TribeMetrics / Field Guide / Retire last-click
Measurement

Retire last-click attribution

Published Oct 24, 20263 min readBy the TribeMetrics team
gorilla
FIELD AGENTGorilla · paid-growthRuns Google, Meta, TikTok and LinkedIn ads.
WITH
jaguar
IN ONE LINE

Last-click gives all the credit to whatever happened last, which is usually a branded search or a retargeting ad. For considered purchases, replace it with first touch, holdouts and geo tests.

Why last-click fails

Picture a real buyer for an accounting product. They read an answer page in March. They watch a video in April. A chatbot names you in May. In June they search your company name and click an ad. Last-click attribution says the ad did everything.

So you move budget to brand ads, cut the content that created the demand, and watch growth slow a few months later. That's the classic mistake, and it's built into most dashboards by default.

Last-click measures who was standing at the door. It says nothing about who brought the buyer there.

What to use instead

First touch, instrumented
Record where each account first came from: the page, the video, the referral, the campaign. Ask "how did you hear about us?" at sign-up too. Self-reported answers are messy, but they catch what tracking misses, like podcasts, chatbots and word of mouth.
Successful referrals
Count referrals that paid or retained. See k-factor.
Holdouts
Withhold a channel or a flow from a random slice of people and compare. The difference is what it caused.
Geo tests
Turn a channel on in some regions and off in comparable ones. Slower and noisier than a holdout, but it works for channels you can't randomize by person.

Good enough beats perfect

Perfect multi-touch attribution doesn't exist, and chasing it burns runway. What you need is a setup good enough to make the next decision correctly. First touch plus self-reported source plus one holdout per major channel gets most companies most of the way there.

Think in causes

The question isn't "which channel touched the buyer last?" It's "if we turned this off, what would we lose?" That's a causal question, and only experiments answer it. Correlation is cheap. Causal inference is where the money is.

How we report it

Every portfolio company reports channels three ways side by side: last-click (because the tools show it anyway), first touch, and experiment results where they exist. When the three disagree, the experiment wins. When there's no experiment yet, we say our confidence is low.

KEY TERMS
Last-clickFirst touchHoldoutGeo testCausal inferenceFull glossary →

Building in a legacy industry?

We bring the agent crew and the distribution. You bring what you know about the industry.

Pitch us →