Reading between the lines of AI hype reveals a retail demand forecasting gem

27 August 2020 | 112 Shares

Source: Shutterstock

Turn on your TV or fire up a web browser and, within just a few minutes, you’ll come across a commercial for a product with some element of artificial intelligence inside it that makes it special.

In consumer goods, it could be a low-end smartphone or a high-end refrigerator. You’ll find plenty of software in commercial settings that “leverages the power of AI” in every sphere of human commercial activity. But of course, most claims of “AI” are little more than hyperbole, comprising a few algorithms that could have been coded by an undergraduate Computing Science freshman. The reasons are obvious enough, especially considering the fact that a business is literally better-valued if it claims to be using artificial intelligence.

Therefore, as retail professionals, how do we look beyond the claims of dubious marketers to actually apply a little machine muscle to operations? After all, modern retail is fantastically data-rich, with technology running underneath just about every aspect of the business. All of this produces information too quickly and in too great a quantity for humans to draw much value from, other than to skim off the surface of the “data lake.”

Real-life uses of (real) artificial intelligence technology in retail are starting to happen, however. As an example, WHSmith’s  retail outlets have been using AI-based forecasting software from RELEX Solutions to help manage demand for their airport stores. The retailer stocks a big inventory of SKUs, including many perishables. By ingesting airport traffic numbers, self-adapting routines predict demand with a degree of accuracy not humanly possible.

Modern retail is so complex that even in a single need-case — in this case, demand management — the number of potential influences is enormous. To understand the full data picture, we need to consider two broad areas of influence on buying patterns: internal and external factors. It’s impossible to formulate a complete list of influences in either case, but consider a handful:

Internal influences on demand
Any in-store element such as floor layout, planograms, and display choices.
Any business decisions regarding pricing and discounts.

External influences on demand
Any occurence that affects a store location, like a sporting event, the weather, or roadworks.
Fashion seasons, trends, influencers’ choices, and discount promotions elsewhere.

In the case of airport retail, dramatic changes to travel volume resulting from COVID-19 restrictions has certainly proven a challenging external factor, one that’s problematic to forecast accurately. Machine learning-based forecasts would provide a valuable advantage for demand planners struggling to adjust to demand fluctuations as travel restrictions evolve.

Asking even highly-experienced staff to consider all the above would be like asking someone to juggle with thousands of objects. And never sleep. But in today’s age of everything technological, data that describes all of those factors is available. It’s “just” a case of sourcing the information, ingesting it as part of the overall picture, and coming up with suggestions based on that processing. (The use of quotes — “just” — in the previous sentence is very deliberate.)

While there’s no artificial intelligence app out there capable of interpreting every conceivable possibility, within retail demand forecasting, products have been specifically developed to address that single area — like RELEX’s. It’s through carefully designed modeling — for example, clever regularization — that machine learning’s self-improving algorithms can be used. But even then, there are no 100% autonomous solutions that can be fired up to provide miraculous answers. Retailers still need to employ their own personnel’s experience to configure and oversee the software, review exceptions, manage novel situations in which no previous data is available, and act a a guiding hand.

The outcomes for WHSmith were better use of very limited storage space on the ground (split delivery management oversight was improved), better control of stock levels (previously, its suppliers had essentially managed deliveries and stock), reduced wastage, and presentation of a better, fresher product to its customers.

The RELEX software used metrics such as airport traffic to better predict demand and better manage the business but, of course, executive control was left to the human beings on the ground at the sharp end of the business.

“The key result is that the 12 weeks pre- vs. post-implementation showed a significant reduction in spoilage [..] at the same time availability increased,” the store chain’s Merchandise Controller said.

In a time where claims about technology go unchallenged, it’s important to mark those instances where cutting-edge thinking is making a significant difference in people’s working lives and the bottom line for businesses. This is one of the many cases in which, by using carefully-focused machine learning routines backed by specialist professionals with significant experience in their chosen retail field, Real-life, working examples make a refreshing change from suspect claims of artificial intelligence poised to start replacing experienced professional staff. The bottom line is that in the case of WHSmith, the company, its staff, and its customers all win.

As we noted above, retail — particularly on-line retail — is a data-rich , so demand-forecasts are only the beginning. There are many other areas that can also be guided by autonomous, self-improving algorithms. Flowing outwards from demand forecasting come automated replenishment systems, planogramming and floor planning, markdown strategizing, and even human resource management (follow the links to read more).

There’s too much ground to cover all the possibilities in a single article (so watch this space), but we recommend you start by reading about discrete use-cases of AI in retail demand forecasting and management in this paper. It’s genuinely one of the most accurate pieces on artificial intelligence in a business environment we’ve read for a few years and comes with a “no hype” guarantee.