Connecting business systems without the nightly export

Most companies connect their systems with a file. Something exports a spreadsheet overnight, something else imports it in the morning, and a person checks occasionally that it happened.

Why file transfer keeps failing

• Failures are silent — the import runs, finds nothing, and reports success
• Data is stale by up to a day, so two systems disagree all afternoon
• A column added upstream breaks the parser without warning
• Nobody owns it, because it is not really a system
• Errors are found by customers rather than by monitoring

It survives because it works most of the time, which is exactly what makes it dangerous.

What to build instead

Talk to the interface, not the export. Nearly every modern business product has an API. It returns errors you can act on, supports incremental updates, and does not break when a column moves.

Push, do not poll, where you can. Webhooks let the source system tell you when something changed. Fewer requests, near-real-time results.

Make it idempotent. Processing the same message twice must not create two invoices. This single property removes most of the operational pain from integration work.

Queue the work. A queue between systems means the destination being briefly unavailable delays the work rather than losing it.

Monitor the flow, not the job. Alert when the expected volume does not arrive. A job that succeeds while transferring nothing is the failure mode that hurts.

Do not connect everything to everything

Six systems joined directly is fifteen connections, each needing maintenance. A single integration layer in the middle turns that into six. It is more work initially and dramatically less afterwards.

Where to start

Take the one flow where staff currently retype data between two systems. That is where the errors are, and it is where the return is easiest to demonstrate. Do it properly — with error handling, retries and monitoring — then use it as the pattern for the rest.

The measure of a good integration is that nobody talks about it. Silence means the errors are being caught before they reach people.

We build these under business automation.