Today’s retail landscape is inherently dynamic. Hundreds of factors can potentially impact demand, making accurate demand forecast generation incredibly complex. In the face of this complexity, inventory management systems that make pragmatic use of machine learning have proven powerful assets in any retailer’s strategy. Machine learning algorithms automate a system’s ability to combine and analyze immense data streams, identify intricate patterns, and produce the highly accurate demand forecasts that retailers require.
In a July 2020 RELEX Industry Talk webinar moderated by Solution Expert Rachel Dutton, Machine Learning in Retail Demand Forecasting, Data Scientist Henrik Aalto and Project Owner Josh Mann got together to discuss:
- The definition of machine learning
- How machine learning can solve retail demand forecasting challenges
- Examples of how machine learning is being used successfully in retail demand forecasting today
- How to make machine learning work for your retail demand planning
Aalto notes that today, nearly everyone takes advantage of some form of artificial intelligence (AI) and machine learning in their daily lives – such as in media applications that predict, based on learned preferences, what song or movie a user would like to hear or view next. However, he explains, as the definition of machine learning evolves, it is often misunderstood. In particular, retailers need to understand that machine learning is more than just a buzzy term – it is a tool that can be used to drive specific business benefits.
Essentially, machine learning—a subset of AI—is simply a collection of self-learning algorithms. Its strength lies in its benefits over traditional rules-based programming, which typically use data from a fixed point in the past (say, week 36 from the previous three years) to predict what demand will likely be for the corresponding period in the future (week 36 for this year). Instead, machine learning allows a system to combine that fixed data with hundreds of dynamic variables—seasonal trends, local weather forecasts, pricing and promotion decisions, and more—and learn from this input over time, updating its forecast calculations as information changes or more data points become available.
Mann describes how machine learning can be leveraged to analyze multiple demand-influencing data points, such as demand patterns, business decisions, and external factors (e.g., holidays, footfall, or weather) to generate a reliable and accurate forecast. Machine learning can, for example, detect instances of cannibalization during a promotional period, then take that learned result into account in future forecasts to make automated adjustments that can minimize spoilage and maximize sales.
However, says Mann, valid demand-influencing data is not always captured in explanatory variables. When this happens, time-series models may consider these incidences as outliers, which can negatively impact forecast accuracy. Machine learning, though, can apply change point detection to sales data, enabling the system to model step changes in demand even in the absence of defined explanatory variables.
Both speakers agree that while machine learning can readily be applied to demand forecasting, it is a powerful tool that can be holistically incorporated into order, markdown, space, and workforce optimization, ideally within a single, unified system. Machine learning provides a higher level of visibility, aids a more collaborative process in retail planning, and assists planners in making better decisions. As today’s retailers seek to use AI to cut costs, enhance their decision making, and automate their processes, Mann sees a pragmatic end-to-end approach to AI and machine learning as necessary for future success.
To learn more about machine learning and how it is being used today to help solve retail demand forecasting challenges, including real-world use cases, check out the full presentation.
