Forecasting is an essential tool for planning and decision-making for businesses to maximise growth and profitability in an increasingly competitive environment. Demand forecasting in particular is central to maintaining healthy stock levels for any major retailer, to avoid both going out of stock and overstocking. This challenge is further exacerbated by factors such as sales seasonality, launch of new products and out-of-stock periods in historical data.
Over the last two years at ASOS, we have built our own AI forecasting system from the ground up. Our Machine Learning (ML) models can now accurately predict the future demand for thousands of products across the business at a very granular level. This allows us to support our merchandisers with their replenishment process so that our customers can buy the products they love on our website. Our AI system is now being used by hundreds of merchandisers in ASOS to make more data-driven decisions and improve our stock positions.
In this talk we will share some of the challenges that we faced throughout this journey, how we solved each of them, and the lessons we learned along the way on product management, ML modelling, and MLOps. We will delve into the modelling challenges, how we built scalable data processing pipelines, enabled fast experimentation through the automation of our MLOps lifecycle, our software engineering best practices, and how we integrated our solution into the workflow of our merchandisers to deliver actionable results.
This talk is aimed at anyone with a basic understanding of ML who is interested in learning about AI forecasting at scale from first-hand experience. Attendees will gain a comprehensive understanding of the steps involved in creating and maintaining an effective AI demand forecasting system to deliver business value.
Technical level: Technical practitioner
Session Length: 40 minutes
Supported by
