ERP sync across an isolated factory floor

The factory floor system was, correctly, not on the public internet. The consequence was that every piece of data moving between it and the cloud ERP moved because a person exported a file and carried it across — which meant the two systems were only ever as current as the last time somebody remembered.

When a transfer failed, it failed silently. The first symptom was usually a downstream number that looked wrong, discovered days later by someone who then had to work backwards to find out which sync had been missed.

A sync service running over a private link between the two environments, on a schedule rather than on a person's memory. It is built to expect failure: a dropped connection or an unavailable endpoint is retried rather than abandoned, and a run that cannot complete surfaces as an alert instead of as silence.

State for every run is reported to a single dashboard, so the question "is the factory in step with the ERP right now" has an answer you can look at.

Manual exports came out of the daily routine. More importantly, sync failures became visible at the moment they happen rather than being inferred later from a number that doesn't add up — which is the difference between a five-minute fix and a two-day reconciliation.