Enterprise Integration
A typical company runs dozens to hundreds of applications — ERP, CRM, HRIS, ITSM, e-commerce, data warehouse — and almost every business process crosses several of them. A closed deal in the CRM must become an order in the ERP, a new hire in the HRIS must become an account in the identity provider, and a shipped order must update the customer portal. Enterprise integration is the discipline of making those handoffs reliable.
Integrations are where data goes missing, duplicates appear, and outages spread. Good integration work treats each connection as a product with a contract, an owner, monitoring, and a way to prove the systems agree.
TL;DR
- Choose a pattern from business needs — timing, volume, consistency, and failure tolerance.
- Every integration needs an owner, a contract, monitoring, and a runbook.
- Prefer APIs and events over shared databases and screen scraping.
- Assume messages will be duplicated, delayed, or reordered — design for idempotency.
- Reconcile. Delivery confirmation doesn't prove the two systems agree.
- Avoid point-to-point sprawl with shared standards and a visible integration catalog.
Quick Example
An integration catalog entry, written before the integration is built:
Core Concepts
Integration Patterns
See Messaging & Event Streaming, Webhooks, and Change Data Capture.
Topologies
- Point-to-point — each pair of systems connected directly. Fast to start; with n systems it tends toward n² fragile connections.
- Hub-and-spoke / ESB — a central enterprise service bus transforms and routes messages. Centralized control, but can become a bottleneck owned by one team.
- iPaaS — a cloud integration platform (MuleSoft, Boomi, Workato, Azure Logic Apps) with connectors, mapping, and monitoring. Faster delivery, vendor dependence.
- API-led / event mesh — systems expose stable APIs and events, and teams compose them. Scales best but needs governance and platform maturity.
Master Data Synchronization
When customers, products, or employees live in several systems, define the system of record, the identifier used across systems, and the direction of flow. Bidirectional sync of the same field is a recipe for loops and overwrites. See Enterprise Applications.
Reconciliation
Reconciliation compares source and target to prove they agree: record counts, sums of amounts, or hashes of key fields over a period. It catches silent failures that per-message monitoring misses — skipped records, mapping bugs, or manual edits.
Best Practices
Define Contracts
Document the schema, required fields, identifiers, error codes, and versioning policy for each integration. Treat breaking changes like API versioning.
Make Every Write Idempotent
Use a stable business key so a retried message updates instead of duplicating. This single practice prevents most integration data-quality incidents.
Monitor Business Outcomes
Alert on "no orders created in the last hour during business hours," not just on HTTP errors. Include correlation IDs so one transaction can be traced across systems.
Handle Failures Explicitly
Retries with backoff, dead-letter queues, alerting, and a documented replay procedure. Someone must own the failed-message queue.
Keep Transformation Logic Reviewable
Mappings buried in a visual iPaaS canvas are hard to review. Export definitions to version control where the platform allows, and test them.
Common Mistakes
Reading Vendor Databases Directly
It works until the vendor upgrades the schema. Use supported APIs and events.
Trusting "Delivered" as "Correct"
A message can arrive and still be mapped wrongly or rejected by business validation. Reconcile.
No Owner After Go-Live
Integrations built by a project team and then abandoned fail silently for months.
Bidirectional Sync Without a Master
Two systems editing the same field and syncing both ways creates overwrite loops and lost updates.
FAQ
What is enterprise integration?
The practice of connecting an organization's applications so data and processes flow between them reliably — using APIs, events, messaging, batch transfers, and integration platforms.
What is the difference between iPaaS and an ESB?
An ESB is typically self-hosted middleware that centralizes routing and transformation. An iPaaS is a cloud-hosted integration platform with prebuilt connectors, visual mapping, and managed runtime. iPaaS trades some control for faster delivery and less infrastructure to run.
When should I use events instead of APIs?
Use events when several systems need to react to the same business fact and the producer shouldn't wait for them. Use synchronous APIs when the caller needs an immediate answer, such as a validation or a price lookup.
How do I prevent duplicate records?
Use idempotency keys based on stable business identifiers, upsert rather than insert in the target, and reconcile periodically to catch anything that slipped through.
Related Topics
- APIs & Integration — API design, versioning, and gateways
- Messaging & Event Streaming — Queues, logs, and event pipelines
- Idempotency — Making retries safe
- Enterprise Applications — The systems being integrated
- Data Quality Management — Keeping shared data trustworthy