The fabric lost at the sampling stage fell by 90 percent.
That is the figure Heilan Home attached to its name when its demand-insight flexible fast-response apparel smart factory was selected into the second national batch of lighthouse plants. It is the first garment factory in the country to make that list, and chairman Zhou Lichen attended the naming ceremony in Nanjing.
Reading the factory's name
The official designation is long, and every word in it is a claim about where the software sits. Demand insight comes before design. Flexible fast response is a promise about batch size. Apparel names the industry, which matters because apparel is the hard case disguised as the easy one.
A car plant holds one architecture for years. A clothing line renegotiates its product every season, on a margin that cannot absorb a bad forecast. If generative tools pay off anywhere, they should pay off here first. If they do not, the case elsewhere gets weaker.
Seven thousand five hundred doors
The disclosed base is more than 7,500 stores in China and abroad, over 200 million online followers, and more than 1 billion records of RFID and 5G-A passive IoT data.
Those three numbers read as scale, and on that reading they are unremarkable for a chain this size. They are better read as an input layer.
Passive IoT is the part that makes the layer usable. A tag that draws no battery can be attached to a garment rather than to a pallet, so the count happens at the rack instead of at the warehouse. A billion records is then not a marketing metric but a training set: what sold, where, at what price, on which day, and what did not move.
That is the signal a pattern cutter needs and almost never gets in time.
Seconds, not days
On top of the data sit an apparel-specific large model, generative design tools and a process knowledge graph. The company describes a dual-engine cross-modal system that turns a design brief into an image in seconds.
Seconds changes the shape of a design meeting. When a rendering is nearly free, the meeting produces forty options and discards thirty-eight, and the scarce resource stops being the sketch and becomes the judgment about which thirty-eight to kill.
The knowledge graph is the quieter half. A generated image is a picture until something connects it to a stitch, a machine and a tolerance. That connection is what turns a concept into something a sample room can actually cut, and it is the part that takes years of process records to assemble.
Where the ninety percent comes from
Sampling is the step between a design and a bulk order. A sample is cut and sewn before anyone commits fabric to a production run, and the material consumed there is pure cost until an order is confirmed.
A 90 percent reduction in that loss is strong for a specific reason: it is measured in meters. A material saving can be checked at the cutting table by anyone with a scale. A productivity gain of 90 percent cannot be audited the same way, which is why material figures travel further between manufacturers than efficiency figures do.
There is also a sequencing argument. Once design iteration costs almost nothing, the binding constraint moves downstream to sampling and fabric. The saving lands exactly where the new bottleneck forms, which is why the seconds and the 90 percent belong in one sentence, not two press releases.
Three hundred thousand is another ledger
The factory's disclosure also puts about 300,000 jobs across headquarters, stores and suppliers upstream and downstream.
That number is doing more work in the coverage than it can bear. It is a stock, not a flow. It counts the people currently attached to a network of stores and suppliers, not the change in headcount that the factory caused.
It is also not a factory payroll. The figure spans retail and supplier plants the company does not own, so it cannot establish that automation was neutral inside the plant, and it cannot establish that it destroyed work either. It sits outside the boundary of the claim in both directions.
This is the distinction worth holding onto. Efficiency and employment are measured on different clocks. A 90 percent material saving shows up in a quarterly cost line, visible and auditable. Employment shows up in hiring decisions taken by thousands of independent suppliers over several years, and almost none of that is reported.
What the company can fairly say is that it automated sampling while still running a 7,500-store network that is itself labor-intensive, and that it still employs about 300,000 people. What it cannot say from this disclosure is that nobody lost work.
What would settle it
Three numbers would move this from a good story to evidence. Headcount in the sample room, before and after. The count of supplier firms and their payrolls. And a sell-through or markdown rate that shows whether better forecasting actually reduced unsold stock.
None of the three is in the release, and their absence is normal. Firms disclose what flatters them, and a waste reduction flatters more cleanly than a hiring table.
My judgment is that the 90 percent is probably real and the 300,000 is probably irrelevant to it. The first is a fact about fabric, measured where fabric is cut. The second is a fact about scale, assembled from stores and suppliers that the factory does not directly control.
Put side by side, they look like a balance. They are not. One describes what a machine now saves. The other describes how many people the business still needs to sell what the machine designs. The question that matters, whether the first can keep growing without shrinking the second, is precisely the question the disclosure leaves open.
*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*
