Why does Meta claim credit for conversions my analytics never shows?
Meta reports on its own terms: it credits conversions to the date of the ad exposure, counts view-throughs by default in some setups, and fills gaps in observable data with modeling. Analytics tools usually use last non-direct click on the conversion date. Both are internally consistent and neither is wrong; they answer different questions. Do not try to reconcile them. Compare direction over time, and settle real questions with holdout tests.
Three design choices that create most of the gap
Meta's numbers are not inflated by accident. They follow from documented reporting choices that differ from how analytics platforms count.
- Impression-date reporting. A conversion is credited to the day the person saw or clicked the ad, not the day they converted. This is why last week's numbers keep changing.
- View-through credit. The default click window plus a one-day view window means a person who scrolled past your ad, never touched it, and later found you another way can be counted.
- Modeling. When device-level signal is unavailable, Meta estimates conversions statistically. Modeled results are directional, not auditable, and they cannot be matched to a specific customer record.
Why your analytics disagrees in the opposite direction
Most analytics tools credit the last non-direct source before the conversion. Meta traffic frequently arrives through an in-app browser, cookies get partitioned or cleared, and the click identifier is often the only surviving evidence of the visit. The result is that analytics systematically undercounts Meta the same way Meta systematically overcounts itself.
So you have one number that is too high and one that is too low. Averaging them is not a method. The useful move is to stop treating either as truth and to hold both alongside the outcome you can actually verify, which is booked work in your own system.
What to do instead of reconciling
Three practices cover most decisions honestly.
First, pick one attribution setting and leave it alone, so week-over-week movement means something even if the absolute level is unreliable. Second, report Meta on your own funnel, from lead through appointment through completed job, using the lead id you captured. Third, when the stakes are high enough, run a holdout: turn Meta off in a matched set of zip codes or a matched market for a defined period and watch total lead volume and branded search, not just Meta-reported results.
The halo you are probably not counting
Meta's most valuable effect for local service companies is frequently indirect. Someone sees the ad, does nothing, and searches your company name two weeks later. That conversion is credited to branded search, which makes brand campaigns look brilliant and Meta look mediocre.
Watching branded search volume and direct traffic when Meta spend changes is a cheap, imperfect check on this. It is also the argument for looking at channels together rather than separately, which is what correlation reporting inside business intelligence is for. Single-channel dashboards will keep telling you the same wrong story, which is why marketing intelligence reports across sources against one revenue record.
Topics: attribution · measurement · Meta · modeled conversions · incrementality
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.