Funnel Segmentation

Why this matters

Segmentation turns a single funnel into many focused views so you can see who converts, where they drop, and why. Product Analysts use segmented funnels to:

• Prioritize fixes: e.g., identify that mobile Safari users fail at payment step.
• Allocate budget: e.g., see which acquisition channels bring high-intent users.
• Tailor onboarding: e.g., new users vs. returning users need different guidance.
• Measure experiment impact across cohorts: avoid hiding effects in averages.

Concept explained simply

A funnel is a sequence of steps users complete to reach a goal (e.g., Visit → Add to cart → Purchase). Segmentation means splitting that funnel by attributes (source, device, country, cohort, or behavior) to compare conversion at each step.

Mental model

Imagine a set of parallel lanes flowing through the same pipe. Each lane is a segment. You don’t just compare lane sizes at the end; you watch where each lane leaks. The biggest differences between lanes point to the most actionable fixes.

Who this is for

• Product Analysts and Growth Analysts who diagnose drop-offs.
• PMs who need evidence to prioritize fixes.
• Data-savvy marketers focusing on channel quality.

Prerequisites

• Know how to define a funnel clearly (events, order, and time window).
• Basic SQL or familiarity with product analytics tools.
• Understanding of user vs. session scope.

Learning path

1. Define the base funnel clearly (events, order, window, unit of analysis).
2. Add one segmentation dimension at a time (e.g., device), then layer more if needed.
3. Quantify differences, search for root causes, and propose targeted actions.

Worked examples

How to run a segmented funnel (step-by-step)

1. Define your base funnel
   • Steps: name each required event and its order.
   • Window: choose the time allowed between steps.
   • Scope: user or session. Stay consistent across steps.
2. Choose one segmentation dimension
   • Start with high-impact: device, channel, country, browser, cohort (e.g., signup month), or behavior (e.g., first-time vs. returning).
3. Compute per-step and overall conversion per segment
   • Per-step i: count(step i) / count(step i−1).
   • Overall: count(final) / count(start).
4. Compare segments and locate the biggest relative drop
   • Look for the step with the largest gap between best and worst segments.
5. Validate sample sizes and noise
   • Avoid overinterpreting tiny segments; combine or extend timeframe.
6. Form a hypothesis and propose an action
   • UX fix, targeting change, copy update, performance improvement, or A/B test.

Exercises (do these before the Quick Test)

These mirror the exercises below so you can check answers.

-- Pseudo-SQL outline
WITH first_view AS (
  SELECT user_id, MIN(event_time) AS fv_time
  FROM events
  WHERE event_name = 'view_product'
  GROUP BY user_id
), scoped AS (
  SELECT e.user_id,
         e.event_name,
         e.event_time,
         e.device,
         fv.fv_time
  FROM events e
  JOIN first_view fv USING (user_id)
  WHERE e.event_time BETWEEN fv.fv_time AND fv.fv_time + INTERVAL '3 day'
), steps AS (
  SELECT user_id,
         MIN(CASE WHEN event_name = 'view_product' THEN event_time END) AS s1,
         MIN(CASE WHEN event_name = 'add_to_cart' THEN event_time END) AS s2,
         MIN(CASE WHEN event_name = 'checkout' THEN event_time END) AS s3,
         MIN(CASE WHEN event_name = 'purchase' THEN event_time END) AS s4,
         ANY_VALUE(device) AS device -- device at first event or most frequent
  FROM scoped
  GROUP BY user_id
)
SELECT device,
  COUNT(*) FILTER (WHERE s1 IS NOT NULL) AS step1,
  COUNT(*) FILTER (WHERE s2 IS NOT NULL) AS step2,
  COUNT(*) FILTER (WHERE s3 IS NOT NULL) AS step3,
  COUNT(*) FILTER (WHERE s4 IS NOT NULL) AS step4,
  ROUND(COUNT(*) FILTER (WHERE s2 IS NOT NULL)::numeric / NULLIF(COUNT(*) FILTER (WHERE s1 IS NOT NULL),0), 3) AS rate_1_to_2,
  ROUND(COUNT(*) FILTER (WHERE s3 IS NOT NULL)::numeric / NULLIF(COUNT(*) FILTER (WHERE s2 IS NOT NULL),0), 3) AS rate_2_to_3,
  ROUND(COUNT(*) FILTER (WHERE s4 IS NOT NULL)::numeric / NULLIF(COUNT(*) FILTER (WHERE s3 IS NOT NULL),0), 3) AS rate_3_to_4,
  ROUND(COUNT(*) FILTER (WHERE s4 IS NOT NULL)::numeric / NULLIF(COUNT(*) FILTER (WHERE s1 IS NOT NULL),0), 3) AS rate_overall
FROM steps
GROUP BY device
ORDER BY device;

Common mistakes and how to self-check

• Mixing units (user vs. session). Self-check: Are denominators consistent across steps?
• Changing definitions mid-analysis. Self-check: Are event definitions identical across segments/time?
• Ignoring tiny sample segments. Self-check: Flag segments below your minimum sample threshold.
• Simpson’s paradox. Self-check: If overall rates disagree with segmented rates, check for hidden confounders (e.g., device mix differs by source).
• Wrong window. Self-check: Confirm time windows reflect realistic behavior and are the same across segments.
• Attribution drift. Self-check: Ensure each user/session has one stable source value for this analysis.

Practical projects

• Mobile checkout deep-dive: Segment by device and browser; capture error codes; propose top 2 fixes with expected impact.
• Channel quality score: Build a segmented funnel from Landing to Activation by source/medium; compute weighted score to inform budget allocation.
• Onboarding simplification: Segment new vs. returning users; run a guided checklist variant for new users and measure changes at the sticky step.

Mini challenge

Your overall purchase rate is flat at 10%, but after segmenting by country you see: Country A 14%, Country B 6%. Payment failures are 3% in A vs. 12% in B. In one paragraph, state a hypothesis and the first fix you will prioritize. Keep it specific (e.g., payment provider routing, currency display, 3DS flow on mobile).

Next steps

• Apply one segmentation to a live funnel and document your top 3 insights with suggested actions.
• Share with your PM/engineering partner; pick one fix to ship or A/B test.
• Then take the Quick Test to check your understanding. The Quick Test is available to everyone; only logged-in users get saved progress.