Some software companies promise retailers that they can improve their inventory accuracy in stores without the use of radio frequency identification technology. You can understand why this pitch might be appealing: no complex hardware to install and no ongoing investment in tags, with algorithms telling you precisely what’s on the shelves. But does it work?
First, let’s explain what an algorithm is. In computer science, it’s a set of steps a computer goes through to accomplish a task. Artificial intelligence, or machine learning, involves refining these steps to achieve increasingly better results. So inventory algorithms would decrement inventory, just as a normal inventory-management system would, but would also adjust inventory based on historical theft rates and employee scanning error rates, in order to obtain a more realistic inventory number.
Does it work? Not really. Most of these systems simply create new metrics, or redefine “inventory accuracy.” For example, a product in one of these systems might have an inventory error tolerance of plus or minus one and still be labeled accurate, while another might have a tolerance of plus or minus two and be considered accurate.
By manipulating tolerances, inventory accuracy automatically goes up, even if the actual inventory positions haven’t changed. The problem is that this doesn’t really solve the problem. If your system says you have an item in stock and it isn’t there, but you think your inventory is accurate, you might try to sell that product to someone who wants to buy it online and pick it up in the store. When he or she goes to the store to get the item, it won’t be there, which means you’ll have an unhappy customer—and, trust me on this, that customer won’t care that your system says your inventory is accurate.
That’s not to say these systems have no value. Retailers can make some incremental improvements to inventory accuracy by analyzing point-of-sale and returns data more closely, and then adjusting inventory based on historical shrink rates. But you would not want to show online customers the last item in their local store based on this data, because there’s no guarantee that item is actually in stock.
Algorithms can work well when combined with RFID. If you were to take an inventory count via RFID every two weeks for three years, for instance, you would have detailed data regarding the average number of items stolen in each category for every month of the year, along with sales of each item, employee theft and other useful information. If you then fed this information into a good algorithm, it could suggest accurate adjustments to inventory in between actual stock-taking—or it could forecast, with a high degree of accuracy, the projected inventory, as well as suggest ways to improve replenishment.
“Algorithms are good, but they are dependent on the data they you give to them,” says Dr. Bill Hardgrave, the dean of Auburn University. “At the end of the day, there is not going to be a single solution that solves inventory accuracy. It is going to be a combination of technology and solutions depending on the category, store and retailer.”
Unfortunately, many retailers will opt to try algorithm-based solutions in the hope of achieving an easy fix for their inventory problems. Eventually, they will come to realize that these solutions only provide minor benefits and are not good enough to enable true omnichannel retailing.
Mark Roberti is the founder and editor of RFID Journal. If you would like to comment on this article, click on the link below. To read more of Mark’s opinions, visit the RFID Journal Blog or the Editor’s Note archive.