The Last-Click Attribution Trap
Last-click attribution assigns credit to the final recorded touchpoint. It is easy to explain, but it can understate earlier interactions—especially when a customer journey spans channels or devices.
Real Example: The Hidden Journey
A user's actual journey to install your app:
Day 1: Sees a social ad, taps, and leaves without installing.
Day 3: Returns through a second campaign touch.
Day 5: Searches for the app and installs.
Last-click report: credits the recorded search touch.
What it cannot establish: whether earlier touches caused or assisted the decision.
Multi-Touch Attribution Models
1. Linear Attribution
Splits credit evenly across included touchpoints. This is transparent, but assumes each touch contributes equally.
2. Time-Decay Attribution
Weights recent touchpoints more heavily according to a chosen decay rule. The chosen window and weighting are assumptions to test, not universal truths.
3. U-Shaped (Position-Based)
A position-based heuristic assigns more credit to the first and last touches, with the remainder shared between the middle touches. The 40/40/20 split is one convention—not an industry standard or a causal measurement.
When to Use Each Model
| Model | Best For | Example Use Case |
|---|---|---|
| Last-Click | Direct response | Flash sales, limited offers |
| Linear | Long sales cycles | B2B apps, enterprise |
| Time-Decay | Recency matters | E-commerce, travel |
| U-Shaped | Most apps | Consumer apps, gaming |
Use attribution models as decision aids
Before comparing models, define which events count as touches, the lookback window, how duplicate events are handled, and which conversion outcome receives credit. Document these choices so reports can be interpreted consistently.
Compare model outputs with incrementality experiments or other causal evidence where possible. A model redistributes observed credit; it does not prove that a touch caused the conversion.
Evaluate a model change
When a different model changes channel credit, compare the underlying touch data, coverage, and conversion totals before changing budget. Treat any lift in reported credit as a measurement change until an experiment shows an incremental business effect.
Conclusion
Multi-touch models can make assumptions visible and help teams explore journeys that a last-touch report hides. Use them alongside platform-level constraints, data-quality checks, and experiments—not as a complete or causal record of every customer journey.