Interval Analysis
Interval Analysis (Interval) measures the distribution of time intervals between two specified events, answering "how long does it usually take from A to B". It corresponds to the backend AnalyticsService.Interval.
1. What Questions It Answers
- How long does it take on average from "register" to "first payment"? What are P50 / P90?
- What does the decision-time distribution from "browse product" to "place order" look like?
- Is there a long tail in the interval from "login" to "key action completed"?
2. Key Metrics
| Metric | Meaning |
|---|---|
| Bucketed distribution | Share of samples per interval length range |
| P50 | Median interval (hours) |
| P90 | 90th-percentile interval (hours, long tail) |
| avg | Average interval (hours) |
Bucket labels: instant (immediate) / lt_1h (within 1 hour) / 1_24h (1-24 hours) / 1_7d (1-7 days) / 7d_plus (over 7 days).
3. Backend API
gRPC: AnalyticsService.Interval
| Field | Type | Description |
|---|---|---|
time_range | TimeRange | Analysis time range |
event_from | string | Start event name (e.g. register) |
event_to | string | End event name (e.g. pay_success) |
app_id | uint32 (optional) | Filter by app |
Response IntervalResponse: buckets[] (IntervalBucket contains bucket / count / percentage), p50_hours, p90_hours, avg_hours, count.
HTTP (admin proxy)
POST /admin/v1/analytics/interval
Content-Type: application/json
{
"timeRange": { "startMs": 1718169600000, "endMs": 1718774399000 },
"eventFrom": "register",
"eventTo": "pay_success"
}
4. Difference from Distribution Analysis
| Model | Measures | Input |
|---|---|---|
| Distribution | The duration of a single event | One event |
| Interval | The time difference between two events | Start + end event |
5. Typical Scenarios
Time from registration to first payment
eventFrom=register, eventTo=pay_success: view the decision cycle from registration to first payment. A short P50 means fast conversion; a high 7d_plus share means many users postpone paying — consider limited-time newbie offers.
Browse-to-order decision
eventFrom=view_product, eventTo=submit_order: evaluate purchase decision time and tell impulse buying (instant/lt_1h) from deliberate comparison shopping (1_24h+).
Adding a time dimension to conversion funnels
Combine with Funnel Analysis: the funnel tells you "what the conversion rate is", while Interval tells you "how fast converters convert and how long non-converters stall".
6. Notes and Caveats
- Both events must be specified:
event_fromandevent_toare both required. - Paired per user: the interval is the same user's time difference from
event_fromto (the first subsequent)event_to; if a user triggers multiple pairs, the first pairing is used. - No cross-user pairing: user A's "from" is never paired with user B's "to".
- Unit is hours: P50/P90/avg are all in hours; bucket labels mix seconds/hours/days — read them by their label semantics.
- Negative/abnormal values: if
event_toprecedesevent_from(data anomaly or backfill), it is usually filtered out.