Conversion Metrics

• Define macro and micro conversions for a product
• Compute overall and step conversion with explicit denominators
• Avoid the denominator traps that corrupt conversion comparisons

• Micro conversions predict macro ones — cart-add rate leads purchase rate and is measurable sooner.
• Rate and count answer different questions: rate tells quality, count tells scale.
• Denominator choices (all visitors vs qualified vs session-entering) change a 3% into a 30%.
• Rate comparisons across tests are only valid with identical definitions.

Getting handed a menu (micro) is a faint signal; the bill closed with a card (macro) is the number that pays the rent. Tracking the menu-handing tells you the bill may follow.

• Name the macro conversion (purchase) and 1–3 micro conversions on its path (search use, add to cart, saved item).
• Define entrants: users who entered the funnel start in the analysis window; apply dedupe and identity rules.
• Rate = unique converters / unique eligible entrants within the window.
• Report step conversion alongside overall so friction inside the funnel is visible.
• Normalize across time periods (daily/weekly) and state seasonality assumptions.

• A checkout team tracks 'add-to-cart' micro conversion because it leads purchase rate by two weeks and pages early on landing-page changes.
• An advertising team reports CVR by campaign with an explicit 'qualified entrant' denominator, ending inflated cross-campaign comparisons.
• A SaaS board reviews macro conversion (paid start) while product reviews micro conversions (team invite sent) as its leading indicator.

• Document the denominator and window beside every rate; an unqualified rate is meaningless.
• Prefer unique converters over conversion events for rate — retconverted users double count.
• Benchmark rates on rolling baselines so seasonality is visible.
• For experiments, read rates with CIs, not point estimates alone.

• Dividing by all site visitors when only search-or-qualified users enter the funnel — the rate is fabricated low.
• Changing the window between reports until the number lands where you want.
• Using event counts for converters without dedup (a refunded purchase still counts if dedupe is on events).
• Reporting conversion rate up while converter count fell (denominator shrank), missing the real story.

• A falling rate with a rising count means the denominator outgrew converters; report both.
• Recompute the rate from raw events with one explicit filter set when two tools disagree.
• Cart-add micro conversions lag after a tracking change; verify the event fires per product, not per render.
• Cross-check CVR against the payments ledger when macro is purchase — the ledger is truth.