Skip to content
SVHSoftware

Operations

Interfaces don't fail at go-live. They fail in month seven.

The cutover gets all the attention, but the incidents that hurt happen long after — and they're operational, not technical.

Every interface project plans obsessively for go-live: parallel runs, rollback plans, war rooms. Almost none of them plan for month seven — when the war room is disbanded, the project team has moved on, and an upstream vendor ships an update that quietly changes a code set. The message still parses. The interface still runs. The data is now subtly wrong.

Why month seven, not day one

Go-live failures are loud: messages reject, queues stop, someone calls. The dangerous failures are quiet, and they cluster around three events that reliably happen months later:

The upstream change nobody announced. An EHR upgrade, a lab analyzer replacement, a vendor “minor patch” — and a field you mapped now carries different values. Nothing errors; the mapping is simply no longer true.

The queue that backs up on a weekend. Volume grows, a certificate expires, a VPN flaps. If your first indication is a clinician reporting missing results on Monday, the interface has been failing for 60 hours.

The reprocessing decision under pressure. Messages failed; some were resent; some weren’t. Replaying the wrong window creates duplicate results in charts. Without a rehearsed runbook, this decision gets made at 2am by whoever answered the phone.

What operational readiness actually looks like

The fix isn’t better go-live testing — it’s treating the interface as a production service with an operational contract:

This is the entire reason our managed services practice refuses to take a pager before the baseline phase: you can’t detect abnormal until you’ve written down normal. Interfaces are infrastructure. Fund them like it — the alternative is finding out in month seven that nobody did.

← All insights

Tell us what needs to talk to what.

Bring us an integration problem, a device fleet, or a product idea — we will come back with an approach, not a slide deck.

Start a conversation