Pinterest Ads for a small store: test it, or judge it only on assisted conversions?

Pinterest keeps coming up in ecommerce threads, and I think it's because people there are planning, not scrolling to buy. Someone pins a sofa today and purchases in three weeks, so the first click often looks like a loss in last-click ROAS.

My guess is it only pays off for stores with visual, seasonal products like home decor, apparel or wedding goods, and that a small store should judge it on assisted conversions in GA4 instead of last-click. What I'm less sure about is whether a 30-day window captures that lag properly, and whether catalog-based Shopping ads beat simple promoted pins at small budgets. Has anyone run a test long enough to actually see the delay?

Before you judge the channel, check the window. Look at how long Pinterest-touched buyers take to go from first click to purchase. If a real share of them convert after day 14 or day 30, a 30-day lookback is undercounting exactly the planners you're describing. GA4's attribution settings let you extend the conversion lookback to 90 days, so use that for the analysis.

For the test itself, run it for at least twice your typical delay. If most buyers take three weeks, plan on six to eight. Assisted conversions are useful, but they tend to flatter any channel that sits early in the path, so don't lean on them alone. A pause test works better: switch pins off for a week or two and see whether revenue from new visitors drops. I'm honestly not sure whether catalog ads beat plain promoted pins at small budgets, so test that separately rather than assume it.

The pause test makes sense, but it fails slowly. Someone who pinned your sofa in week one often comes back in week four through branded search or a direct visit, so when you turn pins off the drop is lagged and can look like noise for the first week or two. Watch branded search impressions in Search Console next to new-user revenue in GA4, and don't call the test after a short pause. It probably needs to run longer than the delay you measured.

Branded search impressions are a shaky signal here. Anyone who sees a pin and later searches your brand is exactly the planner we're trying to measure, so branded search is partly an effect of the pins themselves. That means the metric drops when pins go off even if they were doing little, and it rises from anything else that gets people talking about you. On a small store, brand query volume is noisy enough that a week of movement can mean almost nothing.

I'd also skip a plain pause on a seasonal catalog. Switching off mid-season costs real sales and still tells you little about the channel. A holdout is cleaner. Either exclude a random slice of your audience from Pinterest, or alternate on and off in two-week blocks and compare the gaps. You still have to wait out the delay, but you're comparing groups exposed to different things at the same time, not this week against last week.

I'm not sure two-week blocks hold up here. With a lag of around three weeks, a purchase in an "off" block often traces back to a pin clicked during the "on" block, so the gap you're comparing is already mixed with the delay you're trying to measure. Blocks only work if they're at least as long as the delay, or if you group orders by the date of the first pin click instead of the purchase date. A random holdout is still the cleaner design, if your setup lets you split the audience.

First-click grouping only works if the pin click actually gets recorded, and planners are exactly the people who break that. Someone pins on their phone, reads reviews on the laptop, and buys a month later from a direct visit. GA4 sees a direct session with no pin attached, so the order lands in the wrong block.

For a small store I'd trust the holdout more, even though it's harder to set up, because it doesn't depend on catching the first touch.

On a small store, a random audience split often falls over because the holdout arm might only produce a handful of orders a month, so you can't tell a real gap from noise. A geo holdout tends to work better at small budgets. Pick two or three regions with similar sales history, pause pins in one group, and keep them running in the others. Then compare new-user revenue in each region against its own pre-test baseline. You lose some precision because regions aren't identical, but you get enough orders to actually read the result. It also gets around the cross-device problem you raised, since the split is by location rather than by a click that might never get recorded.

How many orders does the store do in a typical month?

Geo holdouts are the right call at small volume, but pick your regions carefully. Pinterest audiences travel, and someone in one city is often buying a gift for someone in another. So either choose regions with a real geographic gap between them, or accept that part of the holdout region will still get exposed, and say that when you read the numbers.

Spillover is a real issue, but the baseline worries me more. A geo holdout only reads cleanly if the paired regions moved together before the test, and with seasonal decor or apparel that's rarely true for long. A region that spiked last November will look like a winner or a loser for reasons that have nothing to do with pins. I'd pull at least a full year of weekly new-user revenue per region and check whether each pair tracks the other week to week, not just whether the totals look similar. If a pair drifts apart, drop it before you spend two months waiting on it.

Also, pins clicked just before the test starts will keep producing orders for weeks, so the first stretch of the test is partly measuring the old setup. Does the store ship nationwide, or is it mostly local? That decides whether gift spillover matters much at all.

Raw week-to-week correlation will flatter any pair of regions that share a seasonal shape, because both spike in November and look matched. Compare each week to the same week last year, or detrend the series first, and see whether the pair still moves together.