Short answer
Data integration for decision intelligence connects enterprise applications, data platforms, industrial systems and documents where they live — using APIs, change-data-capture, streaming and industrial or healthcare protocols — resolves the same entities across systems, and writes approved decisions back into the systems of record. It is decision-scoped: it integrates what each decision needs rather than copying everything into a new store.
Integration patterns
| Pattern | When to use | Examples |
|---|---|---|
| APIs (REST, OData, GraphQL) | Read and write business objects | SAP S/4HANA, Salesforce, ServiceNow, Dynamics |
| Change-data-capture | Near-real-time changes from databases | Oracle, SQL Server, PostgreSQL |
| Streaming | High-volume events and telemetry | Kafka, Event Hubs, Pub/Sub, Kinesis |
| Industrial protocols | Machines, sensors, historians | OPC UA, MQTT, AVEVA PI, Modbus gateways |
| Healthcare standards | Clinical and claims data | HL7 v2, FHIR, X12 EDI |
| Federated query | Large analytical data in place | Snowflake, Databricks, BigQuery, Fabric |
| Documents | SOPs, contracts, manuals, email | SharePoint, file shares, PDFs, inboxes |
From raw data to decision-ready context
- Connect read-only firstStart without write access; prove value in shadow mode.
- Map to a semantic layerBusiness names for tables, tags and fields.
- Resolve entitiesOne asset, order or customer across every system.
- Score data quality per decisionFreshness, completeness and conflicts shown on every decision card.
- Add write-back lastIdempotent, audited and reversible actions into systems of record.
Real-time where it matters
Not every decision needs streaming. Maintenance timing tolerates minutes; a robot fleet or a fraud check needs sub-second signals. Decision-scoped integration picks the latency each decision actually needs — which keeps cost and complexity down.
Key takeaways
- Integrate for decisions, not for completeness.
- Read-only first, write-back last.
- Entity resolution is what makes data from many systems usable.
Frequently asked questions
Do we need a data warehouse before decision intelligence?
No. The platform connects to data where it lives. A warehouse or lakehouse helps for analytics but is not a prerequisite.
How does it connect to industrial equipment?
Through OPC UA, MQTT, historians such as AVEVA PI, and edge gateways — read-only by default.
Can it write decisions back?
Yes — as draft or approved work orders, holds, tasks or updates, through each system’s API, with an audit trail.