Eight Office Seats, 41,000 Cartons a Year: When a Cut-and-Sew Plant Should Stop Paying Per User

The bill that made a plant manager do the math

Thirty dollars a user, a month. Eight office seats — controller, two customer service reps, the production manager, the sourcing merchandiser, the EDI and shipping coordinator, the sample room manager, and you. That is $240 a month, $2,880 a year, and nobody argues about it. It is less than one week of thread.

The argument starts when someone asks the obvious question: the work we actually want done is not office work. It is 1,900 purchase orders a year, 41,000 cartons, 3,400 pages of sew-by instructions that need to exist in Spanish and Vietnamese, and a tech pack question from a line supervisor roughly every 20 minutes on the floor. None of that belongs to a person with a seat. It belongs to a carton, a style number, a bundle ticket.

Per-seat software pricing works when the work is shaped like a person's day; it stops working when the work is shaped like your volume — thousands of small, repetitive reads that belong to no individual employee. That sentence is the whole decision.

Where per-seat pricing quietly breaks in a sewing plant

Walk it through with a real headcount. A 62-employee cut-and-sew operation has maybe 8 to 12 people who would ever sit at a computer and type a request. The other 50 are operators, cutters, pressers, packers and floor supervisors. Under per-seat pricing you have two bad options: license only the office, which means the floor work never gets touched, or license everyone, which means paying for 50 people who will use it twice a month.

Meanwhile the volume work sits there. Reading an incoming EDI 850 and drafting the cut ticket. Checking a customer's routing guide against what shipping is about to do. Translating a revised sewing instruction for line 3 before first break. Answering "what is the topstitch margin on style 4420 in the navy" without pulling the production manager off the floor. Those are thousands of small jobs a month, each one worth about ninety seconds of somebody's attention, and no per-seat licence prices them sensibly.

What actually changed in the open tier in 2026

Until recently the free-to-run models were noticeably worse than the paid cloud ones, and for anything involving a customer's specification you did not want noticeably worse. That gap narrowed hard this year. Moonshot AI released Kimi K3 — 2.8 trillion internal weights, the largest openly available model in the world, and built as a committee of specialists so that only a small share of it wakes up for any single question, which is exactly why it can run on hardware a manufacturer can actually buy. The rest of the open tier moved with it.

What that means for you in plain terms: for the routine, repetitive, high-volume reading and writing that a plant generates, the free-to-run models are now good enough, and running them on your own machine costs roughly 90 percent less than sending the same volume to a cloud service. Frontier cloud models also got about ten times cheaper than they were in 2025, so this is not a cliff — it is a choice with real numbers on both sides.

flowchart TD
  A["New request from the floor or the office"] --> B{"Does this repeat more than 200 times a month?"}
  B -->|No| C["Office staff uses the per-seat tool"]
  B -->|Yes| D["Route to the plant's own model on the box in the server closet"]
  D --> E["Tech pack Q and A for line supervisors"]
  D --> F["Sew-by instructions into Spanish and Vietnamese"]
  D --> G["EDI 850 read into a draft cut ticket"]
  E --> H["Production manager checks before it reaches the line"]
  F --> H
  G --> H

What it looks like on the floor at 7 a.m.

Line 3 starts a new style. The supervisor has a revised sewing instruction from the sample room, in English, written by a patternmaker in a hurry. She puts it in on a tablet mounted at the end of the line and gets it back in Spanish and Vietnamese in under a minute, printed on the bundle ticket sheet. That used to mean waiting for the bilingual floor lead to come off her own line, which cost two people fifteen minutes each and often did not happen until after first break.

At 7:20 the shipping coordinator has four purchase orders that came in overnight. The plant's own model reads each one against that retailer's routing guide and flags the two that specify a different carton label placement than the one the packing line uses by default. She checks both, changes one, and the whole thing takes eleven minutes instead of an hour of PDF scrolling.

At 7:45 an operator asks about seam allowance on a placket. The supervisor asks the same tablet, gets the answer with the tech pack page it came from, and does not walk to the office. That last detail — it shows you which document it pulled from — is what makes a production manager willing to trust it at all.

What running your own actually costs, in people

The hardware is the easy part and the smallest number. The real cost is that somebody has to own it. Here is an honest illustration for a 62-person plant.

Look at what that table actually says. The saving is under $2,000 — nothing. The reason to run your own is not the licence bill; it is that column two buys you unlimited volume. Every carton, every purchase order, every sewing instruction, all day, with no counter ticking. Under per-seat you would never let it read all 41,000 cartons because nobody would authorise that many seats. That is the real difference, and it only pays off if you genuinely have the volume.

The size below which you should not do this

If your shop has fewer than about 15 people who would ever touch it, and your document volume is a few hundred purchase orders a year, buy the seats and stop reading. The internal owner line in that table is the one that sinks small operations. It has to be a real person with real hours — usually the production systems coordinator or whoever already owns the ERP relationship — not "the plant manager's nephew who is good with computers." When that person leaves, an unowned machine in a closet becomes a liability during your next customer audit.

The other honest limit: hosting your own does not automatically make you compliant with anything. It keeps your customers' tech packs and costing sheets inside your building, which is a genuine advantage when a brand's vendor agreement has confidentiality terms with teeth. But you still have to write down what it does, who can use it, and what it is not allowed to touch.

Where a human still signs

Three things should never leave a person's hands in this trade. Lab dip and shade approvals — colour judgment against a physical standard under a light box is not a job for software, and your customer's quality department will not accept it. Anything that goes to a customer in writing as a commitment: ship dates, price quotes, capacity promises. And any translation that carries a safety instruction, because a machine-guarding notice mistranslated on a cutting room sheet is not a productivity problem, it is an OSHA problem.

Use it for drafts and lookups. Let the production manager, the coordinator or the supervisor put their name on the output. That division has held up in every plant I have seen it work in.

Frequently asked questions

Do we need a data centre? Where does this physically live?

One machine, roughly the size of a desktop tower, in the same locked closet as your ERP server and your network gear. It needs steady power and cooling, which in most plants means it should not go anywhere near the cutting room where lint gets into everything. Lint and fans are a real problem; ask anyone who has opened up a shop-floor PC.

Our customers send tech packs under confidentiality agreements. Does self-hosting help?

It helps, and it is often the deciding argument. If the documents never leave your building, the answer you give a brand's vendor-compliance auditor is simpler. Get it in writing in your own procedure document, because "we run it ourselves" said out loud during an audit is worth nothing without a page behind it.

What happens when a better open model comes out in six months?

You swap it, usually in an afternoon, and this is the underrated advantage. With per-seat software you get whatever the vendor decided to ship. Running your own, you choose when to change and you can keep the old one running until you have checked that the new one still reads your routing guides correctly.

Can it read our older paper records — the ones in the filing cabinet?

Scanned costing sheets and old cut tickets, yes, including reasonably bad handwriting in the margins. Faded thermal fax paper from the 1990s, patchily. Do not build a project around retrieving history from a cabinet; build it around the documents arriving this week, which is where the volume and the money are.

A note on where CallSphere fits. The same volume logic applies to the phone. A contract shop's line rings all day with reorder questions, ship-date checks and freight carriers, and it rings hardest during the exact hours the office staff are out on the floor. CallSphere builds AI voice and chat agents that answer the phone and the website chat 24/7, capture who called and what they needed, and book the callback — so the volume that does not fit a per-seat headcount does not turn into a pile of voicemails either.