Viewer Performance and Journey

This chapter explains the two cards that follow the visitor trends under 3D statistics: Model load performance and Viewer journey. Views tell you how many people opened the viewer; these two cards answer whether it came up properly and whether people actually looked once it did.

How load performance is measured

• Every time the viewer page is opened or refreshed counts as a new load attempt, so repeated refreshes by the same visitor are counted separately
• For embedded viewers, only attempts where the viewer area entered the screen by at least 10% are counted. A viewer embedded low on a product page that preloads before the visitor scrolls to it is excluded, because nobody was waiting for it
• Load time runs from the moment the viewer enters the screen until the 3D model is ready

Model load performance metrics

• Total attempts / settled attempts — settled attempts ended in success, error or exit-before-ready. Attempts that just started and are still loading are hidden briefly
• Load success rate — share of settled attempts that succeeded
• Exit-before-ready rate — share of visitors who left before the model appeared. If this is high, look at the "wait before exit" in the journey card below
• Success within 10 s — share ready within 10 seconds. On mobile product pages this is the first number to watch
• Typical load time (p50) — half of visitors saw the model within this time
• Average load time — heavily influenced by a slow minority, so read it together with p50
• Slow load time (p95) — what the slowest 5 in 100 visitors experienced. The benchmark for slow connections and low-end devices
• Load outcomes — ready · exit before ready · result missing (no result arrived within a set time after loading started, e.g. the tab was force-closed)
• Load time distribution — successful loads per bucket from under 2 s to 60 s and over. Two peaks mean the experience is splitting by device or connection

Administrators additionally see end-to-end load time (from the viewer page request to model ready) and server bootstrap time (reading the viewer settings from the database). Bootstrap time is a part of end-to-end time, so do not add the two.

When loading is slow

• In the editor, enable LOD under Normal edit → Rendering tab, or delete unnecessary regions in Splat edit to reduce size
• Creating with SH (Spherical Harmonics) level 0 reduces file size substantially
• Customize the loading screen in style editing to reduce exits while people wait

Viewer journey — exits before load

How long visitors who left before the model appeared waited before leaving.

• All exits / analyzed exits — analyzed exits are visits where the viewer area stayed on screen and the tab stayed active until they left
• Average · typical (p50) · long wait (p75, p90) before exit
• Wait-before-exit distribution — under 3 s · 3–10 s · 10–30 s · 30–60 s · 60 s and over. Many "under 3 s" exits are visitors who never meant to look at the 3D (scrolled past); many "10–30 s" exits are visitors who gave up waiting, which calls for faster loading

Viewer journey — time after load

How long visitors actually looked once the model was ready. There are three times, and always estimated viewable time ≤ active tab time ≤ total open time.

• Estimated viewable time — time the viewer was at least 10% visible in an active tab. The most conservative and the most meaningful value
• Active tab time — time the tab was active, including while the viewer was scrolled off screen
• Total open time — everything, including time spent on other tabs

Each time comes with average · typical (p50) · long (p75) · very long (p90), and viewable retention is the share of visits where the viewer was visible for at least 10 s · 30 s · 1 min · 3 min.

• Visits with post-load time / visits with final time — final visits are those whose end signal was received. A low end-signal rate means many browsers could not send it (force-close and similar)

Measurement methods v1 · v2

Viewable time is measured in two ways depending on the browser, and the card lets you toggle v1/v2 to compare (at least one is always on).

• Precise (v2) — Chromium browsers such as Chrome and Edge on desktop and Android. Detects not only whether the viewer is on screen but also whether it is covered by another element such as a modal
• Estimated (v1) — Safari, Firefox and most browsers on iPhone and iPad. Estimates from the intersection area, so covered time is not filtered out and values can run long

Sample counts and shares for both methods are shown, so for items with a large iOS audience keep in mind that v1 values are somewhat generous.

Linked item transitions (item statistics only)

Records of moving from item A to item B through a connection button.

• Total transition clicks · transition load success rate · typical (p50) · slow (p90) transition time
• The table lists the path (A → B), connection kind (open/close · product variant), clicks, share within source item (this path's share of all connection clicks from the same source item) and success rate
• Outcomes are ready · error · exit before ready · superseded (another item was selected during load) · pending

Load time and exit data are shown only from the date collection started. The date is noted at the top of the card.