Business Observability: Business Events, Flows and Conversion
Business Observability tests whether you can capture business events, model a multi-step business process, and answer commercial questions - conversion, drop-off, revenue, cycle time - with the same rigour as technical SLIs. Questions typically contrast business events with logs and with generic events.
Capture paths
- OneAgent business-event capture rules instrument method calls (for example Java or .NET) and emit events at the point business logic completes.
- Smart capture derives business events from HTTP requests when instrumenting code is impractical.
- The business events ingest API accepts events from anything outside monitored code, such as a batch job or a SaaS platform.
- Events are stored in Grail and queried with
fetch bizevents. They are not logs and not generic events: retention, processing and permissions are configured separately. - OpenPipeline can process, mask, route and split business events before storage, so an event can be dropped or altered before you ever query it.
Design rules
- Each event needs a stable correlation identifier (order id, customer id), a source timestamp, and a small set of low-cardinality dimensions you will actually segment by.
- Business Flow models a multi-step process across event types, showing volume, conversion and drop-off per step, and supports thresholds on those steps.
- Business metrics can back dashboards, SLOs and alerting profiles, bringing business SLIs into the same alerting pipeline as technical ones.
Worked example
An order funnel captures order.created, payment.authorized and shipment.dispatched, each carrying the order id. A conversion query counts distinct order ids per stage within a fixed window and joins on the identifier to find orders that stalled after authorisation. Reasoning: the funnel answers a business question only if the identifier is stable and the window matches the customer journey; without a join key you measure stage volumes, not conversion.
Traps that cost marks
- Ingest time versus source event time: missing timestamps silently become ingest time and skew the funnel.
- High-cardinality dimensions (raw user ids, free-text fields) inflating storage and query cost.
- Assuming OpenPipeline transformations are reversible or non-destructive.
- Confusing business events with logs when reasoning about retention and access.