March 2017 | Commentary | The Lean Supply Chain

Predictive Analytics Takes Forecasting to a New Level

Tags: Retail, Lean, Logistics, Supply Chain

Paul A. Myerson is Professor of Practice in Supply Chain Management at Lehigh University and author of books on Lean for McGraw-Hill, and supply chain for Pearson, 610-758-1576

Technology users in 2012 generated 2.5 exabytes of data per day, three-quarters of which is text, audio, or video messages. That's a lot of data for companies to review and potentially leverage, and is one reason for the increasing use of predictive analytics, a cost-effective way to filter through and utilize large amounts of data.

Predictive analytics utilizes statistical techniques and models that analyze current and historical facts to make predictions. This distinguishes it from traditional statistical demand forecasting that relies heavily on past demand data.

Predictive analytics goes beyond traditional statistical forecasting methods by producing a predictive score for each customer. Statistical forecasting typically provides aggregate estimates, such as the total number of purchases for an item next month.

For example, traditional forecasting might estimate the total number of cases of laundry detergent to be purchased in a certain region, while predictive analytics tells you which individual customers are likely to buy a specific type of detergent.

The most common best practice today is to use a blend of traditional quantitative forecasting methods integrated with qualitative information such as planner knowledge and intuition, sales force estimates, and customer information, including current inventory levels and point-of-sale data. This is where the art and science of forecasting really come together. The addition of predictive analytics takes it to another level.

Unlimited Potential

More prevalent today are forecasting tools and processes that leverage and combine advanced cognitive algorithms, predictive modeling, and statistical analysis, including information not only from point-of-sale data but also from unstructured text, such as emails, customer feedback, and social media comments. Forecasts need to consider not just general seasonality indices, but also external factors such as seasonal fluctuations or variations in sales volume attributable to customer demographics.

Tools should also be able to include insights through mobile capabilities. There are no limits to the potential applications of predictive analytics for supply chain optimization and forecasting.

Weighing the Risks

Using predictive analytics presents some risks, as companies can tend to rely too heavily on this set of tools. One must understand that predictive analytics involves probabilities and correlation, which are not absolute and must attempt to filter out noise to ensure accurate and repeatable modeling results as well as bad data. You must ask the right questions that are more narrowly focused and test assumptions, eliminating ones that don't hold up under more scrutiny.

Companies must standardize and quantify their data, which can be challenging when using non-numeric and/or dynamic data. There are also privacy and security risks when utilizing data collected from individuals.

Where predictive analytics will head in terms of demand forecasting is still not clear, but we have perhaps seen a glimpse into the future through Amazon's example. In some cases, Amazon ships products before it receives customer orders based upon predictive models that it has developed from customer viewing and purchasing history.