Ecommerce Attribution After iOS Privacy Changes

Stop trying to rebuild the perfect click path. It is gone. After the iOS privacy changes gutted third-party tracking, ecommerce attribution became a modeling problem, not a tracking problem. The brands still winning do not have better pixels. They have a blended model, a habit of running holdout tests, and the discipline to trust contribution margin over a dashboard number. If your agency is still promising last-click precision, they are selling you a fiction that stopped being true years ago.

Why ecommerce attribution broke

The old model assumed you could follow one shopper from an ad impression to a purchase, deterministically, across devices and sessions. That assumption died. Apple's App Tracking Transparency cut off the identifiers that made platform-reported conversions reliable. Browsers cap cookie lifetimes. Shoppers research on a phone and buy on a laptop three days later.

So now Meta reports one number, Google reports another, your Shopify dashboard reports a third, and they add up to more sales than you actually made. Every platform claims the same order. That is not a bug you can fix with better UTM hygiene. It is the structural reality of a privacy-first web.

The mistake I see most is treating platform-reported ROAS as truth. It is a self-graded exam. Meta decides which orders Meta caused. Of course the number looks good.

Build a blended model, not a perfect one

The honest replacement is a blended view. Take total revenue for a period, subtract everything you can attribute to organic and returning customers, and look at what your paid spend produced against the remainder. Marketing efficiency ratio, total revenue divided by total ad spend, is crude but it does not lie to you the way channel-level ROAS does. It cannot double-count, because there is only one top line.

Layer platform data on top for direction, not for truth. If Meta says a campaign is your best performer, that is a hypothesis worth testing, not a fact worth reallocating budget on. The blended number keeps you honest about whether marketing as a whole is working. The channel numbers help you guess where.

This is the same discipline I apply everywhere else in the portfolio: one number that cannot be gamed beats a dozen that each flatter their own source. I wrote about that pattern in why one system of record beats a dozen dashboards.

Run holdout tests to find real incrementality

Modeling tells you correlation. Incrementality testing tells you causation. The only way to know if a channel actually drives orders is to turn it off for a slice of your audience and watch what happens.

Geo holdouts are the cleanest tool for ecommerce. Pause a channel in a set of matched regions, keep it running everywhere else, and compare. If sales in the paused regions hold steady, that channel was harvesting demand you already had. If they drop, it was creating demand. That difference is worth more than any attribution window setting.

Brands hate this because it feels like leaving money on the table. It is not. It is buying certainty. One well-run holdout can save you six figures in wasted spend on a channel that was taking credit for organic sales. That is the exact trap I describe in demand generation mistakes that burn your budget.

Judge marketing on margin, not ROAS

Here is the part that changes how you run the whole business. ROAS is revenue over spend. It ignores product cost, shipping, returns, and discounts. A 4x ROAS on a product with a 30 percent margin can still lose money once you account for the free shipping and the return rate.

Move your reporting to contribution margin after acquisition cost. New customer CAC against first-order margin, then against lifetime value, tells you whether growth is actually building the business or just renting revenue. Most ecommerce dashboards will not show you this because it requires stitching ad spend to real unit economics. That stitch is exactly the work a serious agency should own. It is the core of how to measure marketing agency ROI honestly.

What to ask your agency

If you are hiring out your marketing, three questions separate the operators from the reporters. Do you measure blended efficiency, or only platform-reported ROAS? When did you last run a holdout test, and what did it show? Do your reports use contribution margin or top-line revenue?

If the answers are vague, you have found a reporter, not an operator. Attribution after privacy changes is a modeling and testing discipline, and it is the difference between spending confidently and spending in the dark. This is the way I build measurement for every brand at Girard Media, because the number you trust decides the budget you set, and a lying number sets a losing budget. If you also want to stop renting your own data, read own your analytics data instead of renting Google's and put your own measurement layer under all of it with Girard Media.