Meta Attribution After November 12: Check Eligibility, Then Baseline Your Conversions
Since November 12, Meta's on-device conversion setting is on by default for eligible pixels. Check which pixels qualify, export your pre-change baseline, then use the ratio rule to keep it on or opt out.
Last checked: November 16, 2026.
Since November 12, 2026, Meta can measure some website conversions on the customer's own device, through a Meta Pixel setting called "Measure conversions on customer devices" that Meta's Business Help Center says is turned on by default. Meta attribution numbers, reported conversions and ROAS, can rise with no change in real sales. Check whether each pixel is eligible, export a baseline of the weeks before November 12 now, and decide dataset by dataset whether to keep the setting on.
The November 12 date isn't on Meta's help pages. It comes from the grey label under the toggle in Events Manager, "Measurement starting November 12, 2026," as PPC Land reported on October 27 and stape reported on October 22. If you report Meta CPA or ROAS to a CEO or a board this quarter, that number may already be moving for a reason that has nothing to do with your ads. And if you're still deciding whether Meta moves revenue at all, start with my guide to incrementality testing for B2B SaaS.
Key takeaways
- Meta's "Measure conversions on customer devices" setting is on by default for eligible pixels, and the Events Manager label dated the start of measurement to November 12, 2026.
- Meta's U.S. English setup page lists two conditions: the Pixel installed directly on the page, not in an iframe or as an image-only pixel, and a website that requires people to log in with an email or mobile phone number.
- Meta says the attribution window and attribution model don't change, and the feature isn't active for people in the EU and UK.
- A recovered conversion is a purchase that already happened and that Meta can now tie to an ad. It isn't a new sale, and it isn't proof the ad caused the sale.
- Export October 13 to November 11 now as a daily baseline of Meta-attributed purchases next to backend orders, then check the ratio for each day since November 12 and reconcile weekly.
What changed on November 12?
On November 12, 2026, Meta added a second place to match an ad to a purchase: the customer's device. Until then, the Meta Pixel sent a purchase event to Meta's servers, and Meta tried to connect it there to someone who saw or clicked an ad. With the new setting on, Meta says "the system measures conversions on your customer's device. It uses information they already shared with you (like an email) and temporary, non-identifying device information."
Meta's help pages confirm the rest:
- The setting "is turned on by default," and the setup page adds that it's on by default for eligible pixels and you can turn it off at any time.
- It "is not active for people located in the EU and UK."
- "Your attribution window and attribution model don't change." Standard settings stay where they were: 1-day or 7-day click-through, 1-day view-through and 1-day engage-through, per Meta's attribution settings page.
- It "works alongside your existing Pixel and Conversions API setup and doesn't replace it."
- The toggle sits in Events Manager under Data sources, then your pixel, then Settings, below the Meta Pixel settings section.
The Events Manager panel says the measurement "happens on your customer's device instead of exclusively on Meta servers." PPC Land founder Luís Rijo points out that "instead of exclusively" adds a device-side path next to the server one rather than replacing it. The timing is tight too. November 12 falls 15 days before Black Friday on November 27, and it's the third change to Meta's counting this year. On January 12 the Ads Insights API stopped returning 7-day and 28-day view-through windows, and on March 3 Meta limited click-through attribution to link clicks, both per PPC Land.
Meta hasn't announced the rollout on any public page. When I re-read both help pages on November 16, neither gave a start date. Wortfilter, the German e-commerce blog that reported Meta's notice to advertisers on October 18, wrote that Meta would roll the setting out gradually from November 12, per PPC Land. So a pixel that shows no change yet may still switch on later.
Is your pixel eligible for on-device measurement?
Meta's U.S. English setup page lists two requirements, and the second one rules out a lot of websites. The first: "You must install the Meta Pixel on your website. Install the pixel directly on your page, not inside an iframe or as an image-only (img) pixel." The second: "Your website must require people to log in to an account using an email or mobile phone number."
Read the U.S. English version of the page. Not every render of Meta's default page has shown the login line, and that line is the one that decides most SaaS cases. Jamie Isabel, founder of MetricMaven, listed the same login condition in an October 19 guide.
Here's how I read that for a SaaS company. Most SaaS companies have two places a pixel can live. The marketing site, where a visitor books a demo or starts a trial without signing in, doesn't require a login, so on Meta's stated terms it doesn't qualify. The product, where a user signs in with an email before upgrading to a paid plan, does. If your Meta Purchase event fires on an in-app upgrade or checkout page, that's the dataset likely to change. If your only pixel sits on a demo-led marketing site, expect little or no change.
Two more limits come from Meta's own wording. Meta describes the feature entirely in terms of purchases: its example is a lamp, and Meta says it "only receives confirmation that a purchase happened." Meta hasn't documented whether Lead, CompleteRegistration or StartTrial events qualify, so an account that optimizes only for lead events may see little or no change until Meta says otherwise. And Meta says the match happens on the device "where both the ad view and the purchase are present." Read literally, that helps when someone sees the ad and buys on the same device. A buyer who sees an Instagram ad on a phone and upgrades from a work laptop is a different case, and Meta hasn't said how it handles it.
Why can your Meta numbers rise when sales don't?
Your Meta numbers can rise because the setting changes what Meta can connect, not what customers buy. Meta's own example: "someone sees your ad for a lamp on Instagram, then visits your website two days later and buys it. Meta may have a record of the ad view and receive the purchase event, but can't always connect them to attribute the purchase to your campaign." With the setting on, "that match happens on the customer's device."

So the purchase event already reached Meta. What changes is whether Meta can credit it to an ad. Stape put the consequence plainly: "An increase in attributed conversions does not necessarily mean an increase in the actual number of conversions." It adds that reported CPA could fall and ROAS rise "even if actual sales and advertising spend remain the same." Isabel makes the same point from the measurement side: "The attribution rules stay the same, so the counts can move even when performance has not changed."
Recovered isn't the same as caused. These are real purchases, and more of them now carry Meta's credit. Whether the ad caused any of them is a separate question that no attribution window answers. Meta has argued for years that standard reporting undercounts its ads. Its "Suite of Truth" white paper, based on 307 studies across 54 advertisers, said rules-based attribution undervalues Meta by 31% at the median, per PPC Land. This setting moves Meta's own reports in that direction.
What hasn't Meta said yet?
Four days after the start date, Meta still hasn't said whether recovered conversions feed bidding, whether purchase values travel with them, which browsers are covered, or how the device match interacts with Conversions API events. Both help pages read the same on November 16 as they did when I checked them on November 4. Treat each one as an open question, not a fact.
- Optimization: Meta's reasons to keep the setting on include "Improve campaign performance with a stronger signal." That points at delivery, but Meta hasn't said whether recovered conversions count toward bidding, value optimization or the learning phase. It matters because Meta said on October 21 that ad sets need roughly 50 conversions a week to learn, and that ad sets below that level ran about 28% more expensive per purchase in its own analysis, per PPC Land.
- Value: Meta says it receives "confirmation that a purchase happened and which campaign to attribute it to." It hasn't said whether purchase value comes with it, which decides whether ROAS moves or only conversion counts.
- Coverage: Meta hasn't said which browsers and devices are supported. It also hasn't said whether the match runs in Meta's apps, the browser or both. Safari is the case to watch. WebKit said on October 19 that Safari 27 will block identity resolution providers, the firms that turn email addresses into tracking IDs, per PPC Land. Meta hasn't said whether its feature works in Safari.
- Conversions API: Stape notes that "Meta's documentation doesn't explain how on-device match interacts with server-delivered events."
- Rollout and size: Meta's help pages give no start date, no rollout schedule and no estimate of how many conversions the setting adds.
- Turning it off later: Meta says you can turn the setting off at any time. It hasn't said whether that changes conversions already reported.
How do you baseline your conversions around November 12?
Export October 13 to November 11 now: daily Meta-attributed purchases and purchase value for each dataset, next to the same days from your backend. Write the date you pulled the data into the file. Then compare the ratio for each day since November 12 against the baseline's normal range.
The pull date matters because Meta hasn't said whether days before November 12 can pick up recovered conversions later. Re-pull the same days in a few weeks. If Meta's side of the baseline comes back higher, those earlier days were revised, and your first file is the cleaner comparison.
- Record which datasets have the setting on. Open Events Manager, then Data sources, then each pixel's Settings, and note the state of "Measure conversions on customer devices." It's set per pixel, so check every one.
- Export Meta's side for October 13 to November 11: daily purchases and purchase value per dataset at your current attribution setting. That's 30 days. If your business runs on a monthly cycle, go back 60 days instead. Don't change attribution settings this month.
- Pull the backend side for the same days: new paid subscriptions and first charges from Stripe, orders from Shopify, or self-serve closed-won deals from the CRM. Use the system finance closes the month on. It's the same discipline as counting every enrollment source against the whole channel mix for a Texas residential energy sales organization: the business's own records are the denominator, not the platform's.
- Compute the ratio of Meta-attributed purchases to backend purchases, daily and as a 30-day average, and note its normal range so you know what ordinary noise looks like.
- Check every day since November 12 against that range. Pull the same two numbers for each new day and compute its ratio. A single day outside the range is noise. A run of days above it while backend orders hold flat is the counting change showing up.
- Annotate November 12 in every dashboard and report that shows Meta CPA or ROAS, including the board deck template. Add January 12 and March 3 while you're there.
- Keep the Conversions API running, and make sure the Pixel fires once per purchase. Meta says the new setting works alongside the Pixel and Conversions API. Meta also says advertisers who run its one-click Conversions API alongside the Pixel see an average 17.8% lower cost per result, per PPC Land.

Here's what the check looks like for an illustrative self-serve SaaS product with $21,000 of Meta spend in each 30-day window. These numbers are made up to show the arithmetic. They aren't a client result.
| 30-day window | Meta spend | Meta-attributed purchases | Stripe new paid subscriptions | Meta-to-backend ratio | Meta-reported CPA |
|---|---|---|---|---|---|
| October 13 to November 11 (baseline) | $21,000 | 140 | 410 | 34% | $150 |
| November 12 to December 11 (illustration) | $21,000 | 168 | 410 | 41% | $125 |
Meta-reported CPA fell 16.7%, from $150 to $125, and Stripe didn't add a single subscriber. That jump of about seven points in the ratio is counting, not growth. Without the baseline, it reads as a win in the next board meeting.
Should you keep the setting on or opt out?
Use what I call the ratio rule: keep the setting on for any dataset whose Meta-to-backend ratio moves by an amount you can explain and annotate, and opt that dataset out when the ratio moves and you can't. The rule only works if you reconcile weekly against backend orders. Opt out too if a board report depends on period-over-period Meta CPA that you can't annotate.
| Situation | The ratio rule says |
|---|---|
| Dataset isn't eligible (no login, or an iframe or image-only pixel) | Nothing to decide. Note it in the baseline file. |
| Eligible, reconciled weekly, ratio moves by an amount you can annotate | Keep it on. |
| Eligible, ratio moves and you can't explain it | Opt that dataset out and investigate. |
| A board report compares Meta CPA period over period and can't carry an annotation | Opt out until the report can, or move the report to backend CAC. |

I default to on because more of the purchases Meta credits will be real purchases it used to miss, and if Meta does use them as signal, delivery may improve. Yiqi Wu, chief executive of Aimerce, reached the same place. Wu's X profile lists ten years of engineering work at Meta, Instagram and Reddit. "I wouldn't opt out by default," Wu wrote. Wu's case for turning it off is a reconciliation failure, when "your numbers stop matching the orders and you can't figure out why," as PPC Land quoted the post. Aimerce sells Pixel tracking to Shopify merchants, so weigh that. The logic holds without it.
How do you prove lift without trusting Ads Manager?
You prove lift with a holdout. Meta's Conversion Lift, or a geo holdout read on backend outcomes, measures what the ads caused. Ads Manager, before or after November 12, measures what Meta can attribute. Those are different numbers, and this change only moves the second one.
The research has been clear on that gap for years. Gordon, Zettelmeyer, Bhargava and Chapsky compared observational methods with 15 large randomized experiments run at Facebook. Their study, published in Marketing Science in 2019, found that the observational methods often failed to recover the lift those experiments measured. Karan Dhir, a marketing measurement product lead at Genentech, drew on that line of research. Dhir argues that observational methods frequently miss experimental results by a factor of three or more, sometimes high and sometimes low, per PPC Land.
At SaaS volumes, a geo holdout read on Stripe subscriptions or pipeline is usually more practical than a platform lift study. Meta's open-source GeoLift package does the analysis for free, and my incrementality testing guide walks through the power check and the read. Meta also offers an incremental attribution model that "optimizes delivery for incremental conversions using models that predict whether a conversion is caused by an ad." That's a model's prediction, not a test, so it doesn't replace a holdout.
What should you tell your CEO?
Tell your CEO now, before the next report, in two sentences. Something like this:
"Since November 12, Meta has been counting some purchases it couldn't connect to an ad before, so part of any Meta CPA or ROAS improvement after that date comes from Meta changing what it counts. I'm reconciling Meta against Stripe every week and will report growth on the Stripe number."
If the board deck shows Meta CPA quarter over quarter, add the annotation to the Q4 slide now. The number that matters for a SaaS company is still the backend one: CAC payback on paid subscriptions, or pipeline from self-serve accounts. LinkedIn raises the same question for B2B pipeline, which my LinkedIn Conversions API decision guide covers. If you'd rather have the reconciliation built for you, that's the measurement and attribution work I do alongside paid media programs.
Frequently asked questions
Does on-device measurement change my Meta attribution window?
No. Meta says "Your attribution window and attribution model don't change." The setting changes which conversions can be measured, not how they're credited to your ads.
Is it active in the EU and UK?
No. Meta says the feature "is not active for people located in the EU and UK," so expect any change to show up only in traffic from other regions.
Is the setting already on for my pixel?
Probably, if the pixel is eligible. The Events Manager label gave November 12, 2026 as the start of measurement, and Meta's help pages say the setting is on by default for eligible pixels. Meta hasn't confirmed the rollout on a public page, and Wortfilter reported a gradual rollout, per PPC Land, so open each pixel's Settings in Events Manager and record what it shows.
Is it too late to take a baseline?
No. Export October 13 to November 11 now and write the pull date in the file. Then compare each day since November 12 against that range. Meta hasn't said whether earlier days can pick up recovered conversions, so re-pull the same days later and check that they still match.
Do I need to change any code?
Probably not. Meta's notice to advertisers, as PPC Land reported it, says no code changes or integration updates are needed, and the help pages describe a setting that's on by default for eligible pixels. The Pixel does have to be installed directly on the page, not in an iframe or as an image-only pixel, and the site has to require a login with an email or mobile phone number.
Where is the toggle?
In Events Manager, go to Data sources, select your pixel, open Settings, and find "Measure conversions on customer devices" below the Meta Pixel settings section. It's set per pixel.
Does it affect lead generation campaigns?
Probably very little. Meta's help pages describe purchases only. If your account optimizes only for Lead or other non-purchase events, you may see little or no change unless Meta documents that those events are eligible.
Will it raise my reported ROAS?
It might. Stape says advertisers may see reported conversion value rise, but Meta still hasn't said whether purchase value travels with recovered conversions, so ROAS could move less than conversion counts do.