The short answer
Define the eligible unit, required steps, order, entry period and completion window, then inspect the joined permitted event path. Repeated events and separate dashboard totals are not a person-level funnel. Keep a verified final outcome separate from a checkout click, and preserve legitimate outcomes outside the particular path being reviewed.
What you’ll take away
- Use a matched path, not a ratio of unrelated totals.
- Choose only steps that are required for the question.
- A missing event does not identify why someone stopped.
Define the actual journey and completion
Choose a narrow question, such as whether eligible aid users complete the required aid-to-CTA-to-verified-outcome path. State the unit and final evidence. A valid enquiry, checkout start and verified payment are different results; do not label them all conversion without explaining which is meant.
Decide which steps are genuinely required. Someone may legitimately purchase without using a tool, so a tool-specific funnel is not the whole business ledger. Use only authorized existing identity/event relationships, preserve consent and exclusions, and keep outcomes on other paths rather than treating them as nonexistent.
Match the unit, order and repeated-event rule
PostHog distinguishes sequential order with intervening events allowed, strict order without them, and any order. Read the actual choice. A payment before the recorded CTA cannot complete a declared forward path merely because both event names exist for the same unit.
Deduplicate at the declared unit and stage, not by similar text or a convenient aggregate. Several checkout events from one unit do not create several entrants. A missing identifier, blocked collection or unavailable backend receipt may leave a path unmeasured; do not fingerprint the reader or infer a payment from a browser return page.
Keep entry, completion and reporting cutoff distinct
A conversion window limits time after entry; the report’s date range can impose another cutoff. Current PostHog documentation says the date range applies to every step. A completion can be within seven days of entry yet excluded because it occurs after the report ends. Preserve entry filtering and a complete observation horizon.
Keep overall conversion relative to entry separate from conversion relative to the previous step. In the constructed ledger, six entrants yield four qualifying tool users, three qualifying CTAs and one within-window verified outcome. Overall final conversion is 1/6≈16.67%; final-step conversion is 1/3≈33.33%. Neither is a count of all customers.
Resolve three constructed path audits
F1 — Constructed six-unit ledger: A completes entry/tool/CTA/outcome; B reaches CTA; C has an outcome without the required tool event; D reaches tool; E only enters; F reaches CTA but completes after the window. Decision: retain the 6→4→3→1 required path and report C’s other path and F’s later outcome separately. No genuine outcome is erased to tidy the funnel.
F2 — Constructed cutoff: entry is 30 October, completion 2 November, a seven-day window is allowed, but the report ends 31 October. Decision: record the report cutoff, extend the observation horizon when legitimately available and keep entry restricted to the intended cohort. Outcome: no false claim that the unit never completed.
F3 — Constructed repetition: one eligible unit triggers checkout three times. Decision: count one qualifying unit at that stage under this declared unique-unit funnel, preserve the event evidence and reconcile the actual outcome independently. Outcome: not three people, payments or new customers.
Use a drop-off to choose a source check
A lower stage count can suggest a question about destination, instructions, access or measurement. It does not by itself establish friction, disinterest, an error or causal harm from a recent release. Inspect the actual authorized path and a specific reproducible condition before prescribing a product change.
Breakdown attribution can also change the counted subset. PostHog’s first/last/specific-step property choices concern where a breakdown value comes from, not necessarily marketing-credit attribution. Some modes can put one unit into more than one breakdown. Retain that scope before adding rows as distinct users.
Keep the path contract and verified outcome
The worksheet keeps unit/eligibility, required events, order, identity limits, entry dates, report cutoff, completion window, deduplication and source-verified outcome. Mark incomplete observation and excluded QA separately. An actual payment report also needs its appropriate refund and customer-deduplication treatment.
Current primary funnel guidance was read on 1 November 2026. The entire six-unit ledger is constructed, not this blog’s real funnel, customers or conversion rate. No instrumentation, account, identity, payment source or actual pilot was queried or changed.
Sources & product references
Product details reflect the publication date. External features and interfaces may change.
