Leveraging data and AI to optimize fulfilment operations planning at IKEA

IKEA planners are tasked to continuously ensure adequate operational capacity to fulfil customer needs. In this talk, we highlight challenges, solutions and learnings from developing an in-house digital fulfilment operations planning tool, including a forecasting engine, to support planners. We discuss the use case, organizational setup, machine learning concepts and solution architecture.

Transcript

1. © Ingka Holding B.V. 2022 Leveraging data and AI to optimize fulfilment operations planning at IKEA Picture by Pixabay 14 September 2022 © Ingka Holding B.V. 2022
2. © Ingka Holding B.V. 2022 Loek Gerrits Machine Learning Engineer BigData Republic Alexander Backus Data & Analytics Leader IKEA Retail (Ingka Group) BILLY
3. “Creating a better everyday life for the many people” Group © Ingka Holding B.V. 2022 465 IKEA stores, shops and planning studios in 32 countries 174,225 co-workers EUR 39.8 billion total revenue Figures from Ingka Group Annual Summary FY21
4. © Ingka Holding B.V. 2022 Towards omnichannel retail Physical store Multichannel Omnichannel
5. © Ingka Holding B.V. 2022 The need for a centralized planning capability Planning system Business objectives World state information Demand plan Supply plan
6. © Ingka Holding B.V. 2022 Three opportunities to improve planning Data-driven Optimisation Automation
7. © Ingka Holding B.V. 2022 Inter IKEA Systems (franchiser) Group Digital Retail Cross-functional product teams A digital product organisation Engineering Data & Analytics Product Xperience & Design
8. © Ingka Holding B.V. 2022 We empower co-workers to make informed planning decisions and positively impact IKEA's future Support the co-worker journey with automation and best-in-class analytics Provide user-friendly and high-performant planning tools for all co-workers Enable co-workers to collaborate with the end-end perspective of an omnichannel environment
9. © Ingka Holding B.V. 2022 Towards human-centred AI-assisted planning [1] The starting point Business objectives World state information Plans Human-centred planning system
10. © Ingka Holding B.V. 2022 Towards human-centred AI-assisted planning [2] A step forward Business objectives World state information Plans Forecasts AI Data
11. © Ingka Holding B.V. 2022
12. © Ingka Holding B.V. 2022 Data Machine learning model Forecast 𝑓 𝑥 = 𝑦 Generating forecasts using three components
13. © Ingka Holding B.V. 2022 Historical demand Events T T - 30 T - 2 T T - 2 T - 1 T + 365 𝑓 𝑥 = 𝑦 Data Our model uses different types of inputs
14. © Ingka Holding B.V. 2022 Expressive Non-linear relationships Scalable & flexible Future iterations Generalizable Minimize subroutines Tailormade IKEA Knowledge 𝑓 𝑥 = 𝑦 Model Design considerations for our machine learning model
15. © Ingka Holding B.V. 2022 Uncertainty Orders Order lines Weight Volume Quantity 𝑓 𝑥 = 𝑦 Forecast Multiple forecasts to support specific plans
16. © Ingka Holding B.V. 2022 Dashboard Model registry Model code Train job Scheduler Predict job Evaluate job App Cloud-based MLOps platform for forecasting and optimisation
17. © Ingka Holding B.V. 2022 Towards human-centred AI-assisted planning [2] A step forward Business objectives World state information Plans Forecasts AI Data
18. © Ingka Holding B.V. 2022 Towards human-centred AI-assisted planning [3] The future Plans AI Reviewing Adding Steering Data
19. Key take-aways Empower cross- functional product teams to deliver planning solutions end-to-end 1 Have a long-term strategy to guide your machine learning algorithm and system design choices 2 3 Together with users, evolve stepwise to human-centred AI- assisted planning © Ingka Holding B.V. 2022
20. © Ingka Holding B.V. 2022 Tack! Picture by Pixabay Find us at booth 53 © Ingka Holding B.V. 2022