The Size Curve Gets Guessed 14 Weeks Before the Cut — 2026 Demand Models Read a Lumpy Reorder History

6:40 a.m. in the roll room, and a commitment you cannot take back

The knit mill wants an answer on the 14,000 yards by Friday. Your merchandiser has three windows open: last season's sell-through from the retailer's vendor portal, a size curve copied forward in the same spreadsheet since 2021, and a color split four people argued about on a call in March. Fourteen weeks after somebody signs that commitment, the yardage becomes cut goods in specific sizes and colors — and the moment the spreader lays the first ply, none of it is reversible.

Everyone in this trade knows the shape of the mistake. You do not run out of the program. You run out of mediums in the good color, and you finish the season with 900 extra 2XLs in the slow color sitting on a pallet behind the shipping dock waiting for a jobber to offer you thirty cents on the wholesale dollar.

Demand forecasting in apparel manufacturing means predicting, before you commit to fabric, how many units of each style, color and size will actually sell — and the 2026 generation of models does that from short, gappy, seasonal reorder histories that classical statistics simply could not use.

What a wrong size curve actually costs a cut-and-sew shop

Fabric is usually 45 to 60 percent of the landed cost of a garment, and it is the part you buy first and cannot un-buy. Once greige goods are dyed to your standard and the marker is nested for a specific size ratio, your flexibility is gone. A shop that misses the curve pays for it three separate times.

• The overcut end. Units in the sizes nobody wanted go to an off-price buyer or sit as finished goods carrying insurance and floor space. If you are factored, they also drag your borrowing base.
• The stockout end. The retailer's replenishment order comes in for exactly the size you are short of, and you either take a cancellation or you buy small-lot fabric at a premium and pay to expedite it.
• The relationship end. Fill-rate sits in the retailer's vendor scorecard next to on-time delivery. A season of short-shipping the hot size shows up as a lower rating and, at some accounts, a markdown allowance you negotiate away in December.

Most shops absorb all three as the cost of doing business, because the alternative — guessing better — was not available on terms they could afford.

Why last year's sell-through kept lying to you

The spreadsheet failed because apparel demand history is a terrible dataset by the standards of classical statistics. It is short: three or four seasons before a style is dropped. It is gappy: the style was out of stock in mediums for five weeks, so the history says nobody bought mediums. It is seasonal in more than one direction at once — back-to-school ships May through July, holiday decorated goods peak in a six-week window, and a January warm snap flattens fleece for a whole region. And it is contaminated by promotions the retailer ran without telling you.

Moving averages and seasonal index methods choke on exactly this. They read the stockout weeks as real demand of zero, and they have no way to know that the reorder pattern for a heavyweight hoodie at a college bookstore looks nothing like the same hoodie at a regional chain. So the merchandiser overrode the math with judgment, and judgment carried the season.

flowchart TD
  A["Retailer's Spring 2027 buy lands"] --> B["Model reads 3 seasons of style, color and size sell-through"]
  B --> C["Size curve rebuilt by door type"]
  B --> D["Color split with stockout weeks corrected"]
  B --> E["Reorder probability week by week"]
  C --> F["Yardage commitment to the knit mill"]
  D --> F
  E --> F
  F --> G["Merchandiser signs the cut ticket"]

What changed in 2026: forecasting is now a 48 percent use, pricing a 72 percent one

Two numbers explain why this stopped being a science project. In manufacturing, demand forecasting runs at roughly 48 percent adoption — behind predictive maintenance at 64 percent and quality control at 58 percent, but climbing fastest. On the retail and e-commerce side, pricing optimisation sits at about 72 percent, and inventory work at 66 percent. Your customers are already using these tools to decide what to buy from you — an asymmetry most manufacturers have not absorbed.

What is different about the 2026 models is not that they are smarter in the abstract. It is that they handle the mess. They take a sparse three-season history with holes in it, notice the medium was out of stock in weeks 14 through 19, and estimate what would have sold. They borrow the shape of demand from a similar style in a similar fabric weight when the new style has no history at all. And handed the retailer's promotional calendar and your freight lead times, they produce a size curve per door type instead of one national curve applied everywhere.

The other half is price. On cut-make-trim or full-package quotes, the same models show where your price has room and where you are quoting yourself out of a reorder — which matters more than it used to, because the buyer across the table is running the same math on their own margin.

A Tuesday in the merchandising office, done the new way

The customer's purchase order arrives as an EDI 850 into the apparel ERP — AIMS 360, ApparelMagic, BlueCherry, Infor Fashion, whatever your shop runs. The forecast runs against it that same morning, pulling three seasons of shipment and reorder history out of the same system plus whatever point-of-sale data the retailer shares through their vendor portal.

What comes back is not a chart. It is a size ratio and a color split with a range attached to each one: 8 percent small, 24 percent medium, 27 percent large, 22 percent XL, 13 percent 2XL, 6 percent 3XL, and a note that the medium figure is the least certain because of last spring's stockout. The merchandiser looks at it against the marker efficiency the cutting room can actually hit on a 60-inch goods width, adjusts the ratio to land on clean ply counts, and the yardage commitment goes to the mill with a documented reason behind every number.

What owners actually notice is that the argument changes. Instead of sales and production each defending an opinion, somebody overrides the number on the record, with a reason — and when the season closes you can go back and see who was right.

The arithmetic on one 12,000-unit program

Assumptions, all illustrative: a 12,000-unit seasonal program, wholesale $11.50, closeout recovery $4.20. Today 9 percent of the cut lands in sizes and colors that do not sell through and goes to a jobber, and you air-freight 600 units of replacement fabric at $1.85 a unit of extra freight. Suppose forecasting cuts the mis-allocated share to 5 percent and halves the expedite.

That is $4,059 recovered on one program, roughly $24,000 across six seasonal programs a year, against a forecast that costs a few hundred dollars a month and about three hours of the merchandiser's week. Your figures will differ. The point is that they are countable, on closeout invoices and freight bills you already have in a drawer.

Where the model is wrong and the merchandiser is right

A forecast cannot see a buyer changing jobs. When the account executive at your largest retail customer moves on and the replacement has different taste, three seasons of history become partly irrelevant, and no amount of arithmetic will tell you that. Your salesperson knows it two months before the data does.

It also cannot see a fabric substitution. If the mill switches to a different yarn lot and the hand of the fabric changes, the garment sells differently and the history no longer describes the same product. Your quality manager and your fabric buyer have to flag that in writing, or the model keeps forecasting a garment that no longer exists.

And it will be wrong on brand-new categories. First-season programs have no history by definition; the model borrows from similar styles, which is a guess dressed up neatly. The rule that works: let it own the routine seasonal repeats, where it beats human judgment consistently and cheaply, and keep the merchandiser in charge of anything new, anything where the customer relationship is shifting, and anything where the downside is a cancelled program rather than a slow-moving pallet.

Frequently asked questions

We only have three seasons of clean history. Is that enough?

For repeat styles, usually yes — three seasons is roughly where these models start beating a copied-forward spreadsheet, because they also borrow the shape of demand from similar styles in your own catalog. For a first-season style it is not enough, and any vendor who says otherwise is selling.

Does this need to connect to our ERP, or can we start with exports?

Start with exports. Pull shipment history by style, color, size and week out of AIMS 360 or BlueCherry, run one style whose season already closed, and see whether the forecast would have beaten what you actually cut. That test costs an afternoon and settles the argument before anyone signs a contract.

Our retail customer already forecasts for us. Why duplicate it?

Because their forecast protects their inventory position, not your fabric commitment. Their number says what they intend to order. Yours has to say what yardage to buy 14 weeks earlier, including the reorder that is not on any purchase order yet.

What about pricing — do we get anything if we are a contract sewer, not a brand?

Yes, on quoting. If you price cut-make-trim by the dozen, the same history tells you which customers reorder regardless of a small increase and which walk at a nickel. Most contract shops have never separated those two groups.

A note on where CallSphere fits. Forecasting fixes the plan; the phone is what happens to the plan. When a buyer calls about a reorder, an expedite or a ship date and everyone in the office is on the floor sorting out a cut, that call becomes a voicemail and sometimes a lost order. CallSphere builds AI voice and chat agents that answer the line and the website chat around the clock, take the details, book the callback and hand your account manager a written record. It does not forecast your size curve — but it makes sure the reorder that would have proved the forecast right actually reaches somebody.