How do I compare attribution models?

The Attribution models page shows the same revenue counted four ways — "Last click", "Linear", "Position-based (40/20/40)", and "Time-decay (7-day half-life)" — side by side, so you can see how credit shifts without changing any stored number.

Before you start

• A workspace with the Attribution module active.
• At least a few tracked conversions in the date range you want to compare — the models need conversions to redistribute credit.

Steps

1. Open Attribution Models

   Use Sidebar → Attribution → Attribution Models. The page is titled "Attribution models".

2. Pick a date range

   Use the date range picker next to the model selector. The comparison defaults to the last 30 days.

3. Read the comparison table

   Under "How each model values the same conversions", each row is a model with its "Conversions", "Attributed revenue", "Spend", and "ROAS". "Last click" carries a "Current" badge — it is the model your live reports use today. The cards above the table show how much each multi-touch model moves revenue versus last-click.

4. Click a row to learn what the model means

   Selecting a row (or using the dropdown) opens a plain-language description of that model and its tradeoff — for example, "Time-decay (7-day half-life)" gives recent touches more weight but hides slow-burning channels.

5. Recompute if multi-touch rows are empty

   If last-click shows revenue but the other rows are zero, multi-touch credit hasn't been calculated for this window yet. Press "Recompute unattributed" in the banner — a "Recomputing…" badge appears next to the comparison title while it runs, and the table fills in when it finishes. (The "Recompute the window" card at the bottom of the page has the same button and shows a "Recompute queued" confirmation instead.)

What happens next

• Switching models on this page only changes the view — it redistributes credit across the same conversions and the same total revenue. Nothing here inflates or rewrites your sales.
• Your stored results stay reproducible: every conversion is stamped with the model version that attributed it, so historic reports keep showing the numbers they showed at the time, even as you explore other models.
• "Recompute the window" only touches conversions that were never attributed. It is safe to run repeatedly and never double-credits a conversion that already has an answer.

Gotchas