Every marketing report is a story about who caused a sale, and attribution models are the narrator's voice. Choose the narrator carelessly and you will defund the campaigns that create demand while overfunding the ones that harvest it. Here is the honest tour of the options.
The classic models and what each one rewards
| Model | Gives credit to | Optimizes for | Typical bias |
|---|---|---|---|
| Last click | Final touchpoint | Closing channels | Brand search, retargeting, coupons |
| First click | Initial touchpoint | Acquisition channels | Cold prospecting overvalued early |
| Linear | Every touch equally | Nothing in particular | Avoids hard choices by making none |
| Time decay | Touches near conversion | Assisted demand capture | Similar to last click, slightly kinder |
| Position-based | First + last, some to middle | Acquisition + closing | Reasonable, arbitrary weights |
| Data-driven | Statistical model of what predicted conversion | Actual influence (in theory) | Needs volume; a black box at low data |
The uncomfortable truth about data-driven attribution
DDA is the best model in principle and the most misused in practice. It requires conversion volume to train, degrades gracefully into the modeled black box nobody can explain at low volume, and — like every model — reflects the channels it can see. If your offline sales calls and WhatsApp closes live outside your analytics, even perfect DDA attributes a partial journey.
A practical policy that works
- Pick one model as your reporting standard (for most: data-driven in GA4 at volume, position-based below it) and say so in every report footer.
- Read last-click alongside it as a floor test — if a campaign fails both lenses, it is really failing.
- Separate budget decisions from reporting: allocate budget on marginal CPA and incrementality tests, not attribution share alone.
- Run one incrementality test per quarter (geo holdout, ghost bids, or platform experiments) — attribution tells you what happened; experiments tell you what caused it.
Cross-platform reality
Each platform claims conversions its siblings also touched. Reconcile with a hierarchy: business outcomes (CRM revenue) > analytics (GA4 with disciplined UTMs) > platform-reported (directional). Discrepancy of 20-40% between platform totals and analytics is normal; discrepancy in trend direction is the alarm. More in cross-platform attribution.
Which attribution model should we use?
Data-driven if you have volume; position-based as a transparent fallback; last-click as a sanity floor. The model matters less than the discipline of committing to one and testing incrementality quarterly.
Why don't platform numbers match GA4?
Different lookback windows, different identity models, different definitions of a conversion — plus each platform credits itself. Use platforms for optimization, analytics for reconciliation.
Attribution fights are budget fights with extra steps. Our tracking & analytics service builds the reconciled view. Talk to us.