Tags: analytics commerce concept

Attribution Windows

Date: 2026-08-16

---

How far back a touchpoint counts. It’s a modelling parameter presented as a settings field — lengthening it manufactures upper-funnel credit, shortening it hands everything to the last click, and neither is more true.

---

What it is

An attribution window is the lookback period during which a touchpoint remains eligible for credit. Anything older is discarded before any model is applied.

The window runs before the model, which is why it’s the more powerful of the two settings and gets a fraction of the attention.

purchase at day 0

  ← 30-day window ──────────────────────────────┤
       paid search    email    organic   direct
       day −21       day −6   day −2    day 0
       ✓ eligible     ✓        ✓         ✓

  ← 7-day window ───────────────┤
                    email    organic   direct
       ✗ discarded    ✓        ✓         ✓

Same purchase, same behaviour. Under the shorter window paid search never existed, so first-click credit moves to email.

The two kinds

• Click-through window — how long after a click a conversion counts. Typically 7 to 30 days
• View-through window — how long after an impression. Typically 1 day, and far more contentious — View-Through Attribution

They’re usually configured separately and should be. A 30-day view-through window credits an ad someone scrolled past a month ago.

Choosing one

Two constraints, and they pull opposite ways.

The consideration cycle. Impulse purchases resolve in a session; a considered purchase may take weeks. A window shorter than your real cycle systematically discards the top of the funnel. Measure the actual gap between first touch and purchase from your own data — the median and the 90th percentile — and set the window from that rather than from a default.

What your identity can support. This is the constraint most people miss. A 30-day window is meaningless if identifiers survive seven days:

window                30 days
Safari cookie cap      7 days   ← the real limit
effective window       7 days

You’ve configured a 30-day lookback and you’re running a 7-day one for a large share of traffic, silently and unevenly by browser. See Browser Privacy Restrictions and Identity Stitching.

In plain terms: the window can’t be longer than your memory. Setting it longer just means the number in the settings field is aspirational.

Why it’s the biggest lever

Because it decides which touchpoints exist, and every model then divides credit among survivors.

Lengthen it and upper-funnel channels — paid social, display, video — gain credit, because more of their impressions and clicks fall inside. Shorten it and everything concentrates on the closers: branded search, direct, email.

Which means a change to the window rewrites every channel’s return on ad spend simultaneously, with no change in performance. If channel budgets are set from attributed revenue, the window is a budget-allocation setting. It’s rarely reviewed and almost never annotated.

Practical rules

• Set it deliberately, once, from your own consideration cycle — and record why
• Never change it mid-analysis. Restate history under both windows before anyone sees a number, or the change reads as performance — Metric Drift
• Compare windows as a diagnostic, not as a search for the flattering one. A channel gaining a lot between 7 and 30 days is an opener; one that’s stable is a closer — same logic as comparing first and last click in Attribution Models
• Never compare across platforms. Each ad platform uses its own window and its own model on its own data. Their numbers won’t reconcile with yours or each other’s, and that’s expected — Walled Garden Reporting
• Annotate any change everywhere the number appears — Annotation and Change Logs