How to judge a supplier who promises a lights-out plant

Proposals for unattended operation have become easier to write. They describe a plant where systems watch the machines overnight, an AI layer decides what to do next, and the morning shift arrives to finished work. As a destination the description is not wrong. What a buyer has to judge is whether the supplier knows the route.

Most of that route is not about technology, and the questions that expose it can be asked in the first meeting.

Where the proposal starts

With machines and models, or with exceptions. Most plants considering unattended hours already have capable machines. What stops unattended runs is the situation nobody wrote a response for. A proposal that spends its pages on platforms and algorithms, and a paragraph on exceptions, has its proportions the wrong way round.

With a plan, or with an assessment. A credible first step is an honest assessment of which parts of the operation can run unattended, which need instrumentation or captured knowledge first, and which should stay staffed. A supplier who arrives with a whole-plant plan before that assessment is guessing.

Infographic

Two kinds of lights-out proposal

Promises the destination

• –Starts with platforms and models
• –Plans the whole plant up front
• –Writes to machines or controllers
• –Handles unknown situations with a best guess
• –Little time planned with the night shift

Knows the route

• →Starts with exceptions and how each is handled
• →Starts with a readiness assessment
• →Reads only, actions go through existing controls
• →Stops and calls somebody when unsure
• →Most of the time spent with the shift teams

Questions that tell them apart

None of these questions is technical, and none needs an engineer in the room to judge the answer. They need somebody who knows how the plant behaves at night, which usually makes that person the most useful one to bring to the meeting.

What does the system do with a situation it does not recognise? The only acceptable answer is that it stops and calls somebody. A system that guesses at three in the morning can turn an unknown deviation into a shift of scrap. Listen for whether the supplier treats halting as a failure to be minimised or as the design.

Does it write to machines? A context layer should read equipment signals, MES and quality data, and record what it concluded. Where an action is needed, it should go through the existing control systems under their own safety case. A supplier who proposes to act inside the control path should be able to show you that safety case. If they cannot, that is the answer.

Who will they spend their time with? The judgement that makes unattended running possible sits with the people who work nights. A plan with a few interviews and a great deal of integration has not understood where the work is. Ask how much time with the shift teams the plan contains, when it happens, and who from the supplier will be there at night.

What will the morning shift see? A decision record: what was observed, what was concluded, what was done. Without it, unattended operation cannot be audited by you, by a customer or by an auditor, and so it cannot be trusted.

How is supervision removed? Gradually, one exception class at a time, once the record shows the system’s calls match people’s, with a defined route back. A proposal that goes from staffed to unattended in one step has no way of finding out it was wrong other than by being wrong.

What do they say should not run unattended? Every plant has lines, products or changeovers where the honest answer is not yet, or never. A supplier who cannot name any has not looked closely.

What it costs you to check

A meeting and a careful reading. The answers are often visible in how the proposal is structured before a single question is asked: in where the pages go, and in whether the night shift appears in it at all. If a proposal is thin on these points, that is not a reason to reject it outright. It is a reason to ask for the assessment first and the plan second.

It is also worth asking for the supplier’s own map of what they will not do. A supplier with clear limits — no control-path work without a safety case, nothing unattended without a route back — is easier to hold to account than one who promises everything.

Questions for your own side

A supplier is only half of the route. Before the first meeting it is worth knowing who in the plant will own the exception responses, whether the night shift has been told what the programme is for, and whether the equipment and quality data are good enough to read from. A good supplier will ask about all three early. One who does not ask is planning to find out later, at your expense.

How we approach it

Dark factory context development is built around those answers: an unattended-readiness assessment first, tacit knowledge captured with the shift teams, an exception taxonomy in which anything unrecognised stops and calls somebody, read-only integration, a decision record, and supervision narrowed one exception class at a time.