The Ecomm Analyst

It’s 2026 and people are still pretending iOS 14 didn’t happen.

Five years on, the privacy changes Apple rolled out in 2021 are baked into how the entire mobile web works. Most ecom operators have adjusted in some way. Most have also quietly accepted that their attribution data is worse than it used to be, without ever fully understanding what broke or what to do about it.

Here’s the version I wish someone had given me at the time.

What you lost

Before April 2021, Meta and most other ad platforms could see, with reasonable precision, that a specific user clicked an ad, browsed your site, added to cart, and bought. That signal flowed from the user’s device through Apple’s IDFA back to the ad network. Attribution was largely deterministic. Match rates were high.

After ATT, users had to opt in to that tracking. The opt-in rate landed somewhere in the low-to-mid teens for most consumer apps. Roughly 85% of iOS users became invisible to deterministic third-party tracking overnight. Match rates collapsed. Reported conversions dropped. Reported ROAS fell off a cliff.

Then platforms started backfilling.

What replaced it (poorly)

Meta’s response was Aggregated Event Measurement, modeled conversions, and a heavier push toward the Conversions API (CAPI). The first two are statistical estimates. They use the smaller set of users who can be tracked deterministically to model what’s probably happening for everyone else. Some of that modeling is decent. Some of it is the platform marking its own homework with more confident handwriting.

CAPI is the more interesting piece. Instead of relying on the browser to send conversion data to Meta, your server sends it directly. That recovers some of the signal because server-side tracking doesn’t get blocked the way pixel-based tracking does.

The catch with CAPI is that brands implement it badly. The most common failure is double-counting, where the pixel fires, CAPI also fires, and both get reported as separate events, inflating apparent conversions. The second most common failure is missing match keys, which means Meta receives the event but can’t tie it to a user.

What actually works in 2026

Three things, in order of importance.

First, server-side tracking done correctly with proper deduplication. If you’re sending events through CAPI (or Google’s equivalent, or TikTok’s), make sure you’re hashing email and phone, sending the event ID on both pixel and server-side calls, and validating in the platform’s diagnostic tools. If your event match quality is below 7 out of 10, fix that before anything else.

Second, first-party data. Capture email earlier in the funnel, not just at checkout. The more first-party identifiers you collect, the more of the attribution gap you can close on your own side, independent of what any platform is willing to share with you.

Third, multi-source attribution that doesn’t depend on any single platform being honest. This is where I use ThoughtMetric. It pulls Shopify order data, platform spend, and first-party identifiers, and stitches them into a customer journey view that doesn’t go dark when ATT kicks in. The reporting is conservative compared to what Meta will tell you, but it’s reproducible and it’s the same methodology applied to every channel.

The honest answer

You’re not getting back to pre-iOS 14 attribution accuracy. Not with CAPI, not with modeled conversions, not with any third-party tool. The signal is genuinely gone for a meaningful chunk of your customers.

The goal isn’t precision. The goal is a consistent measurement framework that lets you compare channels against each other on the same basis, run incrementality tests to validate the big calls, and stop pretending Meta’s reported ROAS is the same number it was in 2020.

Bottom line

Implement server-side tracking properly. Capture first-party data aggressively. Use a third-party attribution layer that’s consistent across channels. Accept that your numbers will be lower than the platforms say, and that this is fine because they’re directionally honest.