Dark Factory: From Shop Floor to Back Office

In Oshino, Japan, there is a factory that has been running without lights since 2001. No lamps. No people. Robots building robots, around the clock, for 30 days without interruption. The company is called FANUC. This is not a proof of concept or a press release. It has been their production model for over two decades.

It is called "lights out manufacturing". You do not need lighting when no human is in the building.

And now the same thing is happening to your office.

What "lights out" actually means

The term comes from manufacturing and describes a factory that can run entirely without human presence. Not "with a skeleton crew". Not "mostly automated". Entirely. Without. People.

FANUC is the most famous example, but they are far from alone. MTU Aero Engines runs autonomous airfoil production for 66 hours without anyone touching the equipment. Xiaomi produces 10 million phones per year in its Beijing plant with virtually zero human involvement. The WSJ and France 24 report on China's "dark factories" as an industrial phenomenon, not a future scenario.

What makes these factories possible is not that they automated individual steps. It is that they automated the entire chain. Every step knows what the next step needs. Every machine can handle deviations within its scope. The system runs on rules, sensors and feedback loops, not on supervisors walking around checking things.

The back office is the new factory floor

Here is the point most people miss.

Look at what actually happens in an e-commerce company's back office. Product data needs enriching in PIM. Customer data needs segmenting in CRM. Orders need routing in OMS. Returns need classifying, crediting and restocking. Campaigns need setting up, evaluating and adjusting. Price changes need syncing across channels.

It is repetitive. It is rule-based. It is measurable. Those are exactly the properties that define work that can run lights out.

Yet people are still sitting there clicking. Not because the tasks require human judgement, but because nobody has built the system that makes the clicking unnecessary.

Four levels from manual to autonomous

TMO Group has a four-stage model that describes where companies sit on the path to autonomous operations. It is brutally honest.

Level 1: Manual operations. Humans do everything. Data is copied between systems. Reports are built in spreadsheets. Returns are handled via email. Most e-commerce companies that claim they have "come a long way with digitalisation" are here. They have merely replaced paper with screens.

Level 2: Automated but human-dependent. Integrations exist between systems. Some processes trigger automatically. But a human must start, monitor and approve every step. The flow stops when someone goes on holiday, calls in sick or forgets. This is where the majority of mid-size e-commerce businesses sit today.

Level 3: Intelligent automation. The system makes simpler decisions on its own. A return policy is applied automatically based on a rule set. Prices adjust according to stock levels and competitor data. But a human still needs to handle exceptions and define the rules. A few companies are here.

Level 4: Fully autonomous operations. The system runs around the clock without human involvement. It handles both standard cases and exceptions. It escalates only what falls outside all known patterns. This is FANUC, but for office work.

Almost nobody is at Level 4. Not because the technology does not exist, but because they have not thought of their office as a factory.

The numbers behind the shift

McKinsey estimates that agentic commerce could orchestrate $3 to $5 trillion in global retail revenue by 2030. The market for AI agents in e-commerce is growing from $7.7 billion in 2024 to $282.6 billion by 2034, a compound annual growth rate of 54.7 per cent. Shopify's AI-powered search orders grew 15x year-on-year through 2025.

These are not future predictions in the usual sense. They are measurements of what is already happening, extrapolated forward.

Retailers implementing AI automation report 20 to 35 per cent lower operational costs and 15 to 25 per cent higher revenue. Not by firing half the staff, but by removing wait times, error handling and manual bottlenecks that cost money every day without showing up on any dashboard.

What a "dark office" looks like in e-commerce

Picture this. It is night. Nobody is at the office. But the business is running.

The PIM system receives new product images from a supplier. One agent checks image quality, background and metadata. Another agent writes product copy in three languages based on the product specification. A third agent publishes to all channels with the correct price matrix per market. Time from supplier delivery to publication: 4 minutes.

The OMS handles orders coming in from six different channels. A routing agent selects the optimal warehouse based on shipping cost, delivery time and stock availability. Split orders are handled automatically. The customer receives tracking information before sunrise.

The returns system receives a return request. The agent classifies the reason, applies the correct return policy, sends a return label, credits the customer and updates stock levels. If the product is damaged, it is routed to secondary stock or disposal without anyone needing to make a decision.

The CRM identifies a customer who has purchased three times in the past six weeks but never bought from a particular category. An agent creates a personalised offer based on the customer's behaviour and margin targets. The offer is scheduled for optimal send time.

All of this runs without a single person clicking a single button. Just as FANUC's robots run without lights.

Why "dark office" does not mean "no people"

This is not a story about replacing all employees. FANUC's factory still has engineers. They do not build robots by hand. They design the processes, set the quality requirements and solve the problems the system cannot handle on its own.

The same applies to the office. People move from clicking to specifying. From executing to defining what quality means. From being inside the process to standing beside it.

It is a shift in what work means. Not a shift in whether it is needed.

Lights Out: what we already do

At Spinout, we call this service Lights Out. We build autonomous processes for e-commerce companies. PIM workflows that run without human involvement. Returns handling that classifies, credits and restocks around the clock. Order flows that optimise routing in real time.

We are doing to office processes what FANUC did to manufacturing in 2001. The difference is that we did not need 20 years to prove it works. The AI agents already exist. The protocols already exist. The integrations already exist.

The only thing missing is the decision to turn off the lights.