23%
Perishable waste reduction
4.2%
Forecast accuracy (MAPE)
£480K
Annual waste cost saving
↓ 31%
Stockout rate
The challenge
Overproduction and spoilage cost the business £2.1M annually. Forecasting was done on spreadsheets updated once a week.
FreshCore Foods produces chilled ready meals for UK supermarkets, operating on 48-hour production cycles with perishable ingredients. Their weekly demand forecasting process — done on spreadsheets by a single analyst — could not respond to the signals that actually drive demand: weather, promotional calendars, competitor activity, school holidays. Spoilage and overproduction cost the business £2.1M in the prior fiscal year.
Our approach
Signal identification
We analysed three years of sales data alongside 40+ external signals — weather forecasts, school calendars, public holidays, supermarket promotional calendars, and social media sentiment. Six signals proved to be consistently predictive; the rest were noise.
Forecasting model
A gradient-boosted ensemble model was trained on the identified signals, with separate models for each SKU-retailer combination. The model generates a daily production recommendation for each of the next 7 days, with confidence intervals that operations managers can act on.
Operations integration
The forecasting output feeds directly into the production planning system. Operations managers receive a morning briefing each day with recommended adjustments versus the baseline plan, with the key signals driving each change explained in plain language.
“Our production manager was sceptical of AI for the first month. By month two he was checking the system before making any production call.”
Rachel Obi
Supply Chain Director, FreshCore Foods
Ready to move forward?
Let's navigate complexity together.
Every great partnership begins with a conversation. Tell us what you're building — we'll show you how to get there.
Begin a conversation