Don't let Meta grade its own homework: reconcile ROAS against a second source
Starting November 12, Meta will report more purchases for the same ad spend, and that is not proof the ads got better. The ad networks are about to grade their own homework with a more generous rubric. A second, independent source of truth stops being optional.
According to PPC Land and Stape, Meta is switching on a setting called "Measure conversions on customer devices" by default for every dataset already sending events. Advertisers have to opt out to stop it. Meta pairs an email address your customer gave you with temporary device data, and returns only the fact of a purchase and the campaign that gets the credit. Attribution windows and models don't change. The count of credited purchases does.
What we know, and what Meta hasn't said
The rollout is scheduled for November 12, about two weeks before Black Friday on November 27. It doesn't apply to people in the EU or UK. Per PPC Land, which cites MetricMaven reading Meta's help pages, the Pixel has to sit directly on the page rather than in an iframe, and the site has to require a login with an email address or mobile number. If you run a login-gated store, a loyalty app or an ordering flow, assume you're eligible.
What Meta hasn't said matters more. It hasn't said whether the recovered conversions feed bidding, whether purchase values are included, or how consent is handled. So you can't yet tell whether this is a reporting change, an optimization change, or both. If it changes bidding, your campaigns will behave differently in the busiest weeks of the year and you'll have no clean before-and-after.
Why a platform can't be its own scorekeeper
Every ad network sells you the outcome it also measures. That's a structural conflict, and it exists whether or not anyone is acting in bad faith. The platform decides which touches count, how long they count for, and now which device signals get matched to your customer list. A higher number from that system tells you the system's counting changed. It doesn't tell you a customer who wouldn't have bought, bought.
To be fair to Meta, there's a real argument that platforms undercount. PPC Land cites Meta's "Suite of Truth" paper, which covers 307 studies across 54 advertisers and puts median undervaluation of ads at 31%. That is also Meta's own research, which proves the point from the other side: both the vendor and the skeptic need a source neither of them controls. The same article notes that ad sets below roughly 50 conversions a week stay in the learning phase and ran 28% more expensive per purchase, so Meta has a real incentive to find more conversions. Read the PPC Land analysis before you decide which side of that argument you're on.
The same logic applies to Google, TikTok, Pinterest, Amazon and every AI-search ad product coming online. As more of the journey moves onto platform-owned surfaces and agents do more of the clicking, the platform's report gets less connected to what shows up in your bank account. Third-party and first-party analytics become the control group for all of it.
Reconcile against a source the ad network doesn't control
The principle is simple: every platform-reported number gets compared to at least one record of what actually happened. That means orders in your commerce system or POS, revenue in your finance tools, and a separate analytics layer that attributes traffic on its own terms. For a restaurant group, that's the POS and online-ordering totals. For ecommerce, it's Shopify orders. For B2B, it's closed pipeline in the CRM.
What you track is the ratio, not either number alone. Divide platform-reported purchases by first-party orders each week. Before November 12, that ratio is your baseline. If it jumps the week the setting turns on while real orders stay flat, you've measured the change in Meta's counting and not a change in performance. Agencies that spotted the setting early are already telling clients the same thing: compare Meta purchases with store orders before trusting the new totals.
Reconciliation tells you the count is off. It doesn't tell you which ads drove sales. For that you need an incrementality test: a holdout audience or region that doesn't see the ads, run for long enough to read. It's slower than a dashboard, and it is the only method that answers the question the dashboard can't.
What to do about it
- Before November 12, pull four weeks of Meta-reported purchases and first-party orders. Compute the ratio and save it as your baseline.
- Check the Events Manager setting under Pixel Settings. Decide per dataset whether to leave it on, and write down the date and the decision so you can explain any change in the numbers later.
- Keep an independent attribution layer running in parallel. Use your analytics tool, server-side events, and a post-purchase "how did you hear about us" question as cross-checks.
- Run a holdout on your biggest campaign. A simple audience or geo split during the Black Friday window gives you incremental lift, not credited conversions.
- Change budget decisions to follow reconciled numbers. Treat any ROAS improvement that shows up only inside the ad platform as unconfirmed until the second source agrees.