How this track works
Eight analyst sessions, one store, one map. You draw Mango Lane's screen map today (b1), measure where sessions enter it (b2), compare the search road against the browse road (b3), then weld the whole journey into a live flow simulator (b4). The back half goes deep where the money is: the product page (b5), the checkout (b6), a full leak diagnosis (b7), and the journey scorecard you hand to the business (b8). The leader track (a1-a6) reads the same journey without computing anything. New here? The prequel learn-ecommerce-metrics-with-phoebe defines every metric this course builds on.
Six screens carry the whole store 8 min live
Strip away every pixel of design and Mango Lane is six screen types: Home, Search results, PLP (product listing page - a category grid), PDP (product detail page - one product's own screen), Cart, and Checkout. Every shopper journey is some walk across these six. The counts below are the same month you decomposed in the metrics course - now placed on the map.
LiveWhy the map beats the funnel4 min▶
The metrics course drew Mango Lane as a straight funnel: sessions → PDP → cart → checkout → order. True, but flattened. The map restores what the funnel hides:
- Multiple doors. Only half of sessions start at Home. 54,000 land directly on a PDP from an ad, a push notification, or a shared link - they skip the top of your "funnel" entirely.
- Multiple roads. A shopper reaches the PDP by searching, by browsing categories, or by landing on it. Same gate in the funnel, completely different experiences to fix (b3 shows they convert very differently).
- Dead zones. 36,000 sessions are account checks and order tracking - they were never going to buy. A "conversion rate" that includes them understates how well the shopping journey actually works.
This is what GA4's path exploration draws for you. Its tree graph visualizes the sequences of screens and events users actually walk - forward from any starting point, or backward from a goal ("how did users get to purchase?"). It shows the top 5 next screens per step, expandable to 20. The map you drew by hand today is the mental model that makes that report readable instead of overwhelming.
Self-studyThe map's numbers, cross-checked3 min read▶
Every count on the map reconciles with the metrics course - that is the point of a canon store:
| Map fact | Cross-check |
|---|---|
| Entries: 180K + 54K + 54K + 36K + 36K | = 360,000 sessions, the b1 baseline |
| PDP reach: 36K + 54K + 27K + 27K + 0 | = 144,000 = 40% PDP penetration |
| Cart 36,000 · Checkout 16,200 | = the 25% and 45% gates |
| Checkout x 60% completion | = 9,720 orders, 2.7% session CR |
b2 unpacks the per-door reach rates (Home entries reach a PDP only 20% of the time; search landings 75%) - the map's most actionable hidden numbers.
The five doors: where sessions begin 8 min live
Analytics tools call the first screen of a session the landing page (or landing screen), and they count entrances: the number of sessions whose first event happened on that screen. Mango Lane's five doors: Home 50%, PDP direct 15%, PLP 15%, Search landing 10%, Other 10%. The mix matters because door quality differs wildly - a session that lands on a PDP is already looking at a product; a session that lands on Home still has to find one.
LiveWho sent them? Three scopes of "source"4 min▶
Each entering session also carries where it came from, and GA4 records that at three scopes - mixing them up is a classic reporting bug:
- First user source (user-scoped): the origin of this person's first-ever visit. Never changes afterward. Answers "which channel finds us new customers?"
- Session source (session-scoped): assigned fresh to each session. Answers "which channel drove this month's visits?"
- Event/key-event source (event-scoped): credit for a conversion under an attribution model. Answers "which channel gets credit for the order?" - the deep version of this is its own course (learn-marketing-attribution-with-phoebe).
Self-studyThe re-engagement doors: push and deeplinks3 min read▶
PDP-direct entries are mostly manufactured: an ad click, a shared link, or a push notification deep-linking straight to a product. Vendor benchmarks (flagged as vendor data - Airship's report covers hundreds of billions of pushes) find rich image notifications lift direct opens by about 56%, and users who receive any push in their first 90 days retain at roughly 3x the rate of those who receive none. The metric names worth knowing: a direct open (user tapped the push) versus an influenced open (user opened the app soon after receiving one). No credible benchmark exists for "deeplink conversion vs organic" - be suspicious of anyone quoting one.
The events that draw the map ★ 15 min · everyone works
The map is only as good as its instrumentation. Two events record every step: on the app, screen_view fires automatically each time the visible screen changes (carrying firebase_screen_class, plus a screen_name you set); on the web, page_view does the same job. Layer the ecommerce events from the metrics course on top (view_item, add_to_cart, begin_checkout, purchase) and every journey fact in this course becomes computable.
LiveWalk one session through the event log6 min▶
Maya's Tuesday session, as Mango Lane's analytics sees it:
session_start + screen_view (screen_name: home) - she opens the app. Entrance counted for Home.
view_search_results (search_term: "linen dress") - she searches. The search road begins; the term itself is now data.
screen_view (pdp) + view_item - she opens a dress. This session now counts toward PDP penetration.
add_to_cart → screen_view (cart) → begin_checkout → purchase - the money path, four events, each one a gate from the metrics course.
Read the pairing: screen_view events draw the map; ecommerce events mark progress on it. You need both - screens without commerce events cannot compute conversion, commerce events without screens cannot locate a problem.
LiveThe three instrumentation traps5 min▶
- Unnamed screens. Auto screen tracking reports the class name (
ProductViewController) unless you set a humanscreen_name. A path report full of class names is technically instrumented and practically unreadable. - Screens that do not fire. Auto-logging keys on native screen changes; modern declarative UI (e.g. SwiftUI) and modal sheets often need manual
screen_viewlogging. The symptom: a screen everyone uses that shows near-zero views. Trust the map only after you have walked one real session against the log - exactly the Maya drill above. - Search that does not register. Web search tracking triggers on URL query params (q, s, search, query, keyword by default). An app or SPA whose search never changes the URL silently vanishes from search analytics - the search road disappears from your map while shoppers walk it all day.
The instrumentation audit is always finding number one. Nearly every journey-analytics engagement starts the same way: walk five real sessions by hand against the event log, find the screen or search box that is not firing, and fix the map before trusting any number drawn on it. b8's scorecard bakes this in as a standing checklist row.
Try it yourself - this week ◐ 15-20 min total
- Draw your own product's screen map: the 5-8 screen types, arrows for the main roads. One page, boxes and arrows, no tooling.
- Find your five doors: your analytics tool's landing page/screen report, sorted by entrances. What share is your Home?
- Walk one real session (yours) against the event log. Note every screen you visited that the log missed.
- Memorize the entry canon: Home 180K, PDP-direct 54K, PLP 54K, Search 36K, Other 36K. Every remaining session uses these doors.
Sources this session draws on
Journey analytics lives in the platforms' own documentation; this course applies it to one store end to end. This page draws on:
Three questions before you go 🎯 ◐ 90 seconds
1 · 54,000 sessions land directly on a PDP. What does the plain funnel view get wrong about them?
The funnel flattens the map. PDP-direct entries (ads, pushes, shared links) enter at the product itself - their journey, and their fixes, look nothing like a Home entry's.
2 · A screen everyone uses shows near-zero screen_view counts. Most likely cause?
Auto screen tracking keys on native screen changes; SwiftUI-style UI and modal sheets often slip past it. Walk a real session against the log before trusting the map.
3 · You want to know which channel finds NEW customers. Which source scope answers it?
User-scoped first source never changes and answers acquisition questions. Session source answers this month's traffic; event scope answers conversion credit (the attribution course's territory).