What this checkpoint covers
The second module of the intermediate analytics level: what an AttributionAssigning conversion credit to touchpoints. model decides and what it cannot, how models and lookback windows change the numbers, and the causes of the gap between platform and analytics figures. It builds on the attribution models exercise, where you applied the rules by hand.
Task
- Answer every question in a sentence or two before reading the explanations, giving your reason.
- Compare your reasoning with each explanation and note what you missed.
- Finish by writing one sentence that explains a platform and analytics discrepancy to a non-specialist. A sample sentence follows the answers; write yours before reading it.
Questions
- The finance director asks which attribution model is the correct one, so that the argument can stop. What do you answer?
- A paid social platform reports far more conversions than the paid social channel in GA4Google's event-based analytics platform.. The head of marketing assumes the tag is broken. What do you check before touching the tag?
- Adding up the conversions reported by every ad platform gives more than the number of orders in the order system. Is something wrong?
- Mid-year, the team switched the analytics reporting model. Now they want to compare the first half with the second. What do you advise?
- A channel earns much more credit under data-driven attribution than under last click. The channel manager asks to double its budget on that basis. What do you say?
- You need to reconcile an ad platform's figures with analytics. Which attribution view do you set on each side?
- On Monday morning you compare last week's platform conversions with analytics. By Thursday the platform figure for the same week has risen. Why?
- The gap between platform and analytics conversions is much wider in one market than in the others. Campaign setup is identical. Where do you look first?
Answers explained
- None is correct; the right model is the one that fits the decision. Each model makes one part of the path visible and hides the rest, and every model sees only tracked touchpoints. The attribution models framework turns the question into which decision the number is for, then fixes one model for business reporting, written down and dated.
- Whether the two numbers are measuring the same thing. Check view-through conversions, the platform's window against the analytics model, consent and modelled conversions, and whether you are comparing like units. The causes table in the guide to platform and analytics discrepancies is the order to work through; a tag fault is only one row of it.
- No. Each platform credits the conversions its own ads touched, and several platforms touched the same orders. The order system is the complete count; platform totals are claims, not shares of it.
- Restate the first half under the new model, or report the two halves as separate series. Splicing two rules together breaks the series at the change date, and the attribution models comparison recommends restating history rather than mixing models.
- The shift in credit is an assist signal, not proof. Data-driven credit is estimated from observed paths, so a channel reaching people already on their way to buy can still earn it. Before moving budget, run an IncrementalityConversions caused by marketing. test that withholds the channel from a comparable group.
- Last click, click-only, with matching windows where the tools allow. It is the one rule every tool can produce in the same shape. Record whatever could not be matched, and treat the residual as views and shared credit.
- Many platforms report a conversion on the date of the click that earned it, and keep attributing for days. Analytics reports the event on the day it happened and finalises on its own schedule. Reconcile complete periods only, after both systems' documented processing has finished, as the discrepancies guide sets out.
- The consent rate in that market. Declined consent removes visitors from analytics, while the platform may still match or model the conversion. A low consent rate widens the gap without any change in campaign performance.
Where to go next
If you missed the model questions, rework the attribution models exercise with a different rule. If you missed the discrepancy questions, follow the reconciliation method in the discrepancies guide for one platform you work with. Then start the third module with the funnel exploration exercise.
