What is Fallout?
Fallout is the number or rate of users/sessions that leave a defined path before reaching the next step. Example path: Search submitted → Results viewed → Product clicked → Added to cart → Checkout started → Purchase.
Rate formula (step i → i+1): fallout_rate = 1 − (progressions_from_step_i / entrants_to_step_i).
How Fallout analysis works (quick)
- Define steps: Clear, ordered events (e.g.,
search_submit, results_view, pdp_click, add_to_cart, checkout_start, purchase). - Choose unit: Sessions for UX changes; users for longer trends; keep them separate.
- Measure transitions: Count entrants per step and progressions to the next; compute fallout and throughput for each gap.
- Segment: By query intent, device, geo, traffic source, latency, zero-results, and stock state.
- Compare cohorts: A/B tests, new synonyms, relevance releases, pricing/shipping experiments.
Why it matters in e-commerce
- Pinpoints leaks: Shows exactly where shoppers abandon (e.g., results → PDP gap).
- Prioritizes fixes: Tie fallout spikes to zero-results, slow ART/TTFB, weak relevance, facet friction, OOS, or sticker shock.
- Links to money: Small fallout drops on results → PDP or cart → checkout often deliver outsized revenue.
Typical causes & fixes (search-focused)
- Zero or thin results: Add synonyms, fuzzy search, bigram/phrase fields, spelling; tune boosting and category redirects.
- Slow responses: Reduce Average Response Time with caching, query batching, and precomputed facets.
- Facet friction: Surface in-stock, size, and price first; hide empty values; enable search-within-facet.
- Irrelevance at top-k: Use hybrid retrieval (BM25 + vectors) and re-ranking/LTR; cap business boosts.
- UI issues: Low contrast, weak snippets, missing thumbnails/ratings; improve result cards.
- OOS/variant pain: Filter or down-rank out-of-stock; highlight “Available in your size.”
Best practices
- Event hygiene: Stable names, one event per action, de-dup rapid repeats.
- Attribution window: Define time limits between steps (e.g., 30 min session) to avoid false gaps.
- Guardrails: Exclude bot/dev traffic; filter error responses (5xx/timeout).
- Dashboards: Show step-by-step fallout, segments, and trend lines; annotate releases.
- Targets: Own specific gaps (e.g., “Reduce Search → Results fallout from 18% to 12% in Q4”).
Challenges
- Looping paths: Users revisit search or add filters out of order; use flexible funnels that allow backtracking.
- Mixed intents: Informational queries will naturally “fall out”; segment by intent classifier.
- Confounded signals: Price, stock, and latency interact—analyze multi-variate.
Examples
- Results → PDP fallout drops 6pp after deploying bigram matching and better snippets.
- Search → Results fallout improves when zero-result queries trigger suggested terms and popular categories.
- Cart → Checkout fallout shrinks with “guest checkout” and upfront shipping costs.
Summary
Fallout pinpoints where shoppers abandon the journey. Instrument clean events, segment the funnel, and fix the usual culprits—zero results, slowness, weak relevance, facet friction, and OOS—to recover conversions.
