Düsseldorf’s order round closed with two moods in the same street, and the split was not a matter of temperament. What separated the positions was whether a supplier arrived on Kaiserswerther Straße able to show what its goods had done on somebody else’s shop floor last season. German wholesale is sorting into brands with a defensible sell-through record and everyone else, and the showroom is the only venue where that record gets renegotiated. Writing in FashionUnited, Cheryll Mühlen reached for the Ferris wheel turning at the Rheinkirmes across the water: steeply up, then abruptly down, depending on when you looked. That renegotiation is the one moment demand-planning software has no observation of.

The clearest statement of the split came from a supplier who had already won it. Florian Wortmann, chief commercial officer at the Herford apparel house Bugatti, told Mühlen that the time of explaining and proving was over, that the company had delivered the performance, and that the conversation now concerned growth and brand visibility. Read that as a description of the market rather than a boast. Wortmann is saying his accounts have stopped auditing him. Every supplier unable to say the same spent the week being audited.

What the audit measures has shifted. Marc Freyberg, who runs sales and marketing at Brax, told the same reporter that classic merchandise had run badly while anything bolder, more modern or newer had gone through the roof. The s.Oliver Group’s replacement of its Black Label line with Lala Berlin is running more than 40 percent above plan. That figure is permission, not momentum: buyers trimming budgets everywhere else gave shelf space to a repositioned label because its early sell-through gave them something to defend internally. Roughly sixty new accounts followed the evidence, not the collection.

An order does not measure demand; it caps what demand is allowed to become.

Demand-planning software is sold on a promise that assumes otherwise. The figure quoted most often in the category’s marketing traces to a 2017 McKinsey paper on Germany’s industrial sector, which put achievable reductions in forecast error at 20 to 50 percent. Nine years of vendor decks later the number travels without its setting, which was factories and industrial supply chains rather than an agent arguing next spring’s buy with a forty-door independent in Westphalia. The same paper claimed lost sales from unavailable product could fall by up to 65 percent, and that second figure quietly concedes the first one’s weakness: it admits the sales record captures what was on the shelf, not what was wanted.

The censoring problem is well documented, and the sharpest measurement of it comes from groceries. FreshRetailNet-50K, a benchmark posted last year and revised in June, annotates every stockout hour across 50,000 store-product series in fresh retail. Its authors found that forecasting on the uncorrected sales record underestimates true demand by 7.37 percent, a bias that all but vanishes once the censored hours are reconstructed. Fresh food is not apparel, and the censoring there lasts hours. In wholesale the equivalent gap lasts a season and is written by hand. A style the buyer declined to take produces no sales record anywhere, and every model downstream reads that silence as evidence.

Forecasting an item with no history requires a comparable, and Düsseldorf is where comparables get assigned. A conditional diffusion model posted to arXiv in April builds cold-start life-cycle forecasts by borrowing reference trajectories from products judged similar to the new one. Its authors tested the method on Intel processor life cycles and open-model repository adoption, not on womenswear, and the transferable point survives the change of subject: a new item is forecast by inheriting an old item’s curve, so everything turns on which old item you nominate. That nomination happens out loud in a showroom, with an agent arguing this jacket is last season’s best-seller with a different collar and a buyer arguing it is last season’s markdown.

The strongest objection is that none of this requires the software to watch the negotiation. A planner needs the outcome, not the transcript, and the outcome arrives as a signed order line with quantities, prices and delivery dates attached. Over several seasons a competent model learns which accounts habitually over-order, and by how much, and corrects for it. For that objection to hold, the order quantity would have to be an unbiased read on what the season would have sold. It is not, because the quantity determines how much can sell. Cut a brand from six styles to three and the survivors often sell through better, the brand’s total falls, and the model logs an improvement in accuracy beside a decline in volume — it has learned the buyer’s caution and filed it as consumer preference.

Most coverage of this order round read the split as a mood story. Mühlen’s own conclusion was that the industry has no knowledge problem, only an execution problem, and that is a generous reading of a market doing something more mechanical. GermanFashion put the clothing industry’s 2025 revenue at 1.4 percent below the prior year, with domestic sales falling faster than exports. A market contracting that slowly tells you almost nothing about who is contracting. The sorting inside the aggregate is the story, and this week it ran on evidence of sell-through.

The question this leaves is where a supplier’s next euro of systems spend should go. We argued in March that wholesale platforms are moving upstream from order capture into decision capture, and nothing in Düsseldorf contradicts that. What the week clarified is where the migration stops. If German independents keep ordering as selectively as they did this week, and Mühlen reports that spontaneous orders have largely stopped, then the scarce asset for a brand will not be a better forecast but an auditable account of what its goods did after they left the warehouse. Suppliers who can produce that on demand will keep negotiating from evidence. The rest will negotiate from adjectives, and keep hearing that the answer this season is no.