Case studies / Industry · Supply chain

How a thread manufacturer forecasts 49,000 SKUs at a 98% service level

A long-tail catalogue, next-day delivery and lead times of up to 90 days. Probabilistic forecasts and adaptive safety stock, tested against the company’s real stock.

  • 49KSKUs forecast
  • 98%service-level target
  • −37.5%stock in the Q1 2025 backtest
  • 4–90days of lead time handled

The situation

The company makes industrial sewing thread and sells it worldwide. Its catalogue is a long tail of 49,000 SKUs, many of them ordered rarely, and customers expect next-day delivery.

The problem

Replenishment takes between 4 and 90 days and comes with minimum order quantities, so stock is decided long before demand shows up. Planning relied on a static forecast in SAP, recalibrated weekly and adjusted by hand.

What we built

  1. Probabilistic forecast per SKU

    A demand forecast for every SKU that gives a range, not a single number.

  2. Adaptive safety stock

    Safety stock is set SKU by SKU to meet the service-level target, given each lead time and minimum order quantity.

  3. Simulation on real data

    Replenishment is simulated against real demand and the actual stock in SAP before anything changes in the warehouse.

The results

Stock needed to meet the 98% service level · Q1 2025 backtest, actual stock = 100
Actual stock
January−37%
February−28%
March−47%
  • In the Q1 2025 backtest, the model met the 98% service-level target with 37.5% less stock on average.
  • On one example SKU, average stock went from 815 to 323 units.
  • It outperformed the existing SAP forecast, which is recalibrated every week, while forecasting 40 days ahead.

Backtest on 50 SKUs with undisturbed history, comparing the stock the model would have held with the actual stock, against real demand. SKUs that miss the service-level target are excluded and only the central 70% of outcomes is counted.

Client anonymised. Full case file and reference available under NDA.

Tell us the problem.

Start with a free Value Map: three hours with your team, and a fixed-price plan for your first system five days later.