Skip to main content

​
Overview

Flows (also called Sankey diagrams) visualize the paths users take through your product. See what users do before and after any event, discover common patterns, and identify unexpected behaviors.

​
Use Cases

  • What do users do after opening the agent?
  • What paths lead to first successful run?
  • Where do users go after viewing pricing?
  • Where do users go instead of upgrading?
  • What happens after a failed run?
  • How do users find templates vs. custom workflows?
  • What paths correlate with power usage?

​
Quick Start

1

Select Starting Event

Choose the event to center your analysis:
Starting Event: prompt_submitted
2

Choose Direction

Analyze what happens before, after, or both:
Direction: Steps After
Number of steps: 5
3

Configure Display

  • Set number of steps to show
  • Choose to hide specific events
  • Apply breakdowns
4

Analyze Patterns

Identify common paths and unexpected behaviors

​
Flow Directions

​
Steps After

See what users do after an event: Use for:
  • Understanding what users do after key milestones
  • Identifying next-step opportunities
  • Finding unexpected behaviors

​
Steps Before

See what users do before an event: Use for:
  • Understanding what leads to conversions
  • Identifying effective entry points
  • Discovering successful paths

​
Steps Before & After

See the complete journey around an event: Use for:
  • Complete journey understanding
  • Before/after comparisons
  • Holistic flow analysis

​
Reading Flow Diagrams

​
Sankey Structure

Wider paths = More users taking that route Percentages = Share of users from previous step

​
Multi-Step Flows

​
Configuration Options

​
Number of Steps

Control how many steps to display: 
1 step: Immediate next action only
3 steps: Short journey
5 steps: Medium journey (recommended)
10 steps: Full exploration

​
Hide Events

Remove noise by hiding irrelevant events: 
Hide events:
- page_view (too noisy)
- scroll (not meaningful)
- mouse_move (irrelevant)

Keep:
- signup_completed
- agent_run_completed
- integration_connected

​
Breakdown by Property

Split flows by a property: 
Breakdown: device_type

Mobile:
start → run_completed (25%) → success

Desktop:
start → run_completed (45%) → success

→ Compare surfaces and fix mobile friction if needed

​
Filters

Focus on specific user segments or behaviors:

​
Event Filters

Filter the starting event: 
Starting Event: agent_run_started
  └── where: surface = "mobile"

Mobile runs only

​
User Filters

Filter by user properties: 
User Filter:
├── country = "US"
├── AND plan = "pro"

US Pro users only

​
Time Filters

Analyze specific periods: 
Time Range: Last 30 days
Granularity: Daily

See recent behavior patterns

​
Examples

​
After first prompt

Starting Event: prompt_submitted · Direction: 5 steps after Typical branches: agent_run_completed, agent_run_failed, template_applied, docs views, or session end.

​
Before a successful run

Starting Event: agent_run_completed · Direction: 5 steps before Use this to see whether users came from templates, integrations, or support content.

​
Analyzing flows

​
Identifying Patterns

High-Volume Paths: Look for the thickest flows—these are your main user journeys. 
If most users go: landing → docs → signup_completed
→ That path is your primary journey; improve it deliberately.
Unexpected Paths: Look for surprising flows. 
If 20% of users go: checkout → settings → checkout
→ Are they looking for payment options?

​
Identifying Drop-Off

Exit Points: Where do users leave? → Users leaving mid-run—investigate UX and errors Detours: Where do users go instead of converting? → Users may need clearer guidance before publishing

​
Comparing Segments

Use breakdowns to compare: → If mobile underperforms, review layout, latency, and error handling

​
Best Practices

​
Choose Meaningful Starting Events

✅ Good starting events:
- prompt_submitted or agent_run_started (core loop)
- purchase_completed (conversion)
- feature_used (engagement)
- signup_completed (lifecycle)

❌ Poor starting events:
- page_view (too broad)
- click (too generic)
- scroll (not meaningful)

​
Hide Noise

Remove events that obscure patterns: 
Hide:
- Generic page_views (keep specific ones)
- Scroll events
- Mouse events
- Internal/system events

​
Use Appropriate Step Counts

Immediate impact: 1-2 steps
Journey analysis: 3-5 steps
Full exploration: 5-10 steps

More steps = more complexity
Start small, expand as needed

​
Segment Meaningfully

✅ Actionable breakdowns:
- device_type (fix mobile?)
- traffic_source (optimize channel?)
- user_tier (different needs?)

❌ Not actionable:
- random_id
- timestamp

​
Saving & Sharing

​
Save Flow

  1. Configure your flow analysis
  2. Click Save
  3. Name it: “After first prompt — 5 steps”

​
Add to Board

Include flows in dashboards for ongoing monitoring. All saved flows can be added to boards.

​
Next Steps

Boards

Create dashboards with flows

Cohorts

Analyze flows for specific segments