Skip to content

Can Machines Outsmart Human Mischief?

Embedded vision and AI algorithms are in demand as keys to "smart retail" systems. Will Grabango’s check-out system succeed where Amazon's once-touted “just walk out” service failed?
Can Machines Outsmart Human Mischief?
Grabango believes consumers want to avoid long checkout lines, simply 'grab' and 'go' (Image: Grabango)

Share This Post:

By Junko Yoshida

What’s at stake:
Automation sits at a pivotal intersection between men and machines. Developing automated systems that don’t aggravate consumers is an art not yet perfected. Since the debacle of Amazon’s cashierless checkout systems, the new entrant in this game is Grabango, a startup with different technologies. 

Automation has always been a double-edged sword.

On one hand, automation is a priority for most corporations who strive to lower their operational costs, efforts for which they expect to be richly rewarded by Wall Street.

Some consumers who want to associate with the “cool factor” of autonomous vehicles, smart homes, or barcode checkout systems at the supermarket also welcome automation.  

On the other hand, automation also makes people suspicious.  

When it clicks, automation delights consumers. But when things go wrong (and inevitably, things do go wrong), automation – or the machine that enabled it – frustrates people and invites a storm of backlash.

In this light, I recently convened with Will Glaser, CEO at Grabango. Glaser, who founded Grabango in 2016, has developed a checkout-free shopping system.

It should be noted that Glaser is a seasoned entrepreneur who co-founded Pandora (internet radio and recommendation services) and sold it to SiriusXM for $3.5 billion in 2018.

I was on the phone with him last week to prepare for an “on-stage” interview with Glaser this Wednesday at the Embedded Vision Summit 2024.

Glaser’s topic: Why Amazon Failed and the Future of Computer Vision in Retail

Glaser has a good story to tell. No reporter could resist the David and Goliath contrast between startup Grabango and Amazon, the behemoth of the Internet. Obviously, I fell for it.

Before describing his product, Glaser offered his theory about why Amazon’s automated cashier system called “Just Walk Out (JWO)” — introduced in Seattle in 2018 — got yanked from the market. Then came the contrast. He said that Granbango’s system – with a similar idea but different technologies – stands a better chance at market’s acceptance.  

Although still installed only spottily, Grabango’s system is getting attention from retailers, according to the company. Among Grabango’s clients are ALDI, Chevron, Circle K, Copec and 7-Eleven.

Similar concept, different technologies
Both Amazon and Grabango developed a checkout-free system that retailers can install. Both promote the concept that shoppers can grab stuff off the shelf and simply leave, without lining up or making contact with a cashier. Cameras track shoppers and their purchases, tallying the total electronically and providing it to both retailer and shopper.

Fundamental differences between the two systems are twofold, according to Glaser.

First, the biggest problem with Amazon’s JWO was a huge upfront installation cost. Retailers had to close their existing stores and rip them apart to install vision sensors and new shelves equipped with sensors that weigh individual items. Amazon’s system requires sensor fusion – vision and weight – to recognize products picked up by shoppers.

Second, Amazon’s JWO could not detect seasonal, ad hoc items sold in huge volume only on special days, such as flowers on Mothers’ Day or American flags on Independence Day. The system wasn’t calibrated to cope with single-item surges that happen only on special days.

In contrast, Grabango says its advantage is an easy-to-install automated system. Its checkout-free technology enables retailers to install a ubiquitous network of tiny cameras while staying open for business.

More important, Grabango only uses vision cameras as small as those used in smartphones — without other sensor types. No fusion is needed.

Drawing a parallel to Tesla’s vision-only ADAS vehicles, Glaser said, “Of course, for that to work, your vision algorithms must be very good.” Without offering details, Glaser said that Grabango’s cameras use “cascading [vision] algorithms.” It also uses algorithms for orthogonal errors.

When Amazon launched Just Walk Out, tech media and the tech-savvy audience hailed “autonomous checkout” as the harbinger of a retail revolution.

Several years later, reality has set in and the novelty has faded. Some shoppers are left unsatisfied, when the system failed to recognize items. Others have said they are tired of having to adapt to yet another new technology whose main purpose seems to help retailers save money by letting cashiers go.

After hearing Glaser’s account, I recognized that he is pitching technology differences as the key reasons for his company’s victory over Amazon. This might be partly true, but there’s more to the story.

For technologists who contrive to launch “automation” in public spaces like retail stores, here are three takeaways:  

1. You need machines that can cope with erratic human behavior

After covering autonomous vehicles for a decade, I’ve come to a few instructive conclusions. The first is that the success of any automated system hinges on machines that are designed a priori to anticipate erratic and often unexpected human behavior.

As the San Francisco fiasco with Cruise demonstrated, blaming bad human drivers in other cars doesn’t make an AV any better at adapting to unusual road conditions or the unexpected antics of other drivers.

Glaser acknowledged the reality that Grabango’s autonomous checkout system must adjust not only to normal shoppers but also “atypical behavior.” An indecisive shopper might pick up and replace an item several times before deciding — and then put it back on the wrong shelf place. Even more challenging, Glaser noted, is the shopper determined to fool the system, or the prankster intentionally hiding items from the cameras. “Obviously, we need to have a strategy to deal with them,” said Glaser, “to convince investors.”

2. Go for simplicity

Clearly, Amazon’s JWO system is a web of complexity meticulously developed by technologists eager to solve technology problems with more technology. Combining a vision system with “smart” weight sensors might seem clever, but it tends to both double the elements to make it work and increase the system’s cost.

Given that sensor systems will eventually age or need adjustments later on, easy installation, easy repair and easy updates should be a no-brainer.

3. “Automation” needs to be a choice for a foreseeable future

You can’t shove autonomy down the public’s throat. It must be presented as a choice. One of Amazon’s crucial mistakes with the JWO system was the elimination of an option to keep some traditional checkout counters and human cashiers in stores.

Grabango’s autonomous checkout system, on the other hand, co-exists with flesh-and-blood cashiers, according to Glaser.

Bottom line:
One advantage with autonomous checkout systems is that unlike autonomous vehicles, they don’t need 99.9999999 percent (nine nines) reliability. Current checkout systems manned by human cashiers reportedly achieve 95 percent accuracy, a plateau that Glaser believes his company’s system can match or exceed.

Related video:
Saturday Night Live: Commercial Parodies: Tech

Junko Yoshida is the editor in chief of The Ojo-Yoshida Report. She can be reached at

Copyright permission/reprint service of a full Ojo-Yoshida Report story is available for promotional use on your website, marketing materials and social media promotions. Please send us an email at for details.

Share This Post: