The Segment That Won't Match
An analyst builds a segment called "High-Value Repeat Buyers" defined as: Purchases >= 3 AND Revenue >= 500, scoped to Visitor, with a date range of "Last 90 days." The segment shows 12,400 visitors. The client's CRM says there are 31,000 customers meeting those criteria in the same period. Separately, the analyst builds a calculated metric "Revenue per Visitor" as Revenue / Unique Visitors and notices it produces unexpectedly high values on some pages, and that a "Cart Abandonment Rate" metric (Cart Abandons / Cart Additions) sometimes exceeds 100%. The client is losing confidence in the numbers.
1. Why does the segment undercount versus CRM, and what would you check?
Read the worked answer
The segment is scoped to Visitor, but the criteria (3+ purchases, $500+ revenue) are visit- or hit-level facts aggregated per visitor. The key issue is scope and container logic: a Visitor-scoped segment with hit-level conditions requires the conditions to be satisfied by hits within the segment's date range, and the visitor must have qualifying hits. Common undercount causes:
- Date range interaction. "Last 90 days" limits which hits are evaluated. A customer whose 3rd purchase happened 95 days ago, or whose purchases straddle the boundary, is excluded even though CRM counts them in the period. Confirm the segment's date range versus the reporting window.
- Identity gap. Visitors who purchase in-store or via a channel not tracked in Analytics are invisible. CRM counts them; Analytics cannot.
- Bot/internal traffic filtering removing some visitors.
- Attribution of the segment condition - whether the conditions must co-occur on the same hit vs. anywhere in the visitor's history within range.
Check: Compare a Visitor-scoped segment to a Visit-scoped version, test the same logic with an expanded date range, and reconcile a sample of CRM customer IDs against Analytics visitor IDs to quantify the identity gap. The 12,400 vs 31,000 gap is large enough that identity coverage is likely the dominant factor, not segment logic alone.
2. Explain the odd calculated-metric results and how to correct the metric definitions.
Read the worked answer
Both metrics suffer from mixing metrics of different scopes/denominators.
Revenue per Visitor: Dividing a summable metric (Revenue) by Unique Visitors is valid only when the denominator is counted at the same granularity as the report. On a page-level report, Unique Visitors is de-duplicated per page, so a visitor who sees 5 pages is counted once per page - the ratio is mathematically fine per row but the totals will not sum, and rows can look inflated relative to a visitor-level expectation. If the analyst expected a visitor-level ratio, the correct approach is to use a Visitor-scoped context or to compute the ratio at the report level the client actually wants. Confirm whether the intent is "revenue per unique visitor" (row-level) or "revenue per visitor across the site" (site-level); they are different metrics.
Cart Abandonment Rate > 100%: Cart Abandons and Cart Additions are independent counters. If a visitor abandons a cart and later adds items again, abandons can exceed additions in a period, or the events can be counted on different hit types. The fix is to define abandonment as a ratio of abandons to carts opened with both events counted on the same trigger, and to verify both events fire exactly once per cart lifecycle. Test by pulling both raw event counts side by side; if abandons exceed additions, the event logic is wrong, not the math.