Attribution
📖 3 min readUpdated 2026-04-19
Attribution models assign credit to touchpoints in the customer journey. Multi-touch attribution is ideal; reality post-privacy is multi-touch plus MMM plus holdouts.
Models
- Last-click (simple, biased toward bottom of funnel)
- First-click (biased toward top)
- Linear (equal across all touches)
- Data-driven (algorithmic)
Tools
- Platform self-report (biased)
- GA4 data-driven
- MMM for macro view
- Holdout tests for ground truth
Pragmatic approach
Use platform data directionally, MER for truth, holdouts for validation.
What to do with this
- Stack attribution methods, each answers a different question, using one alone is how teams get misled
- Use platform attribution for tactical daily decisions only, treat its absolute numbers as directional not truth
- Make MER your weekly North Star, it's noisy but it's the truth platform dashboards can't give you
- Run a channel holdout test quarterly, actual revenue drop is the only way to calibrate platform claims
- Skip MMM until you have 12+ months of clean data + $100K+/mo spend, under that it produces noise