all came from that pile.
- Run Ailis's Flora test as asked. Move all four Flora ads from Test 3 into a main ad set together, same posts, pause the Test 3 originals, read after seven days on both numbers above. Put them in Gifting or Everything Else, not the shrinking Christmas set, or the read will be about budget again.
- Give the Fir copies a fair week. Either restore Christmas ad set headroom or pause the Fir originals in the two test campaigns so the budget follows the posts into main. Read on 4 November, not before.