One thing that I find interesting in the AI movement is the creative use of data and algorithms to solve old problems in a new way or to solve new problems.
My colleague, Vikram Srinivasan talked about one such new problem in a previous blog post: AI is Changing Everything We Thought About Forecasting Demand.
Vik discussed how AI is helping improve demand forecasting with additional data and new algorithms. Digging deeper, he mentioned an area that I would classify as an entirely new problem: near term forecasting.
What is near term forecasting? I would describe it as trying to predict what your customer’s next order will be. This problem is even more prevalent for those selling through fast growing e-commerce channels. That is, if you are a manufacturer selling through Amazon, Walmart.com, Home Depot.com, etc., the question you would like to know is what they will order from you next week (or next few days).
Traditional forecasting (even traditional forecasting enhanced by AI) helps you better plan production, inventory, and capacity. And, it tells you when to expect bumps in demand (like for seasonality) and what growth you should plan for.
Near term forecasting is different. Here our goals are short-term and tactical. We simply care about what the next order will look like. We care about this because this is how we get measured and penalized. If the order comes in for 100 and we only deliver 50, we incur penalties. And, if we short the order in one week, we can expect even more volatility in the weeks to come.
E-commerce retailers do provide a longer-range forecast on what they think demand for product will be. But this forecast tends to be at a global level (aggregated across all sites) and not very reliable. Manufactures are often caught off-guard by the very erratic site level orders they receive each week.
In truth, E-commerce retailers often have a machine learning algorithm that place these orders. The algorithms account for not only the expected consumer demand, but also consider things like how many people are visiting the product webpages, whether the item was recently out-of-stock, and how the vendor has been performing in general.
So how can manufacturers keep up? What if they fought fire with fire? Or machine learning with machine learning in this case. If manufacturers could learn to “reverse engineer” these order algorithms they could significantly improve their insights into coming orders and proactively adjust operations to avoid being caught off guard.
AI can help reverse engineer these algorithms. It predicts what the order will be by using the same data the retailer has -or, at least whatever data the e-commerce retailer currently shares with its vendors.
In this sense, this problem is also closely related to the marketing AI problem of providing the next best action or the marketplace AI problem of giving a recommendation. Ultimately, we are looking for our algorithm to tell us what the next action is likely to be.
Near term forecasting requires both internal data (previous orders, previous delivery performance) and data from the e-commerce retailer’s portal (out of stock information, pageview information, pricing changes for the product).
AI is opening up many new problems to solve. And, with the continued rise of e-commerce, areas like near term forecasting will be more and more common.
Mike Watson is Senior Vice President Opex Analytics Division of LLamasoft. With a PhD in Industrial Engineering and 20+ years of experience leading global business teams at LogicTools, ILOG and IBM, Mike brings deep analytics and supply chain optimization expertise to the team. He pairs technical expertise with a personal approach to helping customers design and implement successful enterprise solutions.