Top Conversion Paths
Top Conversion Paths (PathSankey) mines the most frequently occurring event sequences from group behavior, answering "what are the mainstream conversion paths of users". It corresponds to the backend AnalyticsService.PathSankey.
1. What Questions It Answers
- From entering the app to completing a purchase, which are the mainstream paths?
- Which path has the highest/lowest conversion rate?
- Are there unexpected "detour" paths (potential drop-off points)?
2. Difference from Funnels
| Model | Logic | Use Case |
|---|---|---|
| Funnel | Predefined fixed ordered steps, measuring conversion | Key path is known; validating conversion |
| PathSankey | Automatically discovers the most common real event sequences | Exploring users' actual behavior paths |
3. Backend API
gRPC: AnalyticsService.PathSankey
| Field | Type | Description |
|---|---|---|
time_range | TimeRange | Analysis time range |
app_id | uint32 (optional) | Filter by app |
top_n | uint32 (optional) | Return the top N popular paths, default 20 |
Response PathSankeyResponse: paths[] (each PathBucket contains the event_sequence string / support_count times the path occurs / unique_users distinct users / conversion_rate).
HTTP (admin proxy)
POST /admin/v1/analytics/path-sankey
Content-Type: application/json
{
"timeRange": { "startMs": 1718169600000, "endMs": 1718774399000 },
"topN": 15
}
4. Interpreting Results
event_sequence is a comma-separated event sequence, e.g. "view_home,view_product,add_to_cart,pay_success":
support_count: how many (user ×) samples completed this full sequence.unique_users: the number of distinct users who walked this path.conversion_rate: the conversion rate of this path (share that completed the full sequence).
Results are sorted by support_count descending; the frontend usually renders them as a Sankey diagram to show path branching.
5. Typical Scenarios
E-commerce main-flow path mining
Without predefined steps, look at users' real paths starting from view_home: is it view_home → view_product → pay (efficient), or view_home → search → view_product → compare → pay (detour)?
Pre-churn paths
Focus on sequences with high support_count that break off midway, pinpointing the key pages before drop-off.
6. Notes and Caveats
- Path length: overly long sequences grow explosively, so the backend usually limits sequence length / top N.
- Event sequence basis: events are chained per user in chronological order; behaviors across sessions and days are included too — make sure
time_rangecovers them. - Paths ≠ funnels: path mining is descriptive; funnels are prescriptive validation. They complement each other.
- Empty data: first confirm events have been stored (see Getting Started · Current State of Data Storage).