See the waste before it happens.
Forecast fabric offcuts, assess routes for unsold stock, and keep a clearer record of what happens next.
| Model estimate (P50) | – |
| Flat 15% rule of thumb | – |
| Fabric in | – |
Waste decisions start long before disposal.
Cutting plans create offcuts. Unsold stock accumulates across seasons. The information needed to act is often spread across teams, files and suppliers.
- Before the run
The cutting plan
Marker software reports planned efficiency, but the cutting room loses more: ends, splices, defects.
- After the season
The unsold batch
Lines that stopped selling sit in the warehouse until someone decides what happens to them.
- Afterwards
The record
What was considered, what was chosen, and which documents support it, usually rebuilt from email.
Estimate the offcuts.
Model expected fabric waste from planned cutting runs. Compare the estimate with a simple baseline and inspect the inputs behind it.
Prototype trained and evaluated on 8,000 simulated cutting runs. The first pilot goal is to stay within ±15% on real weighed output.
| Method | Mean error, offcut kg | Runs within ±15% |
|---|---|---|
| Marker efficiency alone | 23.3% | 11.1% |
| Flat 15% rule of thumb | 20.3% | 39.2% |
| Model, order data only (this demo) | 8.8% | 82.4% |
| Model, with planned marker | 4.9% | 96.7% |
Find the next route for unsold stock.
Spot the lines likely to stop selling while there is still time to act, then review resale, donation and reprocessing routes for each batch.
| Line | Launched | Units / week, first 12 wks | Stall risk | Units / week, wks 12–25 | Outcome |
|---|
Off-price and resale
Compare expected recovery for the batch with the brand's usual liquidator.
Donation
Record the offer, the receiving organisation and its response.
Reprocessing
A last resort, with the reason and the operator's statement kept on file.
Keep the evidence with the decision.
Organise batch-level records so teams can trace what was considered, what was chosen and which documents support the outcome.
Planned workflow
Illustrative data.
One material story, from plan to outcome.
- 01
Enter production or inventory details
Start with the cutting plan or the unsold batch.
Prototype - 02
Review estimates and options
See projected offcuts, or the lines most likely to stop selling.
Prototype - 03
Record the next step
Keep decisions and supporting information together.
Planned - 04
Learn from patterns
Use recurring material and stock signals to inform the next season.
Planned
Made for the teams responsible for what remains.
Sustainability and compliance
Keep a clearer trail of stock outcomes and supporting records.
Operations and production
See potential cutting waste before the run begins.
Inventory and merchandising
Compare options for stock that has not sold, and see which lines are slowing.
Two kinds of waste, decided at different times.
Offcuts happen at production
Fabric left over from cutting is a material stream. Knowing its likely weight and fibre before the run makes it easier to plan a buyer for it.
Unsold stock happens after it
Finished products that don't sell carry their full cost. Under EU Regulation 2024/1781, large companies may no longer destroy unsold clothing and footwear from 19 July 2026, outside limited exceptions, and must disclose what they discard each year.
Medium-sized companies follow from 2030. This is general information, not legal advice.
"Can we destroy stock if a donation is refused?"
- Art 25(5)Derogations may be set for reasons including non-acceptance of products offered for donation.
- Art 24(1)Operators that discard unsold products disclose numbers, weight and reasons each year.
- Art 25(1)From 19 July 2026, destruction of products listed in Annex VII is prohibited.
Paraphrased from the regulation. Search finds the right article in its top three results for 21 of 22 test questions.
Built around a working prototype.
The offcut forecast uses simulated garment-production data. The stock analytics run on real open retail data. Routing and evidence workflows are being designed with early partners.
mean error on offcut weight, vs 20.3% for the 15% rule of thumb (1,600 simulated runs)
ROC AUC for spotting lines about to stop selling, vs 0.52 for a simple sales-trend rule
automated checks on the data and code, run on every change
Code and method notes: github.com/catliugit/ThreadFutures
Why I care.
Over the last five years it feels like we've been buying more than ever. More clothes, more deliveries, more things worn twice and forgotten. What bothers me most is how normal it's become.
Perfectly good clothes get shredded or burned because that's cheaper than finding them a home, and most of us never hear about it. I don't think the answer is making people feel guilty. I think it's making the waste visible early enough to do something about it, and making the better option the easy one.
I've spent my career turning messy data into things people can act on, and I studied AI because I wanted to understand the technology behind the decisions we're starting to hand over to it. Thread Futures is me pointing those skills at a problem I actually care about: using what we already have, instead of making more of what nobody buys.
A bit about me
I'm Cathy Liu, a data analyst from California, now based in London. I've worked across aviation, banking and healthcare for seven years, building treasury forecasts at Bank of Hawaii and workforce dashboards at Hawaiian Airlines. I recently finished an MSc in AI and Data Analytics at Northeastern University London, where my dissertation tested whether AI models ask the right questions before giving financial advice.
Before you ask.
Is the product live?
Not yet. The forecasting demo and the stock analytics are working prototypes. The routing and evidence workflows are being designed with early partners.
Is the forecast based on real factory data?
No. It is trained on 8,000 simulated cutting runs, calibrated to published industry ranges. A pilot would test it against real weighed offcuts.
Does the EU rule apply to UK brands?
It can, if they place clothing or footwear on the EU market and count as a large company. Medium-sized companies follow from July 2030. Check the specifics with your advisers; this is general information, not legal advice.
Do you buy or hold stock?
No. Thread Futures is software. Stock moves directly between a brand and the organisation that takes it.
Bring us a real material problem.
We're looking to learn from fashion teams managing offcuts, unsold inventory and recovery decisions.
Or email Cathy directly