Skip to main content

​
Overview

Breakdowns split a metric by a property so you can compare segments—for example runs by workflow_id, users by plan, or errors by error_class.

​
How breakdowns work

Without breakdown: 
Total agent_run_completed: 10,000
With breakdown by workflow_id: 
├── weekly_report: 4,500 (45%)
├── crm_sync: 2,800 (28%)
├── onboarding: 1,700 (17%)
└── other: 1,000 (10%)

​
Breakdown types

​
Event properties

Property (examples)Use
workflow_id, surface, modelProduct behavior
outcome, error_classQuality
utm_source, utm_campaignAttribution
Custom fields from track()Anything you instrument
Event: agent_run_completed
Breakdown: outcome

→ success vs. cancelled vs. error mix

​
User (profile) properties

Property (examples)Use
country, device_type, planAudience
primary_workspace_id, roleB2B context
Custom traits from people.setYour taxonomy
Metric: unique users (any key event)
Breakdown: plan

​
Time buckets

Break down by hour, day, week, or month to see seasonality and releases.

​
Multiple breakdowns

Adding a second dimension (e.g. workflow_id × device_type) increases cardinality. Use when you have a specific hypothesis; otherwise start with one breakdown.

​
Ordering and limits

  • Sort by value (most common first) or alphabetically
  • Cap “other” for long tails (e.g. top 10 templates)

​
Comparison across segments

Use the same metric and time range, then change only the breakdown to compare channels or plans fairly.

​
Good vs. risky breakdowns

✅ workflow_id (bounded template set)
✅ plan, country, device_type
✅ outcome, error_class (small enums)

❌ Raw user IDs, full URLs, or unbounded free text
✅ Prefer buckets: input_length_bucket, template_category

​
Examples

​
Runs by workflow

  • Metric: count of agent_run_completed
  • Breakdown: workflow_id

​
Activation by channel

  • Metric: unique users with activation_milestone_hit
  • Breakdown: utm_source

​
Time to complete

  • Metric: average duration_seconds on agent_run_completed
  • Breakdown: model or surface

​
Best practices

  1. One primary breakdown per chart until the story is clear.
  2. Align names with Naming conventions.
  3. Save the view to a Board when it becomes a standing review.

​
Next steps

Filters

Narrow before breaking down

Insights

Build charts