The shift handover is your most expensive undocumented process

A shift supervisor finishes at six. What he knows about the fault on machine 4, the batch stuck in QA and the supplier who did not show up lives in his head and in a WhatsApp group. The next shift discovers all of it again, from scratch. That rediscovery is not a communication problem. It is an unpriced operating cost that you pay three times a day.

The use case

An agent sits on the two places the information already lives: the WhatsApp group the supervisors actually use, and the ERP. At the end of each shift it produces a written handover with three sections. What is open. What is blocking. What needs a decision before tomorrow.

Nobody logs into anything new. The supervisor keeps typing where he already types. The artefact is generated, reviewed in ten seconds, and passed on.

What it is worth to the business

You stop paying twice for the same fault. A problem diagnosed at two in the morning does not get re-diagnosed at ten. More importantly, the knowledge of a twenty-year supervisor stops being a personal risk and starts being an asset of the plant. When he takes leave, or retires, the shift does not degrade.

There is a second effect that shows up later. Once handovers are written down, patterns become visible. The same machine appearing in eleven handovers in a month is a maintenance decision that nobody could see when the information was verbal.

Four decisions you have to make

1. Who owns the handover. Production manager or plant manager. Pick a person, not a department. A handover without an owner is dead within a month, and the failure will look like a technology failure when it was an ownership failure.
2. What goes in and what stays out. Three categories, not twelve. Every category you add halves the chance that anyone reads it. Start with open, blocking, needs-decision, and resist the rest for a quarter.
3. Where it lives. In the tool they are already in, or in a system you will have to push them into. The honest answer for most plants is the first one, and the honest cost of the second one is a training programme you did not budget for.
4. What happens when someone does not hand over. An alert to the manager, or nothing. Both are legitimate. Deciding nothing is the option that quietly kills it.

Where to start

One line. One shift pattern. Two weeks. If the supervisors on that line stop the pilot at the end of two weeks, you learned something cheap. If they ask for it on the second line, you have your business case and you did not need a slide to build it.

Technical footnote

Read from the messaging channel, read from the ERP for the objective state, compose with a language model, write back to a record. The interesting engineering is not the model. It is deciding what the agent is allowed to assert on its own and what it must quote from a system of record. Anything it cannot source, it should say it cannot source.