The Challenge
The client came to us wanting the usual: find the right audience, the right creative, the right offer. Before working with us, they were getting 1–2 sales a month through the website — driven by PR and celebrity mentions, not paid marketing. None of that was tracked or attributed, which is its own optimization opportunity for later (we weren't brought in for that piece, though we could help there too).
Two weeks into the engagement, we paused and asked a different question: can we actually answer "find the right audience" with the data we have?
At this price point, the math doesn't work the way it does for mainstream ecom. A $50 product doing 500 orders a month gives you a real signal within weeks. A $5K product doing 2–5 purchases a month doesn't — one extra sale can double ROAS overnight, one lost sale can crater it. Most "this audience performed better" conclusions at this volume aren't insights. They're noise wearing an insight's costume.
The part that stung a little: three months into working with us, the client was still getting 1–3 sales a month. To them, that looked like nothing had moved — no visible trend, no clear before/after. We'd said upfront that anything meaningful needed at least 6 months, and that the first stretch would mostly be testing hypotheses, not harvesting results. We knew that going in. Doesn't make the conversation any less uncomfortable when a client is staring at the same number three months later, asking what they're paying for. That part is on us to manage better next time.
The Approach
We stopped waiting on Purchase and moved the primary read earlier in the funnel.
With 1–3 purchases a month, Purchase isn't a metric — it's a coin flip. Add-to-cart on cold traffic is far more common per user and shows up fast enough to actually read.
We also stopped changing five variables at once. Audience, creative, landing page, offer, and seasonality had all been shifting in the same short windows — when something didn't convert, the instinct was to blame the audience, but the setup never isolated audience as the variable in the first place.
And we stopped asking why Meta "couldn't find buyers." The algorithm optimizes on Purchase events. With 2–5 purchases a month, it simply doesn't have enough signal to learn from, no matter how good the targeting is. The real question was how to feed it enough signal to learn from at all.
What we changed:
- Stopped evaluating campaigns on a 2-week / 3-purchase sample. Moved the primary read to ATC rate on cold traffic — enough volume to be a real signal within days, not months.
- Separated what marketing could actually be responsible for (qualified traffic, retargeting depth, richer pixel signal) from what belonged to brand and pricing (trust, awareness, the actual decision to spend $5K on jewelry from a brand they'd never heard of).
- Ran fewer, longer tests instead of frequent pivots — every pivot resets the algorithm's learning clock.
- Got comfortable saying "we don't know yet" instead of manufacturing an answer to justify the spend.
The Results
For a typical ecom brand with a $100–200 AOV, a healthy cold-audience ATC rate is north of 5%. Given the price point here, we set a lower, more realistic target going in — 2.5%.
The best-performing audience segments we found topped out at 0.6%. The account average sat at 0.15–0.20%.
Purchases sat at 1–3 a month before us and after us. Read on its own, that looks like nothing happened. Read next to a 0.15–0.20% cold-traffic ATC rate against a 2.5% target, it reads differently: the real constraint wasn't audience luck — it was top-of-funnel intent at this price point. That's a slower, structurally different problem than "test another interest group," and it's the number that told us the truth the purchase count couldn't.
The Takeaway
If you're getting single-digit purchases a month, most of the "best practices" written for $50 DTC products won't transfer directly. The volume isn't there to support the same kind of decision-making — but a leading indicator further up the funnel usually is.
The biggest mistake we see in high-ticket ecom — and the one we had to unlearn ourselves — is asking for answers before the data can honestly give you one.
