Why your MES project stalls at the shop floor interface

The selection phase of a manufacturing execution project is comfortable: demos, scoring grids, reference calls, a steering committee that meets on time. The integration phase is where the estimate meets a machine from 2003 with a serial port, an undocumented protocol and a vendor that was bought twice since it was installed.

The pattern is familiar. The software is chosen after months of work, and the connection of a dozen machines from half a dozen vendors is scoped in an afternoon, from a spreadsheet of equipment names. Six months later the schedule has slipped, and almost all of the slip is on the shop floor side.

Where the time actually goes

On the projects we are asked to recover, the overrun is concentrated in a few places.

• Data models that do not match. The machine counts cycles; the MES expects good units per batch, with rejects by reason. Someone has to build the translation, and someone has to validate it.
• Tags that mean something different on each line. “Running” on one filler means motor on; on the next it means producing to specification. Both feed the same OEE figure.
• Equipment that can only export a file. The machine writes a CSV to a local folder every hour, nobody owns the folder, and nobody owns the script that picks it up.
• Interfaces sold as an option. The vendor connector exists, but it needs a licence, a firmware update or a service visit that nobody budgeted.

None of these is a software problem. They are inventory problems that surface as software problems, usually after the go-live date has been announced.

The cheap fix, done early

A low-level assessment of the installed base before the software is chosen changes the shape of the project. For each machine it records the controller and firmware, the protocols actually available, the tags that exist and what they mean, the licences needed, and who owns the equipment and its data. On a typical packaging hall it takes days, not months, and most of it is spent in front of the cabinets with the maintenance team.

It turns unknowns into line items with a cost and an owner. It also changes the vendor demonstrations. The useful question stops being “can it calculate OEE?” and becomes “can it read this machine, this way, with this licence, and what happens when the connection drops?”

On a GMP site the assessment pays twice. The same list is the starting point for the interface specifications, the data integrity review of each data flow and the network design, so the validation team inherits work that is already done instead of reconstructing it during qualification.

Digitalisation projects rarely fail on the digital part. They fail on the inventory nobody did.