Retailers should rely on human expertise to guarantee the accuracy of the models they use to make store decisions—not put blind faith in “black box” software supposedly driven by AI and machine-learning, advises Douglas Bennett of A&G Real Estate Partners in a new analysis blog for chainstoreage.com.
In the piece (“Why Retail Sales Models Still Require a Human Touch”), Bennett notes that tech firms increasingly offer store recommendations without being able to transparently explain how their models arrive at these results. The executive is Senior Managing Director of A&G’s Financial Reporting & Property Analysis Group and a former strategist and analytics leader at Ashley Furniture and Toys “R” Us, among other companies. “Typically, the vendor leans heavily on terms such as ‘AI,’ ‘machine-learning,’ ‘sophisticated algorithms,’ or some combination of the three,” Bennett writes.
He goes on to provide real-world examples that illustrate the importance of using transparent models—as well as asking hard questions about the assumptions that go into them.
In running a regression model on behalf of a retail client, for example, an A&G analyst noticed a nonsensical correlation: Stores in areas with growing populations appeared to be associated with lower sales. “The analyst immediately suspected the model was picking up a proxy for something else,” Bennett writes.
The retailer had tentatively expanded outside its home base in the Northeast, where its stores performed well despite shrinking populations. Some critical thinking and human due diligence revealed that the scattered stores in these new high-growth regions lacked the name-recognition, operational consistency, and distributional support of that retailer’s older locations.
“It was these latter factors that were responsible for those lower sales volumes,” Bennett writes. “Lacking the intuition of an analyst or real estate executive, a black box could miss this distinction and encourage the retailer to steer clear of high-growth markets—terrible advice.”
He also provides examples of how missing information about key differences between stores or restaurants in a company’s portfolio (prototype size, merchandise lines, drive-thru lanes, etc.) can skew a model’s results.
“On multiple occasions, we have seen retailers rely on legacy models that omitted such important store classifiers,” Bennett writes. “Once we corrected for that, these clients were able to pursue previously overlooked growth opportunities.”
After pointing to some of the methodologies that trained data scientists use to ferret out bias in their models, Bennett advises retailers to take a balanced, collaborative approach to store decisions.
In his view, a good analytical team will want to:
- meet with c-suite executives to gain a solid understanding of their real estate goals, ideal demographics and target markets;
- deepen the conversation by engaging in detailed drilldowns involving the finance and real estate groups; and
- ask for the retailer’s past analyses and the assumptions that went into them.
“A deep dive could show that the company needs to rethink longstanding views about its best customers and markets,” Bennett explains.
When both sides ask the right questions, the model is far likelier to be powerfully predictive. “Such transparent approaches can cast light on distortions that otherwise would have stayed hidden in that proverbial black box,” he advises.