Marketing Attribution

What It Is

Attribution assigns credit for conversions across the touchpoints that preceded them. Three model families:

  • Last-click — 100% credit to the final touch. Systematically undervalues upper-funnel/brand. Fine only for short, single-touch funnels.
  • Rules-based (linear, time-decay, position/U-shaped) — you define the rules, so assumptions are baked in, not derived.
  • Data-driven (DDA) — ML distributes credit from observed converting and non-converting paths; needs ~300+ conversions/month and consistent UTMs. GA4's default. In 2026 Google made it effectively binary: data-driven or last-click.

The Hard Truth About MTA

Multi-touch attribution shows the journey and gives directional budget guidance — but it does not provide causal truth (it observes correlated paths), is degraded by walled gardens and cross-device breaks, and is not a substitute for incrementality testing.

The attribution gap (~40% of the journey invisible to lean teams): dark social, offline touchpoints, cross-device breaks, walled gardens, view-through inflation.

What Lean Teams Should Actually Do

  • Acknowledge the gap explicitly ("we see ~60% of the journey").
  • First-click for channel discovery; last-click for conversion optimization.
  • Invest in consistent UTM taxonomy before any MTA tooling — clean first-click beats sophisticated MTA on messy data.
  • Run quarterly holdout tests (turn a channel off, measure revenue impact) for causal signal no model provides.
  • Track revenue by cohort, NRR, CAC payback by segment, Rule of 40.

How It Applies to Marketing Factory

Attribution is the feedback signal that lets the factory reallocate spend across channel-selection — but the factory should be honest about its limits and lean on measurement-resistant metrics (cohort revenue, net-revenue-retention, CAC payback). UTM-tagging compliance is exactly the kind of operational task an agent owns end-to-end; causal interpretation of ambiguous signals is not (see agent-ownership-boundary).

Referenced from: ai-marketing-measurement