The future of retail forecasting and the role of data science

Presentation on demand forecasting for M and S

Retailers currently face many challenges in times of increased competition and the need for providing the ultimate shopping experience to attract customers. Many of those recent developments which include same-day delivery or sophisticated, individualised promotions with competitor price matching generate additional operational stress. This well-attended event, organised by the Centre for Marketing Analytics and Forecasting (CMAF), brought together practitioners from leading UK retailers and academics to discuss this hot topic and their slides are now available for download.

The half-day workshop consisted of three talks. First Dr Stephan Kolassa, Data Science expert at SAP Switzerland, gave a talk on the challenges retail business face due to the disruptive environment and how it impacts the demand planning process. As technologies such as Artificial Intelligence (AI), Internet of Things and Social Media, shape our everyday shopping experience a variety of new data types need forecasting. The focus of demand forecasting will, he says, shift to the individual consumer. This forces the organisations to understand their drivers better, but it also blurs the difference between numerical prediction and categorical classification. Stephan stressed that the desire to forecast count data makes some standard error metrics unsuitable research, a neglected problem. A key issue that remains is data quality – what truly is product availability. He concludes that the accuracy expectations managers have for AI or Machine Learning methods are unrealistic given the complexity of the demand pattern retailers observe. Download Stephan's slides.

In the second talk, Professor Robert Fildes, director of CMAF at Lancaster University, discussed what has been researched in retail forecasting and how it is practised. His talk touched first on the challenges retail forecasters face. For example, the difficulty in forecasting store sales on a strategic level with a shifting UK market to online and in-town shopping. Under these circumstances, data on the current store mix offers a biased picture for shop openings and closures; the result is a high reliance on the judgment. As Stephan Kolassa pointed out, intermittency is a surprisingly neglected problem with typically 70% of SKUs showing such behaviour. Moreover, when it comes to forecasting evaluation companies tend to poorly define their KPIs which are not linked to the decision problem. Robert also highlighted the limited benefits social media has and ways to improve new product forecasting. Practice in retail forecasting lags behind the new theoretical developments. The upside is that new data science teams are being established whose role it is to set down processes and methods for demand planning to follow. Download Robert's slides.

The last talk was given by Matt White, Business Lead for Demand and Fulfilment at Marks and Spencer. His talk elaborated the approach M&S has taken to date to forecast demand for fashion products. Over recent months, M&S implemented new software that improved their analytical capability. In particular, he reflected on organisational learnings, and the vision M&S has in the future by in incorporating advanced analytics into everyday operational processes. Download Matt's slides.

Timely to this event, we would like to point out that the International Institute of Forecasting (IIF) currently offers free access to two articles on the topic of retail forecasting, one of which is authored by Stephan Kolassa.

If you require accessible versions of the slides or documents within this news story, please contact us.

Back to News