Meta retargeting window: 7 days or 30 days for a high-ticket store?

Take a furniture store selling pieces at a few hundred dollars each, where someone wants a 30-day retargeting window because the buying cycle is long. I'm not convinced. A window that wide fills up with people who already bought, or who looked at a different chair and forgot about you, so you keep paying to re-hit them while frequency climbs. A 7-day pool is smaller but usually warmer. If the long research phase really matters, run a separate 30-day pool on a small budget and judge it on its own numbers. How would you split the budget between the two, and how do you tell whether the wider pool adds sales or just takes credit for purchases that were coming anyway?

I don't think the 7-day pool is usually warmer for something like a chair. Someone who clicked yesterday on a few-hundred-dollar item is often just browsing, and the real decision tends to come two or three weeks later, after a partner weighs in or they compare a second store. A 7-day window misses that stretch entirely.

The already-bought problem is a separate issue from window length. Exclude purchasers with a customer list or a purchase event and most of it goes away. Frequency is the one fair worry, so put a cap on the 30-day ad set and keep an eye on it there.

Purchaser exclusion only works if the purchase event fires cleanly and the customer list is current, and with furniture that's often not the case. A lot of purchases happen offline or on a second visit, so the list lags behind reality. Frequency caps help, but they won't stop someone who already bought from seeing the ad for three weeks.

A lagging purchaser list isn't a good reason to drop the 30-day pool. A stale list means the signal is broken, and that's fixable. If sales happen in store or on a second visit, upload offline conversions matched on email or phone so Meta sees who actually bought, and the exclusion catches up within days.

You're right that caps won't stop a past buyer from seeing the ad. But at two or three impressions a week, the wasted spend on a few hundred dollars' worth of audience is usually small. The better test is a holdout slice of the 30-day pool.

Agreed on the holdout, and I'd make it a hard rule: keep the 30-day pool only if its incremental cost per sale, measured against the holdout, comes in below the 7-day pool's CPA. Don't trust the platform-reported sales here. A 30-day window mostly picks up buyers who would have converted anyway, so those numbers will look better than they really are.

Agreed, and I'd sort out the comparison before you even get to the holdout. The 30-day pool gets a longer attribution window, so if you pull CPA from the same report for both pools, you're comparing two different things. Put both on the same click window and the same offline conversion upload, then read the holdout against that.

Meta's Conversion Lift study randomizes at the user level, which is cleaner than a hand-built holdout ad set that leaks into the other audiences. And if you're only getting a few sales a week, set a minimum conversion count up front so nobody calls a winner off a handful of orders.

Conversion Lift is the right tool, but the measurement window matters a lot here. It only counts conversions that land inside the test period, so with a two or three week buying cycle like Sasha described, a short read will undercount the 30-day pool and make it look weaker than it is. Run the test long enough to cover the full cycle, and compare the 7-day pool over that same stretch. At a few sales a week it also takes a long run to reach significance, so let the minimum conversion count decide the end date instead of picking a calendar date up front.

Agreed, and there's one thing the test-window point misses: offline upload lag. With furniture, a lot of the sales that count happen in store and reach the list days or weeks later, once someone matches an order to an email or phone. If the test closes on the last day of the buying cycle, those late uploads never make it into the read, and the 30-day pool looks weaker for a reason that has nothing to do with the ads.

So before setting the end date, pull the gap between in-store sale date and upload date for a sample of recent orders. Add that gap to the two or three week cycle and treat the total as your minimum runtime. After that, let the conversion count decide when to stop, but don't stop before the uploads have caught up.

Base it on the 90th percentile of the lag, not the average. The average hides the slow tail, and those are the orders that get matched weeks later, so a runtime set from the mean will still cut off exactly the sales a 30-day pool is supposed to catch.