CONVERSION

Funnel drop off diagnosis

Use when you have step-by-step funnel numbers and need to decide where to look first.

funnel-drop-off-diagnosis.md

Download .md

You are diagnosing a conversion funnel from step data. Be careful to separate the biggest drop from the most fixable drop.

FUNNEL STEPS AND VOLUMES: {{STEP_VOLUMES}}
TIME PERIOD: {{PERIOD}}
SEGMENT DEFINITION: {{SEGMENT}}
KNOWN CHANGES OR CAMPAIGNS IN THE PERIOD: {{KNOWN_CHANGES}}
BENCHMARK OR PREVIOUS PERIOD IF AVAILABLE: {{BENCHMARK}}

Produce:
1. A table: Step | Entries | Exits | Step conversion rate | Absolute users lost | Notes.
2. The step with the largest absolute loss, and separately the step with the worst rate. Explain why these may not be the same step and which matters more here.
3. For the top two problem steps, list plausible causes split into: measurement problem (tracking, definition, duplicate events), intent problem (wrong traffic arriving), and experience problem (something on the page).
4. For each plausible cause, the cheapest way to confirm or rule it out. Order by cost, cheapest first.
5. What you would check before concluding anything is broken - seasonality, campaign mix, tracking changes in the period.

Constraints: do not assume a large drop is a problem if that step is a natural qualifier. Do not recommend a redesign or a test until the measurement checks in point 5 are done. If a step rate looks impossible, say the data is probably wrong rather than explaining it. No em dashes.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{STEP_VOLUMES}}
{{PERIOD}}
{{SEGMENT}}
{{KNOWN_CHANGES}}
{{BENCHMARK}}

Getting a better result

Always give the previous period, otherwise you cannot tell a problem from a normal rate.
The measurement-problem branch resolves more funnel mysteries than the experience branch.
Note any qualifying steps explicitly, or you will chase a drop that is doing its job.

Questions about this prompt

When do I diagnose the funnel rather than fix the worst step?

Before you pick a step to work on. It separates the largest absolute loss from the worst conversion rate, which are frequently different steps, and it will not recommend a redesign or a test until the measurement checks are done. Run it ahead of the checkout or demo booking audits rather than after them.

What do I need in front of me before running it?

Step volumes, the time period, the segment definition, any campaigns or releases in the window, and the previous period as a benchmark. Without the benchmark you cannot tell a problem from a normal rate. Say explicitly which steps are natural qualifiers, or you will spend a fortnight chasing a drop that is doing its job.

What comes back, and which part is worth acting on?

A step table, the largest absolute loss set against the worst rate with an argument about which matters here, then plausible causes split into measurement, intent and experience problems, each with the cheapest way to rule it out. Work the measurement branch first, since tracking definitions and duplicate events resolve more funnel mysteries than page changes.

What is the mistake that costs me here?

Explaining an impossible step rate instead of doubting it. Where a rate cannot be true the data is probably wrong, and the prompt says so rather than building a theory on it. The other cost is skipping point five, because seasonality and campaign mix explain a good share of what gets escalated as a broken funnel.