SSubida Skills

Esc to close · results from all notes, skills and people

Marketing›Measurement

When last-click attribution is good enough (and when to invest in incrementality)

For small, performance-driven advertisers on one or two channels, last-touch attribution is often ~85% as good as experiments and better than a bad MMM; invest in incrementality tests as spend, channel count and brand effects grow.

Updated 2026-10-24·3 sources·measurementattributionincrementalitymmmsmall-budget

When last-click attribution is good enough

TL;DR

  • Kaminsky (an MMM vendor) argues most businesses don’t need an MMM, and that an average MMM is net-negative because it actively misleads.
  • One study he cites found last-touch about 85% as good as running only incrementality experiments.
  • Use common sense. At your first dollars, or with nearly all spend on one channel, there are few competing claims for each conversion.
  • Escalate when spend is big and mixed. Run a geo test and ask how far off last-click is: about 5% means move on; over 20% means dig in.
  • Fewer channels make measurement easier. One veteran gaming UA lead keeps small budgets on 1–2 channels partly for this reason.

Details

Last-touch is underrated for small advertisers. Michael Kaminsky says last-touch “has lots of problems”, but for small, performance-marketing-driven businesses it often works well, and an MMM would do worse.1 He cites recent studies finding last-touch about 85% as good as only doing incrementality experiments. In small businesses he works with, more Facebook/Google spend visibly produces more conversions in the raw data, and that lines up with last-touch.2 Last-touch starts to fail with complex mixes and much larger budgets.2

The common-sense dimension (Eric Seufert). If you’re spending your first dollar on digital ads, causality isn’t your problem. Misattribution becomes realistic as you add word of mouth, a brand, loyal customers and several digital channels running at once.3 Kaminsky: a business spending $1M/month mostly on Facebook is probably close to 100% incremental, with little baseline. A test there exists to size the error. Off by 5%? It probably doesn’t matter. More than 20%? Dig in.4 Seufert stresses this is a scale question, not an argument against incrementality testing.5

Keep the mix narrow early. Nebojsa Radovic (Zynga) tells six-figure-a-month game advertisers to run just Google on Android and AppLovin on iOS, and do them well. Measurement becomes “really, really easy” with one channel per platform.6

Signals that last-click is no longer enough:

  • Several channels compete for the same users, especially search plus social plus video.
  • Spend moves into view-based channels (YouTube, CTV, chatbot ads) that last-click can’t see.7
  • Channel ROAS looks fine but the top line doesn’t move when you scale.

Disagreements / caveats

  • Olivia Kory (Haus) reports cases where last-click ranks channels in the reverse order of incrementality (Google UAC standard vs video). See How last-click attribution biases channel results. Even a “good enough” last-click can misallocate between channels.
  • The 85% figure is one study Kaminsky mentioned from memory, not a benchmark to rely on.

See also

Footnotes

  1. S07, E: Demystifying open source MMM (with Michael Kaminsky) @ 14:24 — Michael Kaminsky ↩

  2. S07, E: Demystifying open source MMM (with Michael Kaminsky) @ 15:17 — Michael Kaminsky ↩ ↩2

  3. S07, E: Demystifying open source MMM (with Michael Kaminsky) @ 17:40 — Eric Seufert ↩

  4. S07, E: Demystifying open source MMM (with Michael Kaminsky) @ 18:29 — Michael Kaminsky ↩

  5. S07, E: Demystifying open source MMM (with Michael Kaminsky) @ 20:05 — Eric Seufert ↩

  6. S07, E33: An oral history of mobile user acquisition (with Nebojsa Radovic) @ 42:10 — Nebojsa Radovic ↩

  7. S07, E21: AppsFlyer and the new dynamics of mobile attribution (with Olivia Kory) @ 22:17 — Olivia Kory ↩