What’s the truth behind Walmart’s failed facial recognition trial?

Was this project really just an effort to flag shoppers who had previously been suspected of shoplifting?

walmart 100585118 orig

Walmart told Fortune this week that it had surreptitiously used facial-recognition software to try to identify shoppers. This was unusual for Walmart, and in fact it was nearly heresy. That’s because the multi-month-long effort failed, and Walmart just doesn’t admit to failure.

But it’s also because it told Fortune that the effort was part of a security program. In retail, security is simply never discussed. Ever.

Now, I’ve been a Walmart watcher for a long time, and the question I have to ask is this: Was this project really just an effort to flag shoppers who had previously been suspected of shoplifting?

The reason this question comes to mind is that Walmart has never had any sort of CRM program — you know, the kind of thing that gets shoppers to identify themselves so their shopping history will be at the retailer’s fingertips.

I think it’s safe to assume that the world’s largest retailer wants to retain that title, so let’s say that Walmart desperately wants a CRM program. CRM was one of many reasons that Walmart, along with Target and several other major chains, created MCX. That’s a mobile wallet effort, theoretically to be offered to shoppers next year. But MCX isn’t here yet, and it’s not clear that it will get any traction when it arrives (although Chase last month gave the MCX effort a huge boost).

The interesting thing about Walmart and CRM is that the chain seems to have a harder time than most retailers in getting its customers to identify themselves voluntarily. (Believe me, I can back this statement up, but it’s far more entertaining to shoot you over to this Onion bit instead.)

With all of that background, that “facial recognition was going to help us keep an eye on shoplifters” storyline becomes a bit thin, doesn’t it? I mean, facial recognition in thousands of stores is going to be expensive. It’s not unreasonable to think that what Walmart really wanted was a way to identify all of the shoppers in its stores, with a thoroughness that would leave most CRM programs in the dust.

In general, facial-recognition systems work in one of two ways. In both cases, the software tries to match faces in the store with a known-face database. But those stored images can either be ones that were captured in-store earlier (and in Walmart’s case, we’re talking about any of its many, many stores), or they can be ones grabbed from social media or other databases.

Now, the thing about those latter types of images is that they usually come with identifying information: names, at the very least.

But if you’re just matching faces to previously captured faces, you’re going to have to work to link those up with names. Let’s say the system captures an image and tags it as Face 88327209. The system has to wait for an opportunity to link that face to a name. Maybe at checkout time, the shopper will use a payment card or a check. Bingo! Product return presents another opportunity. There’s been talk about retailers extending their facial-recognition gear to parking lots; that would give them a chance to match a face to a make and model of a vehicle. A name could be gotten that way with some digging through registration records, but even without it, that’s valuable demographic information.

The interesting thing about having a database of faces that aren’t attached to names is that they can still be very useful. If the system can identify Face 88327209 every time she walks into any Walmart store, identify the items that she puts into her cart and even those that that she looks at but then rejects, the shopper history becomes extensive. So does the marketing database, filled with insights into shopper preferences, which the facial-recognition software should be able to tie to gender, age and ethnicity.

All right, then, if Walmart was trying to accomplish any of these things, why did the effort fail? One problem is that facial recognition has accuracy problems. Even under ideal conditions, facial recognition’s identification precision is iffy. Beards, sunglasses, hats, makeup and many other elements can throw off the systems. And big-box store lighting is hardly ideal for capturing good images of moving shoppers.

Then there’s the privacy problem. If shoppers suspected that they were being tracked as they pushed their carts through the aisles, would they decide to shop elsewhere?

It could well be that that was the real sticking point for Walmart. The Fortune storyquoted an unnamed company spokesperson as saying “We were looking for a concrete business rationale,” but the program “didn’t have the ROI.”

I love that the company spokesperson goes unnamed. I mean, if you’re going to have a company spokesperson, what’s the point in not naming him? Call me cynical, but the explanation that pops into my head: deniability. Think about it: Walmart reveals a secret and admits a failure, two things it is loath to do. It says that what failed was a security program, something it just never talks about. As the Fortune article noted, no other retailers have admitted to any sort of facial-recognition trials.

All of that is out of character for the Walmart I know.

Which is why I have to think that CRM, not security, was behind the trial. If Walmart is worried that this truth could come out some day, it now can point to that Fortune article and say, “No, no, no, we admitted already that we were trying facial recognition as part of an anti-shoplifting program.”

Because if there’s one thing Walmart knows better than keeping a secret, it’s misdirection.

This article is published as part of the IDG Contributor Network. Want to Join?

The march toward exascale computers
View Comments
Join the discussion
Be the first to comment on this article. Our Commenting Policies