Every brand I talk to has the same story about Meta attribution. It went bad in 2021, everyone blamed Apple, agencies sent out a PDF about "the iOS 14.5 update," and then... nothing. Three years later the numbers are still bad, but nobody's mad about it anymore. They just adjusted their expectations downward and called it the new normal.
Here's the part almost nobody talks about: iOS 17 broke it again in September 2024, and I'd bet money most of the marketers reading this never rebuilt anything after either event. They just got used to worse numbers.
Two separate things broke your data, not one
The story everyone knows is Apple's App Tracking Transparency, which rolled out with iOS 14.5 in April 2021. ATT requires apps to explicitly request permission before tracking users across apps and websites, and when a user declines, Meta loses the ability to match that person's in-app behavior to your ad. Most people said no. The opt-in rate for tracking permission has generally remained below fifty percent globally, though it varies by region and app category. That's the event everyone remembers.
What almost nobody rebuilt their stack for is the second hit: in September 2024, iOS 17 added Link Tracking Protection, a feature that strips tracking parameters — including fbclid, the identifier Meta uses to connect ad clicks to conversions — from URLs when users browse in Safari Private Browsing mode. That's a direct hole in the exact mechanism Meta uses to say "this click led to this sale." And it landed quietly, folded into an OS update most people installed without a second thought.
So you've got two structurally different breaks — one that nukes visibility into iOS app behavior, one that strips the click ID that ties a website conversion back to an ad — stacked on top of each other. Most measurement setups I audit were patched for the first problem in 2021 and never touched again.
Why "the numbers just got worse" was the wrong response
The natural reaction to declining reported conversions is to assume performance declined. That's often false. The gap isn't necessarily in your creative or targeting — it's in what Meta can actually see and stitch together on the back end.
A browser pixel alone was never built for a cookieless, permission-gated world. The Pixel relies on cookies, which can be blocked or deleted by users, leading to data loss, and it may not fire on your store's thank-you page, resulting in missed conversions. Layer Safari's privacy defaults and iOS 17's link stripping on top of that, and a pixel-only setup is reporting on a shrinking fraction of what's actually happening.
This is exactly why Meta built the Conversions API as a parallel, server-side path. Instead of relying solely on the Facebook Pixel, you send conversion events directly from your server to Meta, giving Meta visibility into conversions the pixel misses. CAPI isn't a nice-to-have add-on anymore — it's the thing doing the job the pixel used to do alone, back when browsers weren't actively working against it.
And the two aren't redundant. Because of these limitations, many advertisers use the Pixel and CAPI together, running both to ensure events are captured client-side and passed directly from the server, reducing data loss. If you're only running one, you're only getting half the picture Meta is capable of building.
The "new normal" is a data-quality problem, not a market problem
I've seen teams cut budget on Meta because "ROAS tanked," when the actual issue was measurement infrastructure nobody had touched since the pixel was first installed. Think about a hypothetical DTC brand doing, say, $40K a month on Meta with a healthy Shopify backend — if the platform-reported conversions drop noticeably quarter over quarter while total store revenue holds steady, that's not a demand problem. That's a visibility problem. (To be clear: that's an illustrative scenario, not a client number — but the pattern is one I see constantly.)
The fix isn't complicated, but it does require actually doing the work instead of adjusting your expectations:
- Check your Event Match Quality score. This is the diagnostic most people never look at. Meta scores how confidently it can tie an event back to a real person, and for purchase events you want to be sending enough hashed identifiers — email, phone, name, location — to push that score up, not just relying on a browser cookie to do the work.
- Run Pixel and CAPI together, deduplicated properly. Sending the same event from both sources without a shared event ID inflates your numbers in the other direction — now you're overcounting instead of undercounting. Both are bad. Get the dedup right.
- Stop trusting platform-reported ROAS as your only source of truth. Compare it against your actual backend revenue on a recurring basis. If there's a persistent, meaningful gap, that gap is your attribution blind spot, not your creative team's fault.
- Treat every major iOS release as a measurement event, not just a product update. iOS 17's Link Tracking Protection didn't make headlines in most marketing newsletters. The next one probably won't either. That's on you to track, not on Apple to announce loudly.
Why brands never went back and fixed it
Honestly? Because rebuilding a measurement stack is unglamorous, unbillable-feeling work compared to launching a new creative test or chasing a new ad format. It's plumbing. Nobody gets a promotion for finally implementing server-side tracking correctly. But every dollar of budget decision you make on top of broken measurement is a guess dressed up as a data-driven call.
The brands that are actually winning on Meta right now aren't the ones with the best creative — plenty of average creative is running profitably. They're the ones who stopped accepting degraded numbers as gospel and went back to fix the pipe. If you haven't looked at your Events Manager setup since 2021, I promise you, you're not looking at reality. You're looking at whatever's left after three years of accumulated signal loss, and calling it your baseline.
Where to start this week
Don't try to fix everything at once. Pull up Events Manager, look at your Event Match Quality score for your primary conversion event, and see where it actually sits. If it's mediocre, that's your answer right there — not "the algorithm hates us," not "the audience is fatigued." Just an infrastructure problem wearing a performance-problem costume. Fix the plumbing before you touch the creative.