Tracker Specification
Learn about the Journium Insight Tracker specification. Irrespective of how a tracker is created or maintained, it is always stored as a YAML file.
A Journium InsightTracker is defined using a structured YAML format that follows the InsightTracker schema.
Top-Level Properties
Prop
Type
metadata Properties
Prop
Type
spec Properties
Prop
Type
trigger Properties
Prop
Type
window Properties
Prop
Type
Shared Primitives
These types are referenced by multiple analysis configurations.
PropertyFilter
Used to narrow the set of events considered by an analysis step.
Prop
Type
BreakdownDef
Segments analysis results by a property dimension.
Prop
Type
Analysis Types
The analysis.type field selects the analysis algorithm. Each type computes different metrics
and injects different template variables into your llm.promptTemplate.
| Type | Description |
|---|---|
funnel | Ordered step completion — find where users drop off |
retention | Cohort return-rate curves (D1 / D7 / D30) |
activation | % of new users who reach the first value moment |
churn | Users who were active but have gone silent |
featureAdoption | Fraction of active users who have tried a feature |
anomaly | Automated spike / dip detection vs. a statistical baseline |
conversion | Simple numerator / denominator conversion rate |
cohort | Compare a metric across cohorts by join date or property |
trend | Count events or unique users per time period |
custom | Raw event list analysis — not available in the NL Builder |
See the Analysis Types overview for per-type configuration reference and examples.
llm Properties
Prop
Type
Universal Template Variables
These variables are always available in promptTemplate, regardless of analysis type:
| Variable | Description |
|---|---|
{{windowPeriod}} | Human-readable description of the analysis window, e.g. "last 7 days" |
{{dataAsJson}} | Full structured result from the analysis query, serialised as JSON |
{{executedAt}} | ISO 8601 timestamp of when this execution started |
Type-specific variables are documented on each analysis type's reference page.
privacy Properties
Prop
Type
Full Example
apiVersion: journium.app/v0Beta
kind: InsightTracker
metadata:
name: checkout-funnel
displayName: Checkout Funnel Analysis
description: Track drop-off across the 4-step checkout flow
spec:
type: LLM
trigger:
mode: automatic
schedule: daily
window:
period: last_7d
granularity: day
timezone: America/New_York
analysis:
type: funnel
entity: person_id
steps:
- event: cart_viewed
label: Viewed Cart
- event: checkout_started
label: Started Checkout
- event: payment_entered
label: Entered Payment
- event: order_completed
label: Completed Order
conversionWindow: 2h
breakdowns:
- property: plan
maxCardinality: 5
llm:
promptTemplate: |
Analyze this checkout funnel for {{windowPeriod}}.
{{totalEntrants}} users entered the funnel with an overall conversion of {{overallConversionPct}}%.
Step breakdown: {{stepsSummary}}
Worst performing step: {{worstStep}} ({{worstStepCvr}}% CVR, {{worstStepLoss}} users lost).
Full data: {{dataAsJson}}
Provide 2-3 concrete, actionable recommendations to improve conversion.
maxOutputTokens: 500
guardrails:
citeSources: true
privacy:
excludeProperties:
- email
- phone
hashProperties:
- userIdHow is this guide?
Last updated on
Creating Insight Trackers
Learn how to create Insight Trackers in Journium. You can use a UI based wizard, or create them manually using the tracker specification YAML.
Tracker Repositories
Learn about Journium Tracker Repositories — how to version-control and manage your insight tracker specifications.