Pott Candles · testing review, part 3 · 29 October 2026
Every new ad is a hand of cards. Nobody, Meta included, knows in advance which one will scale. What we can control is how many hands we actually get to see, and how good a seat each one starts from. This page puts every question we have asked about testing in one place, with what the data says and what it cannot tell us yet.
Scorecard
Mostly, yes. Six of nine winners moved one at a time held up, and so did the ad set around them. Two looked worse only in the mid-September week when the whole ad set struggled. And no ad that did badly in testing has ever done well in main.
Detail and every chart: test winners vs main campaign.
No. In April we moved a whole test over, and the order Meta spent on the ads in testing had nothing to do with the order in main. Two ads that got £26 and £15 in the test became main's top two spenders. Ad 707 was only third in its test and became the biggest ad of the summer.
Spend on the ad, April test vs main campaign afterwards.
Only about half the time. In 45 batches launched on the same day, the ad Meta spent most on in week one was still its top spender at the end 29 times. It ended up with the best cost per sale only 20 times.
Week-one top spender is still the top spender later: 29 of 45
Week-one top spender ends with the best cost per sale: 20 of 45
About half never get looked at. Of 220 ads put straight into the main campaigns since June, half never reached £100 of spend. The rest split roughly evenly between good, middling and poor.
Each square is 1% of the 220 ads. Cost per sale is Meta-reported, 7-day click and 1-day view.
Probably not. We found eight ads that Meta starved in one ad set and that later got £300 or more somewhere else. They came in at £45 a sale, the same as ads that were funded everywhere. An ignored ad is an unread hand, not a losing one.
Cost per sale in main. Only eight examples, so this leans one way rather than proves it.
Only when each test ad gets enough budget. The October Christmas tests gave each ad a real budget, and 19 of 22 ads got £100 or more. The summer tests spread a small budget across many ads, and only 15 of 147 did, worse than putting them straight in.
Share of new ads that reached £100 of spend where they started, June to October. A test only improves the odds if every ad in it can actually get seen.
It seems to. New ads on a post that already had 50 or more reactions and comments reached £100 of spend 86% of the time, against 32% on a brand-new post. Part of that is because we only reuse posts from ads that already worked. Strong existing ads in an ad set did not crowd new ones out.
Share of new ads that reached £100 of spend within about three weeks.
No. Across 24 batches, from a single ad to 44, cost per sale two weeks later moved by the same amount as in two-week spells when nothing was added. The day-to-day ups and downs were the same too.
Median change in the main campaign's cost per sale, two weeks after vs two weeks before. Detail: stability page.
No clear difference. One or two ads did not land any more gently than 20 or more.
Median change in cost per sale after the batch. Each group has only 4 to 10 batches.
Mostly straight in. Of the 20 biggest main-campaign ads since June, 18 went straight in and 2 came through testing. But that is mostly because we hardly moved anything: of 111 creatives tested between June and September, only 7 were ever moved into main.
Main-campaign spend since June. Ad 551 combines its two ad sets.
We have never run it properly. The summer tests were too thin to see most ads, and the few winners were rarely moved. October is the first time we have funded tests properly and moved the winners, so the first real answer comes from the Christmas moves over the next two weeks.
The fair measure: ads that reach £1,000 of spend at or under break-even, per 10 ads made, counting test spend as well. Compare the two routes on the same kind of creative.
Two clean examples, one better and one slightly worse. Not enough to decide.
Main campaign cost per sale, two weeks before vs two weeks after the move.
Not checked yet. This came up on our call: a quote box or pottery-story ad is a known type, so why test it? It needs every ad tagged by concept, then a check of whether proven types get funded more often straight in.
Where things stand
| What | So far | Next read |
|---|---|---|
| Icicle 1066 and 1067, moved into the Christmas ad set 26 Oct | £260 for 7 sales (£37) and £537 for 12 (£45). Both getting spend. | 5 Nov |
| Five Fir winners, moved 27 Oct | Only 1029 has had real spend (£126 for 2). The rest under £15 each, because the ad set was down to £700 a day. | 5 Nov |
| Christmas ad set budget | Raised from about £700 to about £1,300 a day on 29 Oct by lowering the other ad sets' minimums. | daily |
| Flora 1063, in Gifting since 27 Oct | £630 for 12 sales (£52). | 5 Nov |
| Flora 1056, 1064, 1065, into the Christmas ad set 29 Oct on their original posts | 1064 sold at £27 and 1056 at £35 in testing. 1065 had only £5. Test 3 kept on at about £150 a day as a backup. | spend check 2 Nov, read 5 Nov |
Read each ad on its own cost per sale and the ad set's over the same days. November break-even is about £43 a sale.
How we improve our odds
What we still need to find out