Machine Learning – Omnichannel Demand’s New Sidekick

Forecasting Omnichannel DemandWho can possibly predict demand today? There’s no way you can forecast orders to the granularity of one exact number, either for sales in their aggregate, by channel, or by individual item, across the dizzying array of SKUs and potential demand points in the omnichannel landscape. The best you can hope for is a consensus forecast based on the realistic input of all stakeholders—bolstered by machine learning’s continuous reassessment of, and adaptation to, the shifting omnichannel demand.

You’re never going to find the right answer to planning demand in a fickle, omnichannel, e-commerce age because there is no single, precise number, but rather a range of answers and outcomes—and most importantly, the probability values that attach a figure to the likelihood that each of those potential outcomes will best guesstimate your incoming demand.

Stochastic (“skillful in aiming,” from its Greek derivation), or probabilistic thinking is driving many of today’s planning decisions, like how much demand to plan for in a product launch or promotion, or where to put a new distribution site for last-mile excellence.

Stochastic planning doesn’t shy from the infinite variability of demand—it understands that variability is a natural state, and that what is unnatural is to expect that you can reduce complexity to a single number, and stake critical business decisions and capital commitments on that figure. Stochastic planning operates off the notion that demand behavior can be analyzed statistically, even though it cannot be predicted exactly.

Forecasting Omnichannel DemandOf course, factoring variability into planning is only the first step, the first pass driving your design, production, staging, and logistics decisions for a new product launch. As a professional boxer once famously stated, everybody’s plan goes out the window once you get hit in the face. In business supply chains, that’s analogous to what actually happens in the marketplace when items are available for sale and demand begins to flow in.

At that point, omnichannel demand’s sidekick of choice is machine learning. As soon as demand hits the fan, predictive analytics powered by machine learning begin to identify trends and patterns that reveal what’s selling, in what quantities, and where the demand originates.

Channel data—like sales velocity, product volumes, channel traffic distribution, the effect of promotions—feeds that analytics pipeline to help predict the sales flow and prescribe inventory replenishment, making recommendations that your planners can evaluate through the lens of their business acumen and knowledge of organizational strategy.

This is how you confidently promise availability at the point of sale when a customer makes a commitment to purchase—because you have the information that allows you to establish, and continually tune, strategic inventory buffers to support that order fulfillment.

Stochastic planning goes beyond the idea that you minimize inventory to free working capital and physical space and mitigate the risk of obsolescence. It helps you factor in variability, so that when a product wins customer acceptance, it won’t stumble because of an inability to promptly fill orders.

John Martin

John Martin writes about technology, business, science, and general-interest topics. A former U.S. correspondent for The Economist (Science & Technology), he writes for the private sector, universities, and media, and can be reached at jm@jmagency.com.

Latest posts by John Martin (see all)