A long-held retail IT fantasy is that complete item-level RFID will be deployed. In theory, this would allow both merchant and shopper to know precisely where every item is, making both inventory and finding that wayward box of strawberry-flavored corn flakes quite easy. But the economics of placing an RFID tag — the cost of which still tends to plateau at about five cents each — have made it nonviable for all but the most expensive products.
Hold that thought for a moment. Now let’s consider Amazon Go, which is Amazon’s attempt at an entirely automated physical store. But instead of RFID tags, it uses cameras and video analytics. It presumably starts with a perfectly accurate snapshot of every item in the store and knows exactly where each one is situated.
Then it watches shoppers and extrapolates from what happens next, whether an item is placed in a cart (assumed purchase intent) or examined and put back. Ideally, it would also note and remember if an item were returned to the wrong SKU spot, the wrong shelf or even the wrong aisle. (How would it go to the wrong aisle? That happens when the shopper puts the item in her cart and then later finds a better alternative in another aisle and just places the now-rejected item wherever she happens to be at that moment.)
One big problem with Amazon Go, as it’s now being used in the initial test stores, is that it’s a “we’ll get it right much of the time” approach to a “we need to get it right almost all of the time” problem. Loss prevention and mobile checkout/payment require a much higher level of precision and accuracy than, for example, inventory and a product-location functionality on shoppers' mobile devices. As we quoted Forrester analyst Brendan Witcher saying last week, it won’t take long before college students figure out that they can “create a huddle and the store will have no idea what they just grabbed.”
Here’s the wacky idea: What if a store used the video analytics approach of Amazon Go as a far-lower-cost alternative to item-level RFID? To be fair, properly done video analytics is expensive to deploy and fine tune, but once it’s fully set up, the recurring costs are much lower than item-level RFID. Removing RFID tags simply slows down the checkout process and adds labor, both to remove the tag and to then reattach it to other products to resume the cycle. In short, reusable is unlikely to be viable.
And item-level RFID also requires complete cooperation from every supplier, which is quite difficult.
The beauty of video analytics is that it requires no help from suppliers. It’s not completely accurate, but as long as it’s better than manual inventory (which it should be) and requires far less labor, it’s a win. Amazon Go’s problem is that it’s trying to use video analytics to solve the wrong problem.
Let’s look at it from the shopper’s perspective. First, to be viable, this requires mobile. It doesn’t depend on an app — the mobile web should suffice — but it should give shoppers a good reason to download the app, which is a marketing win.
Consider a grocery store. Say a shopper is trying to find a very specific item (mixed nuts, but no peanuts and unsalted, or maybe a yogurt pack with no banana flavors). First, this capability would be able to say, “Sorry. The last one was purchased 90 minutes ago. It should be back in the store Tuesday afternoon. Would you like to click here to special-order one so that we can set it aside for you at the Customer Service desk?”
Even better, this approach would allow more precision. Currently, all a retailer knows about an item's availability is what the POS can tell it: We had 12 in inventory yesterday, and 12 were scanned at the POS today, so now we don't have any. With video analytics, the retailer would know when an item has been removed from its SKU spot, but that isn't the end of the story. It would also know whether that item is still sitting in someone’s cart. And it would know whether that cart is standing in a checkout line (and almost certainly unavailable) or merely in an aisle somewhere (probably unavailable, but there’s a still a shot that the shopper will change her mind and remove it from her cart).
And what if she does change her mind and chooses to place that item in whatever random aisle she is standing in at that moment? This app would be able to blow the second shopper’s mind by saying, “We have only one of those specific cans of mixed nuts left, but it’s been misplaced. You’ll find it in the paper goods aisle. Click here and I’ll navigate you to within three inches of its current location.”
Mobile-accessed video analytics could be a wonderful retail technology, but only if it’s used to attack the right problem.