Jan 20, 2019Each year for the past 10 or 12 years, I have attended the NRF Big Show, which is held in New York. I usually write about what I see there because it is useful for RFID Journal's readers to understand the trends in the retail technology market. Every year, there is usually a buzzword that many exhibitors latch onto. Last year (see Update from NRF's Big Show), artificial intelligence (AI) was big, and this year it was dominant. It seemed like two-thirds of the booths or more were selling some kind of AI solution… or perhaps I should say they were selling AAI—artificial artificial intelligence.
Artificial intelligence can generally be described as a computer science discipline that emphasizes the creation of machines that learn and respond like humans. For example, researchers at DeepMind, an AI company owned by Alphabet (Google's parent corporation) developed a game-playing AI computer system called AlphaZero.
AlphaZero is different from other so-called chess engines, in that it wasn't programmed with the basic theories of chess ("try to control the middle of the board," for example), common chess openings or tried-and-true defenses against popular openings. Rather, it was taught the rules and then played against itself until it learned how to play better than any human or any other chess engine. AlphaZero is true AI (see One Giant Step for a Chess-Playing Machine).
With the exception of Google, IBM and maybe one or two other companies, however, this is not what exhibitors were selling at the NRF Big Show. From what I could glean from those exhibitors—most were very vague about what their solutions actually did—companies simply added algorithms that say "If this and this happen, do this." For example, "If a customer has bought this and this in the past, recommend that" or "If an item can't be found in inventory, recommend this other item." This hardly qualifies as AI.
In addition, AI doesn't work unless you have good, accurate data—and most retailers do not have good, accurate data. We know from studies conducted by the Auburn University RFID Lab that apparel retailers have an inventory accuracy of only about 62 percent (and it can drop below 30 percent for some departments). So, imagine that you deployed a recommendation engine that used AI, and that the system recommended the perfect sweater for your customer—only it wasn't in stock. That AI system wouldn't be worth very much if that happened 20 or even 30 percent of the time.
And what about that awesome AI-powered supply chain optimization system you implemented? It truly learns, but since nothing in your supply chain is tagged, your data is a mess. Only 70 percent of the items that should be delivered to a store are actually delivered, even though your system says everything was delivered because the staff picked the wrong items without realizing it. How can your AI system optimize a solution that is riddled with problems?
I'm sure some of these companies might be developing AI systems that deliver some value to retailers, but with physical store channels and online channels merging, almost all retailers are now dealing with the messy challenges of managing inventory levels. That goal cannot be accomplished with a high degree of accuracy without radio frequency identification. RFID needs to be deployed before data analytics and AI systems are put in place. Someday, retailers—and maybe retail software providers—will figure that out.
Read the second part of this column here.
Mark Roberti is the founder and editor of RFID Journal.