Attribution is the most argued about topic in growth, and the least understood. Teams either trust a platform's black-box model blindly or distrust all attribution so much they make decisions on gut feel. Neither works. Here's a pragmatic approach to attribution that's transparent enough to trust and simple enough to actually use.
Accept that perfect attribution doesn't exist
No model — last-click, first-click, linear, data-driven, MMM — captures the full truth of how a customer decided to buy. Every model is a simplification. The goal isn't the perfect model; it's a model whose simplifications you understand and can defend.
Once you accept that, you stop chasing the impossible and start choosing the model that fits your decision. Different questions need different models.
Match the model to the decision
Use last-click for budget reallocation within a single short funnel — it's simple and good enough for 'which ad drove the sale.' Use multi-touch or data-driven for understanding channel contribution across a longer journey. Use MMM for high-level budget allocation across channels when you can't track individuals.
The mistake is using one model for everything. Last-click will starve your brand and content budgets; multi-touch will over-credit touchpoints that didn't matter. Pick the model that answers the question you're actually asking.
Make the model transparent, not magical
A black-box 'data-driven' model that no one can explain is a liability. If you can't explain why a channel got the credit it got, you can't defend the budget decision that follows. Prefer models whose logic you can walk a stakeholder through in five minutes.
Document the model, its assumptions, and its known blind spots. A transparent model with acknowledged gaps beats a magical model everyone pretends is perfect — until it isn't.
Validate with incrementality, not just attribution
Attribution tells you where credit went; incrementality tells you whether the spend caused the outcome. Periodically run holdout tests — pause a channel for a defined audience and measure the revenue delta — to check whether your attribution model is telling you the truth.
This is the checksum that keeps attribution honest. A channel that attribution credits heavily but shows no incrementality is a channel you're overpaying for, and only a holdout will reveal it.
Takeaways
- Perfect attribution doesn't exist — choose a model whose simplifications you understand.
- Match the model to the decision: last-click for short funnels, multi-touch for journeys, MMM for top-level allocation.
- Prefer transparent models you can explain in five minutes over black-box magic.
- Validate attribution with incrementality holdouts — it's the checksum that keeps the numbers honest.