iOS 14 and Signal Loss
What actually broke in 2021, and what the fixes oversell.
April 27, 2021. I remember the date because our Meta purchases fell off a cliff over a weekend and I spent Monday convinced someone had broken the checkout. Shopify was fine. The customers were fine. Apple had simply asked 60 million iPhone users whether they minded being tracked, and they did not need long to think.
The one-sentence version
iOS 14.5 cut the thread between "saw the ad in an app" and "bought on the website" for most iPhone users. The purchases kept happening. The platforms stopped being able to prove they caused them.
The setup
Five years in one slider
Try it: drag through the years
How much of the signal did the pixel lose?
Year
2021
2021: iOS 14.5 ships (April)
App Tracking Transparency: apps must ask permission. Most people decline. Meta's pixel loses the link between an in-app click and a Safari purchase. Windows shrink to 7-day click, 1-day view. (About 38% of purchases invisible to a browser pixel, illustrative.)
The ads never stopped working. Only the receipts did.
What broke, what did not
Separate the real damage from the folklore
| Claim | Verdict | Why |
|---|---|---|
| Meta lost most iPhone attribution | True | ATT opt-in rates sit around 25 to 35%. The rest are dark to the pixel. |
| Lookalikes got worse | True | Fewer observed buyers means a smaller, more biased seed. |
| Meta ads stopped working | False | Delivery still reaches the same people. Reporting and learning suffered. |
| A CAPI checkbox fixes it | Mostly false | CAPI only helps with match keys, de-duplication and verified events. See the Conversions API page. |
| Android saved us | Partly | No permission prompt, but blockers, consent and cross-device still bite. |
| Modeled conversions are fake | Unfair | They are estimates, often decent, never verifiable per order. Do not budget on them. |
Why it matters
A biased sample steers your spend
Meta learns from the purchases it can see. After 2021 it saw mostly the ones that survived: Android, desktop, people who allowed tracking. It went looking for more of those, because that is who "buys". Your iPhone customers were still there. Your ads just stopped being shown to them as often.
That Monday I paused two prospecting campaigns because their ROAS "collapsed". They were our best campaigns. The survey we added later showed it. I still think about the two weeks they were off.
The real fixes
Four that work, in order
1. Server-side, first-party capture
Record clicks and orders on your own domain so nothing depends on a browser cookie surviving.
2. Conversions API with real match keys
Hashed email, click ID, order ID. Match quality above 7. De-duplicated.
3. Post-purchase survey
Catches the in-app impression that never produced a click at all.
4. Order-verified attribution
Decide credit from the Shopify order outward, so the modeled and the double-counted never enter the budget conversation.
Honest footnote
What no fix restores
Nobody gets 2020 back. Deterministic, per-person, cross-app tracking of people who opted out is not coming back and should not. The goal is a verified count of orders and a fair split, not surveillance. That is achievable, and it is what the four steps above do.
Questions people actually ask
Straight answers
Did iOS 14 make Meta ads worse?
It made Meta's reporting worse and its targeting slower to learn. The ads still reach the same people. Stores that restored signal server-side saw delivery recover; stores that did not are still optimising on a partial picture.
What does Meta's 'modeled conversions' label mean?
Meta estimates purchases it could not observe and adds them to your totals. They are statistically reasonable and impossible to verify per order. TrueROAS ignores them and counts Shopify orders instead.
Is Android unaffected?
Mostly, for the in-app permission part. Chrome still respects blockers and consent, and cross-device journeys break on any phone.
Does this affect Google Ads too?
Less from ATT, more from Safari cookie caps and EU consent mode. Google models the gap when volume allows and drops it when it does not.
See your own numbers, verified against your orders.
Free under $10k/month in revenue. 14-day trial above that. Nothing changes in your ad accounts. You will know within a week which platform has been telling stories.
"I built this because I was making $50k-a-month decisions on numbers I knew were wrong. If TrueROAS does not show you something Meta or Google got wrong in the first two weeks, tell me and I will help you cancel."
Rasmus Arvidsson
Founder, TrueROAS. Answers support himself.
Keep going