Ocado
deep-learning demand forecasting for availability and waste reduction
Ocado runs over 70 million supply chain calculations a day with deep-learning models, orders up to 98% of its stock automatically, has 94.5% of recommended purchase orders accepted without intervention, and keeps food waste at 0.49% of all stock handled.
Key points
- Deep-learning demand forecasting for replenishment and waste reduction.
- In-house encoder-decoder networks on Ocado Smart Platform, more than 70 million calculations a day.
- Up to 98% of stock ordered automatically, 0.49% waste, 94.5% of orders accepted.
- Evidence B, confirmed status.
Objective
Order the right quantity of each fresh product to keep availability high without overstock, driving replenishment by predicted demand rather than by fixed planner rules.
The deployment
Ocado sells groceries online in the UK through Ocado.com and licenses its technology (Ocado Smart Platform) to retailers worldwide. The forecasting engine trains deep-learning models on years of sales data to predict, product by product and day by day, what customers will buy. These forecasts trigger automatic supplier orders and set stock levels in the automated warehouses. For perishables like bananas, the system combines stock, remaining shelf life, and expected demand to avoid waste, and launches flash sales when a batch approaches its date. A planner can override a forecast, with visibility into its accuracy history.
Results Proof B
Figures published by Ocado Group on its own product and editorial pages, consistent across several official pages. Since Ocado sells OSP to third parties, these results function as a quantified platform case study. No independent press reproducing these exact figures, hence B and not C.
How it works
Documented architectureThe stack in detail
- plateforme OSP Supply Chain (Ocado Smart Platform) The supply chain module of Ocado Group's commerce platform: forecasting, supplier order generation, stock and waste management. Sold to partner retailers.
- llm Reseaux encoder-decoder (deep learning maison) In-house probabilistic demand-forecasting models trained on years of sales data, product by product and day by day.
- infra Entrepots automatises CFC Ocado's automated Customer Fulfilment Centres, whose stock levels are set by the engine's forecasts.
How it runs, concretely
For ops teams-
1Data ingestion OSP platform / data team
Historical sales, past shortages, promotions, holidays, delivery slots, and product attributes feed the deep-learning control center.
-
2Demand forecasting AI model
The encoder-decoder networks predict demand per product and per day, with accuracy up to 40% higher than conventional systems.
-
3Order generation AI model / OSP
The system calculates the optimal quantity to order and issues automatic supplier orders; up to 98% of stock is ordered without human input.
-
4Planner control supply chain planner
A planner can override a forecast, with visibility into historical accuracy; over 94.5% of recommendations go through as is.
-
5Waste management OSP
For perishables, the system detects purge risk and triggers flash sales to clear stock before expiry, keeping waste at 0.49%.
Predicted demand per product, combined with the remaining shelf life of stock. If sales history or expiry data is missing, the forecast degrades and the system falls back to more conservative rules, at the cost of overstock or shortages.
How your customers perceive this type of use
Sourced studiesC'est la famille la moins acceptee : 68% des Americains jugent inacceptable un score financier personnel calcule par algorithme et 67% l'analyse video automatisee d'entretiens d'embauche (Pew Research, 2018). La demande d'explication et de recours est massive : 83% veulent savoir quelles donnees l'IA utilise et 91% veulent pouvoir corriger des donnees erronees (Consumer Reports, 2024). A l'echelle mondiale, seuls 46% se disent prets a faire confiance aux systemes d'IA et 70% jugent une regulation necessaire (KPMG / Universite de Melbourne, 2025).
Acceptance conditions
- Transparence sur les donnees utilisees : 83% des Americains la reclament (Consumer Reports 2024)
- Droit de correction des donnees erronees : 91% le demandent (Consumer Reports 2024)
- Explication de la logique de decision : 44% des consommateurs sont plus enclins a utiliser un agent IA si sa logique est clairement expliquee (Salesforce 2024)
- L'acceptabilite depend du contexte de la decision : 50% des Americains jugent equitable un score de risque criminel pour la liberation conditionnelle, contre 32% pour un score financier applique aux consommateurs (Pew Research 2018)
Red lines
- La decision opaque et sans recours sur l'emploi, le credit ou le logement : 45% tres mal a l'aise pour l'embauche, 39% pour le pret, 39% pour le logement (Consumer Reports 2024)
- Le scoring des personnes a partir de donnees comportementales : 68% le jugent inacceptable pour les offres financieres (Pew Research 2018)
Sources: Pew Research Center 2018 · Consumer Reports 2024 · KPMG / Universite de Melbourne 2025 · Salesforce 2024
How to replicate
Inference, not sourcedData prerequisites
- granular sales history per product and per day
- real-time stock data
- product attributes and expiry dates for fresh
- promotions and holidays calendar
Org prerequisites
- in-house data science team or a forecasting platform
- replenishment process able to accept automatic orders
Possible stack
- Ocado Smart Platform (for a partner retailer)
- in-house forecasting on cloud (SageMaker, Vertex AI)
- demand planning platforms (Blue Yonder, RELEX, o9)
The plan, step by step
- Step 1Consolidate clean sales history by SKU and by day, with past shortages, promotions, calendar, and expiry dates for fresh.Deliverable: Usable sales and product-attribute dataset
- Step 2Train or configure a forecasting model on a perishable category where waste is visible, and backtest it against the current method.Deliverable: Accuracy report compared vs existing rules
- Step 3Connect the forecasts to replenishment in recommendation mode: the planner validates each proposed order.Deliverable: Recommended orders in the planners' tool, acceptance rate tracked
- Step 4Automate orders above a confidence threshold, keeping the planner override with visibility into historical accuracy.Deliverable: Share of stock ordered automatically, measured by category
- Step 5Extend to all categories and add waste management (flash sales on batches near date).Deliverable: Availability and waste report vs pre-project baseline
First step: Consolidate clean sales history by SKU and by store, then test a forecasting model on a perishable category where waste is visible.
The same pattern, proven in other industries
Every Man Jack
predictive replenishment timing (predicted next order date)
Maersk
predictive monitoring of refrigerated cargo (customer visibility)
T-Mobile US
Intent-driven decisioning platform: an engine reads customer intent and sentiment in real time, decides the next action, and can execute it, beyond a simple chatbot
Sources
- S1 Forecasting the future: using deep learning to optimise online grocery supply chains Primary archive pending
- S2 OSP Supply Chain | Ocado Group Primary archive pending
- S3 Ocado's AI: Reducing food waste across grocery supply chains Primary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.