funnel-diagnosticslisted

> **Provenance.** great-pm-original, authored 2026-05-29, grounded in the > cited sources (Plane, FullSession, Datadog Product Analytics). Host > agent: analytics-analyst. # Funnel Diagnostics — locate, explain, fix, re-measure A funnel is **diagnostic, not descriptive — but only once you segment it.** An aggregate funnel tells you *where* users drop. That is necessary and almost useless on its own: "60% drop at the upload step" prompts a quarter of guesswork. The leap from *description* to *diagnosis* is **segmentation + session review** — splitting the drop by device, source, plan, cohort until the loss concentrates in a group, because the group where the drop concentrates *is the diagnosis*. Then you prioritize the drop that blocks value for the **largest group**, fix it, and **re-measure the specific step** to confirm the curve moved. The single biggest failure in funnel work is reading the aggregate number, "fixing onboarding" generically, and shipping a change that moves nothing because the real loss was a PDF-parser failure for one bank, invisible in the average. ## 1. The 5-step workflow ### Step 1 — Define the journey Write the *ordered* steps that constitute the conversion. Each step is a discrete, instrumented event with an unambiguous fire condition (see `metrics-design` §4 — event names, properties, triggers, dedup). Rules: - **Order matters and must be real.** Steps should be sequential dependencies, not parallel activities lumped together. - **Pick the