04Insights · The Metrics
Conversion rate is a lagging indicator. Below are the eight leading ones — the metrics that tell you whether your real-time activation layer actually exists, runs fast enough, and drives revenue.
The hierarchy
Identity feeds Decisioning feeds Outcome. If identity capture is broken, decisioning has nothing to act on. If decisioning is slow, outcome never moves. Read the stack top-down to find the bottleneck.
Reading the dashboard · where to look first when numbers move
Outcome flat? → check Decisioning.
If lift is flat or negative, latency is too high or the NBA ranker is picking weak actions. Pull P95 latency and selection-vs-random lift first.
Decisioning starved? → check Identity.
If action coverage is below 40%, you do not know enough about the visitor to act. Identification rate and identity drop-off explain why.
Identity capture stuck? → check the funnel step.
The Identity Signal Heatmap shows which step bleeds the most. Most teams find one step that loses 50%+ — fix that, double identification.
Numbers good, revenue flat? → check coverage.
Per-session lift is real but only on covered sessions. If coverage is 30%, total revenue impact is capped at 30% × per-session lift. Expand action library.
01Identity
02Decisioning
03Outcome
Identity
identified sessions / total sessions
Target15–25%+
Why it matters
How much of your traffic do you ever know? Most stores sit at 2–8%. Every percentage point unlocks owned-channel reach.
How to instrument
Count sessions where email, phone, or known login attaches before exit. Divide by total sessions. Track daily.
Identity
median seconds from first event → identity capture
Target< 90s for high-intent
Why it matters
The faster you capture identity, the more session left for personalization. Late capture is half a CDP.
How to instrument
Log timestamp of first event vs identity event. Bucket by intent score and source. P50 and P95.
Identity
1 − (identified at step N / identified at step N-1)
Target< 30% per step
Why it matters
The Identity Signal Heatmap. Shows where the funnel leaks identity — exit intent popup, checkout, loyalty signup, etc. Reveals the highest-leverage capture moment.
How to instrument
Define funnel: landing → engaged → cart → identify → checkout. Plot identification at each step.
Decisioning
ms from event → rendered action
Target< 100ms P50, < 250ms P95
Why it matters
Anything slower and the bounce wins. Latency is the single most under-measured KPI in personalization stacks.
How to instrument
Tag event ingest timestamp. Tag DOM render timestamp. Diff. Pipe to time-series dashboard.
Decisioning
lift of selected action vs random eligible action
Target> 1.5× control
Why it matters
The decisioning brain has a job: pick the action that converts. Holdout-aware lift quantifies whether it does.
How to instrument
Run 10–20% holdout group on each campaign. Compare conversion rate of NBA-selected vs random or rule-default.
Decisioning
sessions with ≥1 eligible action / total sessions
Target> 60%
Why it matters
You cannot influence sessions you have no action for. Coverage is the ceiling on revenue impact. Below 60% means most visitors get no intervention.
How to instrument
Count sessions where at least one campaign / journey is eligible. Divide by total. Bucket by intent.
Outcome
(treated CR − holdout CR) / holdout CR
Target+8–25%
Why it matters
The headline KPI. Did real-time intervention move conversion vs doing nothing? Without a holdout, you are guessing.
How to instrument
Run platform-level holdout (5–10% of sessions see no Convertive action). Compare CR weekly. Statistical significance over 4 weeks.
Outcome
identified sessions that purchase / identified sessions
Target8–18%
Why it matters
Identification only matters if it converts. Tracks the second leg of the funnel: did capturing identity actually pay off?
How to instrument
Cohort by identification source (popup, checkout, loyalty). Track 7-day and 30-day purchase rate.
Outcome
incremental revenue / treated sessions
Target$0.10–$0.80+
Why it matters
Revenue language. Translates lift into a per-session unit so you can compare against ad spend and CAC. The number that gets the CFO nodding.
How to instrument
Revenue from treated cohort − revenue from holdout, divided by treated session count. Roll up monthly.
Questions
Conversion rate is a lagging, multi-causal metric. It moves on traffic mix, seasonality, promo cycles, and stack changes — drowning the signal from any one intervention. The KPIs on this page are leading indicators that isolate the in-session revenue infrastructure.
No. Start with three: identification rate, decision latency P95, and in-session lift (with a holdout). Those three tell you whether the infrastructure exists, runs in real time, and drives revenue. The rest are diagnostics that explain why.
Standard ecommerce KPIs (CR, AOV, sessions, revenue) measure store performance. These measure infrastructure performance — whether your real-time activation layer is doing its job. They sit upstream of CR and AOV and explain movement in those numbers.
Latency and coverage metrics are visible in 24 hours. Identification rate and time-to-identify stabilize in a week. In-session lift requires 2–4 weeks of holdout data depending on traffic volume to reach statistical significance.
Free 12-hr ROI audit benchmarks your store on identification rate, leakage, and recoverable revenue.