Ad Refresh Watch
Every morning this watch looks at each Pott ad that is taking real money and asks one question: is it starting to wear out? When it is, the answer is never to switch it off. It is to have the next version ready and live beside it before the results slip.
The one idea. A good ad does not fail suddenly. First the leading signs move (fewer people clicking, fewer new people seeing it, the cost of each sale creeping up), and only a few weeks later the return falls. The watch reads those early signs so that a refreshed version of the same winning idea can be briefed, shot and launched in time. Winning ads keep running; they hand over to their refresh when it has proved itself.
1. What we watch
Nine signals, read for every ad that is taking real money (at least £300 in the last seven days, or 3% of the account). Each one is judged on the last 7 days against the 28 days before, and against the account's own movement over the same days, so a Christmas auction getting dearer for everyone does not read as one ad tiring. The signals sit in three families, in the order they tend to move.
Attention: are people still stopping for it?
- Click rate
- Clicks to the site per view. Judged against the account's own click rate move.
- Hook rate (video)
- Share of people who watch the first three seconds. Judged against the ad's own past, because it depends on the format.
- Hold rate (video)
- Share of three-second viewers who stay to the end. Tells us the opening still earns the rest.
Audience: is it still finding new people?
- Ad set frequency
- Average times each person has seen the ad set this week. Prospecting ceiling 2.5 to 3; retargeting 5 to 8.
- Cost per thousand views, against the account
- Rising for this ad while the account's holds means Meta is finding it harder to place.
- New people reached per pound
- How many people saw the ad for the first time this week, per £1,000. Falling sharply means the audience is wearing thin.
Return: is it still paying?
- Cost per sale
- Spend divided by purchases, with a chance test: a rise only counts when it is unlikely to be a few quiet days (purchases arrive in lumps). A higher cost per sale with a higher return means bigger baskets, not fatigue.
- Return against its own best week
- Where this week sits against the ad's best seven days. Catches slow decline that no single week shows.
- Purchases
- Sales this week against the 28-day weekly average, same chance test.
A signal is moderate or strong depending on how far it has moved (for example click rate down 10% is moderate, down 20% is strong). One moderate signal on its own never does anything: it is noted and rechecked. Signals are counted by family, not individually, because click rate and hook rate tend to move together.
2. What triggers each stage
Each ad sits on one rung. The rung is an instruction about the next version of the ad, not about switching the current one off. The further down the list, the more urgent.
Pausing is the last resort, and it needs two things at once: the original has sat below breakeven (2.46x) for two full weeks, and a refresh is live and proving itself. Then the instruction is to shift budget to the refresh, which Meta mostly does on its own. No ad is ever cut on one bad week, and no stage on this page ever says "pause".
A stage is sticky for seven days: an ad keeps the highest rung it earned in the last week, so a quiet Tuesday does not make a refresh vanish from the plan. If more than half of all judged ads move together on the same day (a site problem, a tracking wobble, an auction shift), the run is labelled account-wide and only a fresh "Launch" is held back, because that is not fatigue.
3. Where the wear is
Before blaming an ad, the watch asks whether the movement belongs to something bigger. The rule is the same at each level: when more than half of the ads taking real money inside a container move together, it is the container, not any one ad.
| Level | What it looks like | What it usually means | Pott today |
|---|---|---|---|
| Account | Most ads dip at once; click rates hold | The site, tracking or the auction. Check the site before touching creative. | No shared move across the 13 judged ads |
| Campaign | Most ads in one campaign dip; other campaigns hold | Budget pushed too hard, the audience exhausted, or the whole set of ideas tired. Refresh the campaign's creative set, not one ad. | None moving together (see below) |
| Ad set | Frequency climbing and new people per pound falling on every ad in it | The audience is running out. The fix is a broader audience or a new ad set, not new creative. | None moving together (see below) |
| Ad | One ad's early signs move while its neighbours hold | Creative wear. The ladder above. | 1 ad needing a refresh now |
By campaign
| Campaign | Ads judged | 7-day spend | In Build or above | Read |
|---|---|---|---|---|
| Evergreen (old Tomato vine) | 10 | £13,087 | 50% | No shared move across the 10 ads. Anything flagged here is about the individual ad. |
| Gifting Christmas 2026 | 3 | £1,413 | 0% | No shared move across the 3 ads. Anything flagged here is about the individual ad. |
By ad set
| Ad set | Ads judged | 7-day spend | In Build or above | Read |
|---|---|---|---|---|
| Christmas Evergreen (old Tomato vine) | 5 | £9,836 | 66% | No shared move across the 5 ads. Anything flagged here is about the individual ad. |
| Everything Else Evergreen (old Tomato vine) | 2 | £1,861 | 0% | 2 ads taking real money here, too few to read the container; judged ad by ad. |
| Gifting Gifting Christmas 2026 | 3 | £1,413 | 0% | No shared move across the 3 ads. Anything flagged here is about the individual ad. |
| Tomato Evergreen (old Tomato vine) | 2 | £1,013 | 0% | 2 ads taking real money here, too few to read the container; judged ad by ad. |
| Seasonal Evergreen (old Tomato vine) | 1 | £377 | 0% | 1 ad taking real money here, too few to read the container; judged ad by ad. |
"In Build or above" is the share of the container's judged spend sitting on ads at Build, Launch or Hand over. Ad sets with fewer than three judged ads cannot be read as a container, so their ads are judged individually. The 13 ads under seven days old (the Christmas creative tests and the Fir A/B tests among them) are not in these counts yet.
4. The refresh calendar
One row per ad that needs a next version, worked back from today with Pott's production lead time of 14 days from brief to live. The wear-out window is a band for now (set by the stage); per-ad estimates from Pott's own history of how long winners last come next.
Why this stage, in numbers
Timing: late: results are already moving. Dates assume 14 days from brief to live. See the ad.
Watch: 4 ads with something to keep an eye on
| Ad | Ad set | 7-day spend | Return | What is moving | |
|---|---|---|---|---|---|
 | Ad 983 Icicle: New Glaze Fresh From The Kiln | Christmas | £1,343 5% | 1.4x was 2.08x | cost per sale rising (strong); well off its best week (strong) |
 | Ad 940 Tomato Vine: Maker So Happy | Tomato | £396 1% | 0.62x was 1.31x | fewer new people per pound (moderate); cost per sale rising (moderate); well off its best week (strong); Meta is already moving spend off it |
 | Ad 837 Fig: Scent Led | Seasonal | £377 1% | 2.73x was 2.01x | click rate falling (moderate) |
 | Ad 1010 Christmas Collection: Favourite Gift Last Christmas | Gifting | £303 1% | 1.59x was 1.17x | click rate falling (moderate); Meta is already moving spend off it |
Nothing to do: 8 ads clean or improving
| Ad | Ad set | 7-day spend | Return | Age | |
|---|---|---|---|---|---|
 | Ad 922 Seasalt Starter Pack: Buy Once Refill Forever | Everything Else | £1,423 5% | 1.67x was 1.36x | 28 days old, Meta already moving spend off it |
 | Ad 973 Fir: Smells Like A Christmas Tree | Christmas | £814 3% | 3.84x was 2.11x | 16 days old |
 | Ad 934 Tomato Vine: Customer Quote Whole Hall | Tomato | £617 2% | 1.44x was 1.26x | 29 days old |
 | Ad 995 Icicle: Meet Icicle Limited Edition | Christmas | £599 2% | 2.55x was 1.4x | 16 days old |
 | Ad 1008 Starter Pack: Gifted Last Christmas Still Using It | Gifting | £575 2% | 1.92x was 1.04x | 16 days old |
 | Ad 970 Christmas Collection: Christmas Scents Are Here | Christmas | £567 2% | 1.31x was 0.79x | 16 days old |
 | Ad 1006 Starter Pack: Bought It For My Girlfriend | Gifting | £534 2% | 2.36x was 1.46x | 16 days old |
 | Ad 964 Seasalt: Standing By The Sea | Everything Else | £438 2% | 1.78x was 1.6x | 19 days old |
Too new to judge: 13 ads under seven days of history
These are judged from their eighth day. A new ad that becomes a winner can still reach Brief straight away.
| Ad | Ad set | 7-day spend | Return | Age | |
|---|---|---|---|---|---|
 | Ad 1020 Cook Candle: One Candle For Your Kitchen | Ad Set 1 - MAIN | £968 3% | 0.99x was 2.45x | 11 days old |
 | Ad 1025 Tomato Vine: Only Tomato Candle You Need | Tomato | £714 3% | 2.31x was 2.47x | 11 days old |
 | Ad 1063 Starter Pack: T3 SP 02 Gift They Keep Enjoying | Test 3 Starter Pack Scent | £651 2% | 0.82x | 2 days old |
 | Ad 1011 Christmas Collection: Gifted To Lots Of People | Gifting | £644 2% | 1.25x | 16 days old |
| Ad 1035 Holly: Customer Quote Tree Without Needle Drop | Christmas | £459 2% | 1.43x | 6 days old |
 | Ad 1033 Fir Starter Pack: Pine Disinfectant Real Christmas Tree | Fir A/B - Videos | £411 1% | 2.53x | 6 days old |
 | Ad 975 Icicle Starter Pack: Starlight Starter Pack Unboxing | Christmas | £385 1% | 1.13x | 16 days old |
 | Ad 1067 Icicle: T4 Icicle 02 Wintery Not Too Christmassy | Test 4 Icicle Proof | £373 1% | 3.65x | 2 days old |
 | Ad 1028 Fir Grand: Winning Quote New Grand Image | Fir A/B - Statics | £342 1% | 1.15x | 6 days old |
 | Ad 1027 Fir Starter Pack: Short Review Smells Like A Fir Tree | Fir A/B - Statics | £337 1% | 2.1x | 6 days old |
 | Ad 1029 Fir: True To Life Review | Fir A/B - Statics | £337 1% | 2.5x | 6 days old |
 | Ad 1030 Fir: Not Fake Review | Fir A/B - Statics | £335 1% | 1.52x | 6 days old |
 | Ad 1022 Tomato Vine: Best Reason Refill Demo | Tomato | £305 1% | 1.47x was 0.56x | 11 days old |
5. How to use it
Every day
- The watch runs each morning after Meta settles yesterday's numbers.
- Only a change of stage reaches Casey's morning email. A quiet day means nothing to do.
- This page is rebuilt from the same run.
Every Monday
- Casey and Ailis read the calendar above together.
- Agree which refreshes to shoot this week: anything with a brief-by date this week, Launch rows first.
- Ailis's filming day covers the brief-by rows; anything that needs a creator or a customer clip is requested the same day.
Each refresh
- Same idea and same offer as the original. Change the opening frame, the hook, the first line, the thumbnail or the format.
- Launch it in the same ad set, next to the original. Never pause the original cold.
- When the refresh holds 90% of the original's return for a week, it takes over and the original winds down.
What it will not do
- It never pauses an ad or changes a budget. It only reads.
- It never asks for separate ad sets by audience or funnel stage: Meta decides who sees what, and does it better with a wider range of creative to choose from.
- It does not judge new ads for their first week, so a launch week's odd numbers never trigger anything.
6. How the numbers are judged
- Windows. Recent = the last 7 days to yesterday (17 Oct to 23 Oct). Baseline = the 28 days before (19 Sept to 16 Oct). Best week = the ad's best rolling 7 days since 22 Aug, excluding Black Friday week.
- Who is judged. Ads with £300+ in the last 7 days or 3% of the account's spend, still switched on. An ad whose daily spend Meta has already halved is "pulled back" and watched, not alerted.
- Breakeven. 2.46x, Pott's first-order new-customer breakeven on the 41 to 42% contribution margin. Only an ad below this line can reach the top rung on return alone.
- Net of the account. Click rate, CPM and return are judged against the account's own move over the same days, and never read as worse than the ad's own move.
- Chance tests. Cost per sale and purchase falls must be unlikely to be chance (under a 20% probability for moderate, 10% for strong, on the number of sales expected), and the week must have spent enough to buy at least three sales at the old cost.
- Sticky stages and the account-wide guard are described under section 2. Stage thresholds live in
scripts/meta-fatigue-watch/shared.jsonand Pott's own settings inclients/pott-candles/meta-fatigue-watch/config.json.