Derived Event Generation

💼 Business Context

• Business logic like 'cart abandoned' or 'browsing spree' isn't a single raw event — it's a pattern the platform computes once, so every team doesn't reinvent it
• Standardizes business-meaningful moments across the company instead of each team defining its own version of 'abandonment'
• Owned by Platform Engineering, with derivation rules defined in partnership with the business teams that consume them

🔌 Technical Overview

The Stream Worker maintains short-lived windowed state per customer to detect patterns and emit derived events: e.g., cart_abandoned fires when an add_to_cart event has no matching order_paid within a configurable window (default 60 minutes); browsing_spree fires after N product views within M minutes. Derived events are published back onto Azure Event Hubs with the same schema as any other event, so they flow through the rest of the pipeline identically.

💾 Derived Event

{
  "event": "cart_abandoned",
  "event_id": "derived-3c8e...",
  "user_id": "cust_004821",
  "derived_from": ["evt_add_to_cart_991"],
  "properties": { "cart_value": 6998, "minutes_since_add": 62 }
}

🔗 Integration Points

• .NET Core Stream Worker — windowed pattern detection and derivation logic
• Azure Cache for Redis — holds in-flight windowed state per customer
• Azure Event Hubs — derived events are republished here, same as primary events
• Activation API — the most common consumer of derived events (e.g., cart-abandonment recovery)

🧰 Services Consumed

• Owning microservice — Cxos.Processing.Api (see the Full Application Service Map)
• Database — ADLS Gen2 (Iceberg) + Cosmos DB Table API (stream checkpoints)

⚠️ Non-Functional Considerations

• Scale: windowed state is sharded by customer ID and scales with the Stream Worker's partition count
• Latency: derived events fire as soon as their defining window/condition is met, typically within minutes
• Reliability: derivation rules are versioned; a rule change doesn't retroactively alter already-emitted derived events
• Security/Privacy: derived events carry the same classification and consent handling as the raw events they're computed from

🎯 Enterprise Example

A customer adds an item to cart and closes the tab. 62 minutes later, having seen no order_paid event, the Stream Worker emits cart_abandoned — which the Activation API picks up to trigger a recovery email, entirely without the marketing team having to build their own abandonment-detection logic.