Most digital transformation programmes start with a central platform and work backwards to the machine. The architecture diagram is clean, the procurement is simple, and the first dashboard looks convincing. The trouble appears in month seven, when a switch fails, the uplink drops for two hours, and the shift report no longer matches the line.

An edge-first design reverses the dependency. The gateway owns protocol translation, time-stamping and buffering. It can hold seven days of raw data and keep enforcing local stop and alarm rules while the network is down. When connectivity returns, the buffer replays in order, and the central platform receives a complete history rather than a gap.

This matters beyond resilience. Edge processing cuts the volume of data that has to leave the plant. A vibration sensor can compute an FFT on-device and send a feature vector instead of a waveform. A vision system can reject a part in 21 milliseconds without a round trip to a data centre. Bandwidth and cloud costs fall, and control decisions stay inside the control network.

The organisational benefit is less obvious but larger. When the gateway profile is maintained like code, with versions and review, maintenance and IT finally share one artefact. Changes can be tested on a pilot line, rolled back if needed, and audited later. That discipline is what makes the second and third site cheaper than the first.

None of this requires a wholesale replacement of existing controllers. The gateway is added beside equipment that continues to run. The practical rule we use with clients is simple: put intelligence where the decision has to happen, and put history where people need to compare it.