When acquisition costs rise, the question shifts from 'which channel converted' to 'which channel generated sales that wouldn't have happened otherwise.' This is the difference between attribution and incrementality. A large portion of poorly evaluated spending buys sales that were already on their way — and the report shows positive ROI while total revenue remains unchanged.
Why Attribution Misleads
- The brand click appears at the end and takes credit for a decision that started organically.
- Remarketing audiences often include those who would have bought anyway.
- Attribution models distribute credit without measuring causality.
- Bottom-of-funnel campaigns seem efficient because they harvest, not because they plant.
Three Ways to Measure Incrementality
1. Geoexperiment
Divides similar regions into test and control. It invests in one set and keeps the other off, comparing the revenue difference. It is the most robust method for operations with regional presence.
2. Pause Test
Suspends a channel for a defined period and observes the effect on total revenue. Simple, but requires caution with seasonality and established brands.
3. Audience Holdout
Separates a slice of the audience from a campaign and does not expose it, measuring the conversion difference. Useful in reach media, such as video and display.
How to Set Up the Test Without Disrupting Sales
| Step | Definition |
|---|---|
| Hypothesis | Ex.: 30% of remarketing revenue is incremental |
| Design | Regions, period, and minimum viable volume |
| Metric | Total revenue by region, not clicks |
| Duration | 2 to 4 weeks, avoiding atypical dates |
| Reading | Compare variation between test and control |
What to Do with the Results
- High Incrementality: maintain and scale with protected budget.
- Low Incrementality: cut or switch to a discovery channel.
- Tie: test longer; small samples generate random conclusions.
Errors That Invalidate the Test
- Comparing regions with very different profiles.
- Running during a promotion that distorts demand.
- Changing creatives mid-test.
- Ending before reaching statistically useful volume.
- Ignoring brand effect: a turned-off channel can lower organic search later.
Also Read:
Conclusion
With CAC under pressure, incrementality ceases to be an academic exercise and becomes a tool for budget allocation. Companies that test in a disciplined manner discover that a significant portion of the budget was only harvesting existing demand, while another part supports growth — and they start investing where revenue truly grows.
Measuring causality is more labor-intensive than reading the platform dashboard. It is also the only honest way to decide where to place the next dollar.