Dated creative fatigue, not guessed.
This ad stopped working on August 10. You killed it on the 22nd.
Creative decays. Everyone knows that. Almost nobody can name the day a specific asset crossed from working to not working, because the signal sits under budget shifts, audience changes, and seasonality. Creative puts a date on it.
Thirteen real series, one asset that fatigues, detection computed live in your browser.
The problem
The platform is not going to tell you an asset is done.
Delivery reporting is optimised to keep spend flowing. It shows you yesterday against last week, per campaign, blended across every asset in the ad set. A single creative going dead inside a healthy account is a rounding error in the blend, so the number you check weekly never turns. What actually happens is that assets run for weeks past their useful life, or get killed on one bad Tuesday because a manager was looking.
The insight
Fatigue shows up in engagement first. Kill rules watch conversion.
Frequency-driven fatigue is an attention problem before it is a money problem. The same people see the asset again, they stop stopping, and engagement rate and click-through break first. Conversion rate and cost per acquisition only move once the smaller, worse-qualified click stream works through the funnel, which takes days. Every kill rule in production is written on conversion or CPA, so every kill rule is structurally late by exactly that lag, and the spend that goes out during it is unrecoverable. You cannot just threshold the engagement metric instead: it is noisier than conversion, it moves with placement mix and dayparting, and a per-metric threshold across forty assets sends a kill recommendation every day. What you want is the changepoint, dated, with the false alarms already removed.
CUSUM on day-over-day percentage changes for sustained decay, Bayesian Online Changepoint Detection (Adams & MacKay 2007) for abrupt breaks, and Benjamini-Hochberg FDR control across every asset × metric pair in the account, because an account with 40 live assets and 6 metrics is 240 simultaneous tests.
How it works
Four steps, no data science team
Meta, Google, TikTok, or a warehouse table if your agency already centralises it. Asset-level daily metrics, which is the grain the APIs already return.
Engagement rate, thumb-stop rate, click-through, hook rate and frequency on one side. Conversion rate, cost per acquisition, ROAS on the other. Both get tested, but the leading side is what dates the decay.
Each series runs CUSUM and BOCD. Whatever crosses is filtered by Benjamini-Hochberg across the whole account, so what survives survives knowing how many assets you tested.
The date the asset broke, the metric that broke first, what the blended account number did over the same window, and the spend that has gone out since. You still make the call.
Who it is for
The person who owns paid social performance
Performance marketing teams and the agencies that run accounts for them. The shape that hurts most is $200k to $5M a month of paid social across a rotating library of 20 to 100 creative assets.
Pricing
- –Full detection engine
- –90-day history
- –Weekly digest
- –Meta and Google
- –Unlimited assets
- –Slack and email alerts
- –Leading vs lagging split per asset
- –Spend-since-break accounting
- –TikTok and warehouse sources
- –Per-client workspaces
- –White-labelled client reporting
- –API and webhook
- –SSO and audit log
Competition
What exists, and what it does not do
| Who | What they do | The gap |
|---|---|---|
| Meta Ads Manager | Native delivery reporting, frequency, and a creative fatigue hint inside the ad set. | Reports the blend and is built to keep spend flowing. No dated changepoint, no correction for the number of assets you are watching, and nothing that spans platforms. |
| Motion / Atria | Creative analytics dashboards that roll asset-level metrics up by tag and concept. | Better reporting on the same data. You still have to eyeball a chart and decide the line bent, and you still decide on conversion metrics. |
| Triple Whale / Northbeam | Attribution and blended MER across channels. | Answers where the money worked, at account and channel grain. Attribution is not decay detection and it does not run per asset. |
| In-house rules in a spreadsheet | Kill if CPA is above target for three straight days. | Late by construction, because CPA is the lagging metric, and noisy by construction, because three days of a threshold on a noisy series is not a test. |
The honest failure mode: the platforms own this data and could ship a dated fatigue verdict natively, for free, inside the surface where the decision is already made. There is also a buyer problem, because an agency billing a percentage of managed spend is not obviously eager to kill assets faster. The defence is to be cross-platform and to sell to the advertiser rather than the agency, but if Meta ships it well the standalone version of this is a feature.
Market
Priced off the creative production budget, not the media budget
A brand spending $1M a month on paid social already pays a creative studio $20k to $60k a month to feed it. Six hundred dollars to know which assets are dead sits inside that line, not the media line. Five thousand accounts at the Team tier is $36M ARR, and the agency tier is the same customers bought once instead of ten times.