Judge Creative on Sparse Conversions: Honest Proxies
Most new creative dies before it ever earns a verdict. You launch four or five concepts, each gets a handful of purchases in the first few days, and the numbers are too thin to tell a real winner from variance. So operators do the expensive thing: they kill on noise, or they keep funding losers on hope. There’s a better discipline — read leading indicators up the funnel, but never let a proxy impersonate proven profit.
For the surrounding account decisions, compare Modeled Conversions: Meta’s Consent-Gap Fill, Kept Honest and use Iteration vs Net-New Swings: A Disciplined Creative Pipeline as the next diagnostic.
Why sparse conversions break naive testing
Purchases are the rarest event in your funnel. A creative that drives a thousand impressions might produce a few clicks and one or two checkouts. At that volume, the difference between a 1.8 ROAS and a 3.2 ROAS can be a single extra order that happened to land on a Tuesday. You are not measuring the creative; you are measuring luck.
Creative testing at low conversion volume is the default state for many accounts, not an edge case. It gets worse when:
- Average order value is high and purchase frequency is low — fewer conversion events per unit of spend.
- You’re testing many concepts at once — each one starves for signal.
- The optimization event is deep (purchase, not add-to-cart) — Meta needs enough recent optimization-event signal to exit the learning phase, and so do you.
Two mistakes follow. The false negative — killing a creative that would have worked, because three quiet days looked like failure. And the false positive — crowning a “winner” that simply caught a good roll of the dice and regresses the moment you scale it.
The proxy ladder
The fix is to stop staring at the rarest event and read the events that accumulate faster. Every step a user takes is a rung on a ladder, and each rung trades correlation for volume:
| Rung | Signal | Volume | Tie to profit |
|---|---|---|---|
| Hold / thumbstop rate | Does the hook stop the scroll? | Very high | Weak |
| Outbound CTR | Does it compel a click? | High | Moderate |
| Cost per add-to-cart | Does it create intent? | Moderate | Strong |
| CPA / contribution ROAS | Does it actually pay? | Sparse | Truth |
The principle: read the lowest rung you have enough volume to trust. The higher you climb, the faster you get a stable number — and the further you drift from money. A great hook with a dead CTR tells you the creative is candy with no payload. A strong CTR with no add-to-carts tells you the click was a promise the landing page couldn’t keep.
Rung 1 — Hold rate (the hook)
Three-second/thumbstop views over impressions. This fires in the hundreds within hours, so it stabilizes first. Use it for one job only: triage the hook. If hold rate is in the floor of your account’s range, the creative never got a fair hearing — nobody watched long enough to be sold. That’s a hook problem, not a proof of unprofitability. Don’t kill the concept; recut the first two seconds.
Rung 2 — Outbound CTR
Outbound clicks over impressions is your first real intent signal. It accumulates fast enough to compare creatives within a day or two of meaningful delivery. A healthy CTR means the creative earned attention and a click — that’s a creative doing its job at the top. But CTR is where Goodhart’s law starts to bite: optimize hard for clicks and you breed clickbait that converts worse downstream. Read CTR as a gate, not a goal.
Rung 3 — Add-to-cart (the honest workhorse)
Cost per add-to-cart, and the click-to-ATC rate, is the most useful proxy in the ladder. It sits one step from the wallet, so it correlates far better with purchases than CTR — yet it generates several times the volume of completed orders, so it reaches a readable sample days earlier. When conversions are too sparse to rank creatives, rank them by cost per add-to-cart instead, then let the purchase data confirm the order over a longer window.
The caveat keeps you honest: add-to-cart can inflate. Cheap impulse adds from a discount-led hook will pad the metric while the same users abandon at checkout. So always read ATC alongside the ATC-to-purchase ratio. If adds are cheap but checkouts don’t follow, the gap is downstream — price, shipping cost, or checkout friction — and no amount of creative iteration fixes it.
Reading divergence between proxy and reality
The ladder earns its keep when rungs disagree. The pattern of the disagreement is the diagnosis:
- High CTR, low add-to-cart: the ad over-promises or targets the wrong intent. The click was curiosity, not desire. Tighten the message-to-offer match.
- High add-to-cart, low purchase: the creative did its job; the post-click experience is leaking. Audit price presentation, shipping reveal, and checkout speed — not the ad.
- Strong hold, weak CTR: entertaining hook, no reason to act. Add a clearer value proposition or a sharper call to action.
- Every proxy healthy, conversions still zero after real spend: keep funding it to reach decision volume. Healthy proxies are exactly the evidence that says “this deserves more data,” which helps you avoid errors from the false-negative kill.
That last move is the whole point. Proxies are best used to decide what not to kill and where to send more budget to manufacture conversion data — not to declare victory.
A workflow that survives thin data
- Set a floor before you read anything. Pick a minimum spend or impression threshold per creative and refuse to judge below it. This single rule prevents most premature kills. Respect the learning phase — a creative still gathering its first optimization events isn’t a verdict, it’s a work in progress.
- Read top-down, gating. Hold rate, then CTR, then cost per add-to-cart. A clear failure on an upper rung is permission to cut early on a diagnosable reason. Healthy upper rungs with no conversions yet is permission to wait.
- Rank survivors by the lowest rung you can trust — in many cases cost per add-to-cart — and concentrate budget there to accelerate real purchase data.
- Confirm with money, over a longer window. A creative is only a winner when contribution after product cost, shipping, and fees holds up — judged on contribution margin and account-level MER, not platform-reported ROAS on a three-day flicker. Proxies nominate; profit confirms.
- Rotate your proxy targets. The instant a metric becomes the goal, it stops measuring quality. Reward hooks and you get clickbait; reward adds and you get abandoned carts. Keep the proxy a thermometer, not the thesis.
This is also where an always-on operator helps. Bach watches the full ladder per creative, flags when a proxy and the purchase data diverge, and surfaces the diagnosis — while staying read-only until you approve any change. The judgment about what’s a true winner stays yours; the proxy bookkeeping doesn’t have to.
The takeaway
When conversions are too sparse to rank creatives, climb the ladder — hold rate to triage the hook, CTR to confirm intent, cost per add-to-cart as your honest workhorse. Use the lowest rung you have the volume to trust, and read disagreement between rungs as a free diagnosis. But hold the line on the one rule that keeps this honest: a proxy is a leading indicator, never a profit certificate. Let proxies decide what to keep alive and where to spend; let contribution margin decide what actually won.