Logistics is becoming an increasingly complex phenomenon, and it will likely continue to increase in complexity as the impacts of climate change, pandemics, and high customer demands become more challenging. Companies will need to identify areas of their business that they can optimize and focus their efforts on improving those areas through advanced analytics and technologies. Analytics can help provide insights on all aspects and phases of one’s business. Obviously one of the first applications of analytics in logistics pertains to demand planning, which is the focus of this article.
The Role of Demand Planning in Logistics
Demand planning is a process that requires demand signals (data) to predict future demand. The financial outcomes of any logistics business will be driven by the organization’s ability to develop a robust and resilient demand planning process. In mature supply chains, demand planning is supported bystatistical, machine learning, or AI models. Models can be simple (e.g.,based on a moving average) or they can be complex (e.g., recurrent neural network algorithm). The type of model selected will depend on the amount and type of data available as well as the nature of the sales patterns (e.g., presence of seasonality, trends, noise components).
Machine learning and AI models are particularly useful when a substantial amount of data exists in terms of history (number of years available), granularity (WRIN, store, and day level), and features that can serve as demand signals. At a minimum, one would need historical sales at the level of analysis desired. However, additional demand signals such as external local events, promotions or marketing tactics, school calendars, and weather events can provide substantial explanatory and predictive value. In fact,we found that external local events explained more than half of the forecast error for a given store that had erratic foot traffic patterns. However, it is important to notethat obtaining and managing much of this data can be difficult, and it can be costly to purchase data from vendors.
One key consideration in selecting a model for forecasting demand is the need for model explainability and interpretability. With black box models, it can be difficult or impossible to explain what is happening under the hood. Similarly, if the model output needs to be interpretable, then the data scientist is limited to models that provide cause and effect insights (a decrease in price of 20% results in a 30% increase in sales). Another consideration is the technology environment and infrastructure used to support and scale analytics. There are many options available such as Microsoft Azure, AWS, Google Cloud, etc. Regardless of the platform selected, importance should be placed on the tech ecosystem’s capability for scaling analytics.
Considering Trade-offs Associated with Advanced Analytics and Technologies
While analytics are a requirement for increasing organizational capability and growth, there are trade-offs to consider. For example, building and sustaining an analytics or data science team can be difficult and costly. In addition, the technology costs associated withplatforms, data acquisition, data storage and compute need to be considered. In other words, there is no free lunch.Understanding the ROI associated with implementing data analytics is a critical component of the business case and understanding the concept of diminishing returns will help guide decisions around the amount of effort to spend on improving a demand forecast.
"One key consideration in selecting a model for forecasting demand is the need for model explainability and interpretability"
Regardless of the position taken (build, buy, do nothing), organizations that do not leverage advanced analytics andtechnologies will inevitably fall behind, lose market share, and struggle to survive.Improving the organization’s demand planning process is likely a good place to start in terms of improving organizational effectiveness and financial outcomes. Therefore, projects aimed at identifying demand signals and testing out new models and technologies to support demand planning should be part of an organization’s strategic vision and plan.