Parts 1 and 2 were about one station and one piece of paper. This post is about the top of the ladder, so the rest of the series has somewhere to point. It is deliberately ahead of where most shops are. That's the point of a destination.
Lights-out is sold as a machine problem
A dark factory — or a lights-out shift — means the shop runs with nobody in the building: overnight, or a whole Saturday. The pitch for one is usually hardware: the pallet pool, the robot tending the lathe, in-process probing, the tool-life monitor. All real, all available, all installed in shops that still can't run a Saturday without someone in the building.
The reason is rarely on the machine. A machine will happily run all night if it has been told what to run next. Telling it is the hard part. "What's next" is a work order, and a work order is the end of a chain of decisions — what to make, from what material, to which rev, with which fixture, due when, shipping where — that in most shops is made by a person, in an office, during the day.
A dark factory is a shop where that chain runs without the person. Not the machining. The deciding and handing off.
The chain, written out
Here are the handoffs a typical job needs between a customer wanting a part and the customer being invoiced for it. In most shops, each one today is a person, and each one is a document moving from one place to another:
- Intake. The order arrives and becomes structured data — part, quantity, rev, due date, ship-to. (Part 2.)
- Routing. The order becomes a work order with operations in sequence, each with a station, a time, and a setup.
- Material. On-hand is checked, stock is reserved, and if it's short, a purchase is triggered — before the job is released, not when the operator finds the rack empty.
- Tooling and fixtures. The fixture and the tools for each operation are confirmed available and located, not assumed.
- Sequencing. The job is placed in the station's queue in an order that respects due dates and setups — the thing Part 1's tape was doing by hand.
- Program and revision. The right CNC program for the right rev is the one loaded. Not last month's. Not the one with the manual edit on the control that nobody wrote down.
- In-process truth. Each operation reports start, stop, count, and any quality-check results — so the next decision has real data, not a guess.
- Outbound. Packing, labels, the carrier booking, the ship notification to the customer.
- The books. The invoice comes from the work order that was actually run, matched to what actually shipped. No retype.
- Exceptions. When any of the above fails — short material, a probe out of tolerance, a program rev conflict — the right person is told, with enough context to decide from their phone, and the chain waits rather than guesses.
That's the back of house. When people say "automated work orders," this list is what the phrase has to mean. A work order form that fills itself in is step 2 of 10.
Why most "lights-out" projects stall in the office
Notice what isn't on the list: the robot. The machine sits between steps 6 and 7. Every other step is paperwork, and paperwork is where unattended operation stalls — the night shift hits step 3 at 11 p.m., the rack is empty, and the only system that knew it would be empty was the purchasing person's memory. Shops buy the robot and discover they've built a very fast machine that waits for a person. The office was never automated; it was just asked to work later.
The honest sequence
You don't build this chain in one project, and anyone who proposes to is quoting you a year of contingency — padding for a year of unknowns. You climb it, and every rung is a measurement:
- Measure first. The station, the tape, the three numbers. Until flow is visible, automating it automates a mess.
- Structured intake. The PO read once, approved once, born as data. KPI (the number the work is judged by): office minutes per order, keying errors toward zero.
- Floor truth. Scan in, scan out, every station — so steps 5 and 7 stop being guesses. KPI: jobs whose position on the floor the system knows without anyone walking out to look.
- Then the handoffs, one at a time. Material reservation. Program control. Outbound. Each a sprint, each with a number: touches per job — how many times a human had to do something for a job to move — and percent of orders that flowed untouched from intake to invoice.
That last number is the one to watch. A shop that goes from zero to thirty percent of orders flowing untouched hasn't built a dark factory — it's built a day shift that gets a lot more done, which is what many shops actually wanted when they said "lights-out." In our view, the dark Saturday only arrives when that number is close to a hundred, and it gets there one handoff at a time.
Where the chain stops: your wall
Here is the part lights-out pitches rarely mention. Run down the list again and ask, for each step, who is on the other end of the document?
Step 3 is a supplier. Step 8 is a carrier. And somewhere in steps 2 through 7 there is almost always a plater, a heat treater, a grinder across town, or the shop down the road that takes your overflow when the bottleneck station is three weeks out. Each of those is a work order leaving your building — and today it leaves as an email, a PDF, and a phone call to confirm they got it.
A factory can be dark inside its own walls and still stop every time a job has to cross one. Between shops, most work still runs on retyping. Your automated work order hits the plater's inbox and becomes Part 2 all over again, in someone else's office.
So the top rung of this ladder isn't inside the shop at all. It's a work order that can travel — leave one shop and arrive at another as data, not paper, with both sides able to verify it's the same job at the same rev, and neither side having to hand its customer list to a middleman to make that happen. Plain email can't do that. EDI (electronic data interchange, the order-exchange standard big manufacturers use) does some of it, usually through a paid network called a VAN that sits in the middle and charges for the privilege. A spreadsheet of "who can do what" is a start and not a system.
That's where this series is headed, and where our own work is pointed: the shop that runs with fewer hands, then the job that can leave the building without a human re-describing it, then what becomes possible when enough shops can pass work that way. The next posts go rung by rung — floor truth, the measured sprint, when a full overhaul is honestly the right call — before we get to the wall.
For now, take the list. Print it. Walk it with the person who runs your office and count how many of the ten are a human today. That number is your real distance from lights-out, and it has nothing to do with the robot.