Funnel Maturity

Why a steps query over the entrant window is wrong

The funnel dateRange filters the events of every step, not only the entry step. A steps query with a dateRange of one entrant week ignores every conversion after the week ends. A person who enters on the last day of the week then gets less than one day to convert, whatever the saved interval is. Do not score a funnel from a steps query bounded to the entrant window.

The trends-mode query

Trends mode groups entrants by entrant period, and counts each entrant’s conversions inside the conversion interval, also after the period ends.

• Copy the saved funnel’s series, funnelOrderType, funnelWindowInterval, funnelWindowIntervalUnit and filters into the query. Do not change the measured flow.
• Set funnelsFilter.funnelVizType: trends and interval: week (use day only for a daily cadence).
• Set a dateRange that ends now. Start it at the beginning of an entrant period that is at least one conversion interval, the buffer and N+2 entrant periods before now. Then the latest mature point and N mature baseline points remain after you drop the immature points, and the first point is a full period.
• With the exec dispatcher, run call --json query-funnel <json_input>. Put --json before the tool name; do not add output_format to the JSON input. The default optimized output shows only rates and no entrant counts.
• Each result holds days, data (conversion %), reached_from_step_count (entrants) and reached_to_step_count, with matching indexes.
• Each point is one entrant period. Score only the points whose entrant period ended at least one conversion interval, plus the buffer, before now. Use the older mature points as the baseline.
• With interval: day, compare a point only to baseline points on the same weekday.
• Use reached_from_step_count for the volume gate, the steady-denominator check and the report evidence.

Step-to-step rates

Each trends point is the first-to-last rate, and query-funnel has no step selector. Build each step rate from prefix funnels instead, with the same trends query:

• Step 1 → 2: data of the prefix funnel of steps 1..2. Do not query a one-step prefix, because query-funnel rejects a funnel with fewer than two steps.
• Step k-1 → k, for k of 3 or more: query the prefix funnels of steps 1..k-1 and 1..k. Divide reached_to_step_count of the longer prefix by that of the shorter prefix at the same index. For this rate, the denominator is reached_to_step_count of the shorter prefix. Use it for the volume gate and the evidence. Skip a point where it is zero.
• Prefix rates hold only when funnelOrderType is ordered or strict. For an unordered funnel, score only the first-to-last rate.

Funnels this method cannot score

• Exclusions. In trends mode, one fully excluded attempt removes that person’s failed entries from every period in the dateRange. Older baseline points have more later attempts, so their conversion can be too high. Do not author a decline for a funnel with exclusions from trends results.
• Optional steps. A step marked optionalInFunnel is valid only in a steps funnel, so the trends query fails. Do not remove the optional setting, because that changes the measured flow. Do not fall back to a steps query bounded to the entrant window.

Name such a flow as unscored, with the reason, in the watchlist entry and the run summary.

Worked example

A saved ordered funnel Signup → project created → first event, with a 14-day conversion interval and a weekly cadence. Now is Wed 24, and weeks start on Monday.

• The cutoff is Wed 24 minus 6 hours. The newest mature week is Mon 1 – Sun 7, because Sun 7 + 14 days ends on Sun 21. The week Mon 8 – Sun 14 matures only after Sun 28.
• With N = 4 baseline weeks, start the dateRange on a Monday at least 14 days, 6 hours and 6 weeks before now, and end it now.
• Run the prefix funnel Signup → project created and the full funnel in trends mode with call --json query-funnel <json_input>.
• Read the points for weeks that start Mon 1 and earlier. Drop the points for Mon 8, Mon 15 and Mon 22, which look low only because their entrants are still converting.