Measurement · 04 Demo · synthetic data
Attribution comparison
Who gets credit, and does it add up?
The same synthetic customers credited three ways: last-click, data-driven (Shapley), and an MMM-style incremental view. Plus a reconciliation of paid, affiliate, and creator claims against real orders.
Attribution comparison
60,000 synthetic shoppers, 1,153 orders. Same data, three answers.
| Channel | Last-click | Data-driven (Shapley) | MMM-style incremental | Comparison |
|---|---|---|---|---|
| Paid search | 472 | 299 | 106 | |
| Paid social | 255 | 261 | 293 | |
| Affiliate | 225 | 58 | 11 | |
| Creator | 76 | 40 | 55 | |
| Would have bought anyway | 125 | 495 | 688 |
Shoppers with high purchase intent are more likely to click a search ad or an affiliate coupon on the way to checkout, so last-click and even data-driven credit give those channels sales they didn't cause. In the simulator, the incremental view is the known truth. In practice an MMM or a lift test estimates it. Last-click's “would have bought anyway” row is orders with no paid touch.
Reconciling paid, affiliate, and creator claims
Overlap rule
Social claims
Platforms, affiliate networks, and creator codes claim 1,721 orders. The order system booked 1,153.
Claims add up to 149% of real orders. De-duplicating by order ID brings the total back to what was booked.
| Source | Claimed | De-duplicated | Over-claim |
|---|---|---|---|
| Paid search | 618 | 472 | 1.31× |
| Paid social | 724 | 266 | 2.72× |
| Affiliate | 237 | 225 | 1.05× |
| Creator | 142 | 53 | 2.68× |
| No paid claim | — | 137 | — |
| Total | 1,721 | 1,153 | 1.49× |
De-duplication fixes the double count. It doesn't say which touch caused the sale. Compare the de-duplicated column with the incremental view above: the rule you pick moves credit around, and none of the rules recover the truth on their own.
Synthetic numbers generated in your browser. Not client or employer data.
How to read it
What the demo is showing
Last-click rewards whoever touched the customer last, which usually favors search, affiliate coupon sites, and creator codes.
Data-driven credit (Shapley values over the synthetic paths) shares credit across the channels in a path based on how conversion rates change when a channel is present.
The MMM-style view only credits what was incremental, so it also shows how many orders would have happened anyway. In this simulator that view is the ground truth, which is why it's the one to steer by.
How I'd reconcile platform numbers with cashNext demo: Executive MER dashboardAll demos