Meta CAPI for a small budget: is deduplication worth the setup?

Setting up Conversions API through a partner integration or a server container feels like a lot of work at around 1,500 a month in spend. My view is that for a small account the pixel alone gets you most of the way, as long as your purchase event fires cleanly on the thank-you page and you aren't double counting. Plenty of people argue the opposite: browser-only tracking undercounts iOS purchases badly enough that Meta's optimization starves and your CPA looks worse than it really is. Both camps usually skip the same check, though. If you run both, do the event IDs actually match, and did reported conversions drop once you deduplicated? That's the number I'd want to see before deciding.

Your check catches double counting, but it can't tell you whether the pixel is undercounting, and at 1,500 a month that's the bigger question. A drop after deduplication only shows what was duplicated. If iOS purchases are going missing, the reported number can look stable either way.

What I'd run is backend orders against Meta purchases over the same week, with the same attribution window. Say a store logs 200 orders and Meta reports 150 purchases with CAPI off. That 25% gap is what the algorithm is learning from. My rough rule: if the gap stays under about 10 to 15%, the pixel is probably good enough at this budget and CAPI is mostly setup time. Above that, the setup usually pays off, because Meta is optimizing on a thinner signal than the dashboard suggests.

Just make sure both sides count the same way, or the gap is meaningless.

I'm not sure the 10 to 15% threshold is the right thing to anchor on. At 1,500 a month, the size of the gap matters less than whether the missing conversions are random or skewed. If it's mostly iOS buyers dropping out, the algorithm is learning from a biased sample, even when the total gap looks small.

Easy enough to run. Pull backend orders by device or OS for the same week, then compare the iOS share of orders with the iOS share of purchases in Meta's breakdown view. If iOS is 40% of your orders but only 25% of reported purchases, the gap is skewed, and a 12% total gap would hide that completely. Just keep the date range and attribution window the same on both sides, or the split won't tell you much.

The device split is the right test, but filter your backend numbers first. Total orders include direct and organic buyers who never touched a Meta ad, so the gap will look bigger than the pixel could ever capture. Compare against orders from paid Meta sessions, or at least drop direct traffic, before you read the iOS share.

My rough rule: if iOS is under-represented by more than about 5 points in Meta's purchases against that filtered backend number, CAPI is probably worth setting up even at this budget. Anything smaller could just be noise from a week of data.

Dropping direct and organic orders first is the step most people skip. Before comparing anything, make sure both sides assign the device the same way. Meta's breakdown reports the device where the purchase happened, but plenty of people see an ad on their iPhone and check out on a laptop. If your backend tags the order by the device of the first session, the iOS share will look off for reasons that have nothing to do with tracking.

I wouldn't lean on the 5 point threshold, though. At around 150 purchases a week with iOS near 40%, the standard error on that share is roughly 4 points, so a 5 point gap could easily be noise. You'd want closer to 400 purchases before the number means much, which at this budget is probably several weeks of data.

Fair on sample size, but I'd flip the question. Waiting for 400 purchases to settle the iOS gap can take months, and all that time you're running on a weaker signal. If setup is an afternoon through a native integration, it's cheaper to just turn it on with matching event IDs and watch Event Match Quality and reported purchases for four weeks. Turning it off again is easy, so the test runs itself and you don't need certainty first.

EMQ and reported purchases can both mislead you here. EMQ only tells you how well the customer parameters match, not whether the counts are right, so it can look strong while iOS buyers are still missing. Reported purchases will probably go up after you turn CAPI on, but not just because you recovered sales. Attribution shifts and the learning phase move that number too. The check that holds up is comparing backend orders against Meta purchases over the same window, both before and after the switch.