Resources · 69
Advertising: distinguish attribution from incrementality
Understand what reports attribute and what a protocol can estimate as an additional effect.
· 2 min
What this guide helps achieve
- Define the decision
- Establish comparable measurement
- Prepare a protocol
- Avoid forcing a conclusion
Quick check
- Do platforms count the same conversion?
- What forms the control comparison?
- Does the decision account for uncertainty?
Step-by-step method
- 01
Define the decision
Specify the proposed investment, audience, period and business action. Attribution reports allocate credit under their rules; they do not directly describe what would happen without advertising.
Deliverable: question and scope.
- 02
Establish comparable measurement
Fix conversion definitions, deduplication and observation windows. Reconcile spend and outcomes on the same scope. Separate revenue, margin and costs; media spend is not a fully loaded cost.
Deliverable: measurement dictionary.
- 03
Prepare a protocol
Where feasible, compare exposed and control groups under a protocol defined before reading results. Review contamination, initial differences and conversion delays. Lift tools have their own eligibility conditions.
Deliverable: protocol and limitations.
- 04
Avoid forcing a conclusion
Report counts, period, estimate and uncertainty. An inconclusive result does not establish zero effect. Before-and-after differences may also reflect seasonality, price or stock. Record the decision and next check.
Deliverable: decision report.
Reusable worksheet
Complete with your authorised observations. These fields are a working template, not observed results.
| Field | Information to record |
|---|---|
| Decision | Budget, audience, action and period |
| Measurement | Conversion, window, costs and deduplication |
| Protocol | Control, contamination, uncertainty and decision rule |
Worked example
Illustrative situation
Fictional example: two platforms each claim the same order.
Decision and expected evidence
The team deduplicates the business result and prepares a controlled comparison. Attribution reporting and lift estimates remain separate.
Distinguish the mechanisms
| Mechanism | Purpose | Check or limitation |
|---|---|---|
| Attribution | Allocate credit for an observed conversion | The model is not a control group |
| Lift study | Estimate an additional effect | Protocol and assumptions determine interpretation |
Management indicators
| Indicator | What it measures | First action |
|---|---|---|
| Deduplicated conversions | Unique business actions | Reconcile overlapping reports |
| Estimated effect | Protocol-based difference with uncertainty | Retain limitations before allocation |
Common pitfalls
- Add conversions claimed by several platforms
- Reallocate budget without documenting the control
Frequently asked questions
Does high ROAS prove profitability?
It relates attributed value to spend. Margin, other costs, returns and incremental effects need separate checks.
Can all platform reports be added together?
Not without checking duplicates and windows: several platforms can claim the same conversion.
Should a minimum gain be promised?
No. Define a question, protocol and decision rule; measurement may remain inconclusive.
Official references
References consulted: . The method and worksheet propose checks to adapt to your context; they do not constitute certification.
Continue exploring
Related analysis
