ERP Integration Guide: Patterns, Data Ownership and Reliability

Every ERP lives among other systems: web shops, marketplaces, CRMs, payment providers, carriers, banks, BI tools. Integration determines whether data flows automatically or relies on people exporting and importing files.

This guide covers how to design integrations that keep working.

Start with data ownership

For each entity, decide which system is the source of truth:

Most integration bugs we're asked to fix trace back to two systems both believing they own the same field.

Choose a pattern

Point-to-point

Direct connections between two systems. Simple for one or two integrations, messy at scale.

Hub-and-spoke (middleware)

A central integration layer brokers between systems. Better for several integrations: transformations, monitoring and retries live in one place.

Event-driven

Systems publish events (order created, payment received) that others consume. Low latency and loose coupling, at the cost of more infrastructure.

Batch

Scheduled transfers of files or API calls. Fine for data that doesn't need to be real-time.

Build for failure

Integrations fail. APIs time out, tokens expire, data arrives malformed. Design for it:

• Idempotency — every write carries a unique key, so retries never create duplicates.
• Retries with backoff — transient errors retried with increasing delays.
• Dead-letter queues — permanent failures parked visibly with full context.
• Replay — fixed messages can be reprocessed.
• Validation — reject bad data early with clear reasons.

Monitor and alert

An integration without monitoring will fail silently. Log every message with searchable identifiers (order number, SKU). Track throughput and error rates. Alert a named owner when failures exceed thresholds.

Respect rate limits

Most APIs limit request rates. Batch updates instead of pushing every change, and handle throttling responses gracefully.

Security

• Dedicated integration credentials with minimum permissions
• Secrets stored in a secrets manager, not code
• Signature verification on incoming webhooks
• Encrypted transport everywhere

Connector or custom?

Off-the-shelf connectors are worth evaluating first. If a connector fits your rules for customers, taxes, variants and fulfilment, use it. If it requires heavy modification, a purpose-built integration is often cheaper to own.

Testing

Test the happy path, then the failures: duplicate events, missing products, expired credentials, partial outages, peak volumes. Most production incidents come from scenarios nobody tested.

Ownership after go-live

Every integration needs an owner and a runbook. When it fails at 2 a.m., someone should know what to check.

For API-specific detail, read ERP API integration explained. For services, see ERP integration.