Marketing Attribution After the Cookie

Third-party cookies are mostly gone. iOS asks people to opt out of tracking, and browsers block pixels by default. The tidy reports that once told you which ad drove which sale are lying to you now, and marketing attribution has quietly become the thing that keeps media buyers up at night. We run these accounts every day, so here's the honest version: no single number tells you the truth anymore, and pretending one does is how you burn budget.

This isn't a doom post. You can still make good decisions on imperfect data. You just have to change what you ask the numbers to do.

Why last-click was always a bad witness

Last-click gives 100% of the credit to the final touch before purchase. It's popular because it's simple and because the platforms report it by default. It's also wrong in a predictable direction: it overpays the bottom of the funnel and starves the top.

Think about how someone actually buys. They see a TikTok, forget about it, get retargeted on Instagram, then search your brand name a week later and click a branded search ad. Last-click hands the entire sale to branded search. The TikTok that created the demand gets zero. Cut that TikTok spend because it "doesn't convert," and two weeks later your branded search volume caves in. We've watched it happen on real accounts.

Now stack the cookie problem on top. When tracking windows shrink and cross-site cookies vanish, even the last click gets undercounted. Meta can't see a conversion that happens on Safari after the click window closes, so its reported ROAS drifts away from what actually hit your bank account.

No attribution model is correct, and that's fine

The usual reaction is to reach for a fancier model. First-click, linear, time-decay, position-based, data-driven. Each one splits the credit differently, and marketers argue about which is "right" as if a correct answer is hiding somewhere.

It isn't. Every attribution model is a guess about human behavior dressed up as math. Time-decay assumes recent touches matter more, which is sometimes true and sometimes nonsense. Data-driven models need enormous conversion volume, and most accounts don't have it. We got into the mechanics in our guide to marketing attribution, but the short version: pick a model, know its bias, and stop expecting precision it can't give you.

The models still earn their keep. They show relative direction. Which channel is trending up, which creative is pulling weight. They just can't hand you the exact dollar a given ad returned. Read them as a compass, not a GPS.

Blended numbers and the MER view

This is where we spend most of our time now. Instead of trusting any one platform's self-reported return, we zoom out to blended metrics: total revenue divided by total ad spend across everything. That's MER, media efficiency ratio.

MER doesn't care about cookies, opt-outs, or attribution windows. It's money in versus money out. Spend $100k across Meta, Google, and TikTok and make $400k, your MER is 4.0. Push spend up and MER holds or climbs, the machine is working. MER falls, something is off, even while every platform dashboard brags about its own ROAS. We laid out ROAS versus MER and when each one is worth reading.

The catch: MER lags, and it blends. It tells you something slipped, not which channel did it. So we pair it with platform signals and read them together. Blended for the truth about profit, platform data for where to point the fix.

Lightweight modelling and incrementality

Big brands run media mix modelling, months of spend and sales data feeding a statistical estimate of each channel's contribution. You don't need a data science team to borrow the idea. A spreadsheet regression of weekly spend against weekly revenue per channel gets you surprisingly far. It's rough. It's also cookie-proof, because it never touches user-level data.

The cleaner tool is incrementality testing. It answers the only question that matters: if I turned this channel off, how much revenue would I actually lose? You run a holdout. Geo tests are the practical version. Turn a channel off in a set of matched regions, leave it on elsewhere, compare. If sales in the dark regions barely move, that channel was taking credit for demand it didn't create.

We ran this on a brand convinced their branded search was a hero. Paused it in half their markets for three weeks. Revenue there dropped about 4%, not the 30% the last-click report implied. That's a budget decision no dashboard can hand you. It's also how you check whether your work to lower CPA is cutting real cost or just reshuffling credit.

Consistent beats precise

One habit separates buyers who scale from buyers who stall. Pick a measurement approach and hold it steady. A slightly biased number you read the same way every week beats a "perfect" number you recalculate with a new method every month.

Consistency turns noise into a trend. Measure MER the same way every time, and a drop from 3.8 to 3.2 means something. Keep swapping attribution windows and models, and you can't tell a real decline from a definitional one. That's how you make panicked calls on fake signals.

So set the frame. Blended MER as the scoreboard. Platform data for direction. Incrementality tests a few times a year to pressure-test your assumptions. Then leave it alone and let the trend talk.

If your reports and your bank balance keep disagreeing, that gap is usually fixable. We'll run a free ad account audit, map where your marketing attribution is misleading you, and show you the blended numbers your dashboards are hiding.

FAQ

Should I stop using last-click attribution entirely? No, just stop treating it as truth. It's a decent directional read for lower-funnel channels, but pair it with blended MER so you don't overcut the top of the funnel that feeds it.

What MER counts as healthy? It depends entirely on your margins. An 80% gross-margin brand can live on a 2.0, while a low-margin product might need 5.0 or more to clear a profit, so find your own breakeven before benchmarking against anyone else.

Do I need expensive software for incrementality testing? Not to start. A basic geo holdout, one channel off in matched regions and revenue compared against the rest, costs nothing but attention and reads cleaner than most paid attribution tools.