Timing is everything, as the saying goes.
It’s also the story of a startup named Poly, and its visual identification software, says its CEO, Alberto Rizzoli. Two years ago, Rizzoli began developing an AI platform capable of seeing an object – or person – and “visually” identifying it, with a high degree of accuracy.
The use case, then, was the ability to enable smartphones to be more accessible for the visually impaired.
And while today, that may seem like the most logical sort of development project to pursue, it was something that Rizzoli said was considered “an esoteric interest,” given its very specific use and the way in which Poly approached its development.
“At that time, the conventional wisdom was to run everything in the cloud, and we heard a lot of how crazy we were to be building everything locally for a phone,” Rizzoli said. “The cloud is going to just keep getting better and better, and that is where everything is going to be, we were told.”
Time, and a few high-profile data breaches, have a way of changing perspectives. Today, the conventional wisdom – as expressed by luminary tech names like Google, Apple, Samsung and Facebook – is that some types of data are in fact better stored locally than in the digital ether.
And there is a wide range of use cases very suitable to visual recognition technology.
“As we progressed, we kept getting inquiries from retailers,” Rizzoli said. “So instead, we started thinking about what we would do with it if we were to put this tech into the hands of retailers – what would we build?”
What he and the team settled on was the ability for an app and consumer credentials stored on the phone to authenticate the consumer when she enters the store, visually load products taken from store shelves into her virtual cart, build the tab and enable autopayment for those items –without ever having to stop and checkout in the store.
“It’s nice to walk in, pick up an item and let the computer vision do the work, so you can just walk out and know you’ve been charged correctly,” Rizzoli said.
The Right Time And Place Match-Up
There is a reason, Rizzoli said, that 2017-2018 will be the year that Poly’s computer vision technology will get its retail test drive; this is literally the first year the technology has been available in the market at a realistic price point. As recently as a year ago, in 2016, the types of algorithms Poly’s deep learning engine runs to match faces and identities would simply have been beyond what the average retailer was willing to invest in hardware.
And to thank for this innovative advance? Video games.
“The technology is in the graphical processing units – the most typical application for which is the cards they use for video game systems,” said Rizzoli. “Those cards are able to train things a lot faster. Before that technology was available affordably at scale, the personal identification system Poly uses would have required $3,500 worth of computer devices to run a single camera. And that’s a year ago. Today, it requires about a tenth of that.”
Apart from a technological issue, however, Rizzoli noted that the bigger challenge was actually psychological. Consumers – who have shown an incredible amount of comfort with putting microphones into their homes and interacting with them – have historically tended to be a lot more skittish about cameras.
“People associate cameras with surveillance, and that can be scary for people.”
In this regard, Poly owes a debt of gratitude to Apple and its introduction of Face ID. The fact that the data is stored on a device and not in the cloud, and that Apple is strongly associated with vigorous protection of their consumers’ data, went far in “creating trust almost overnight.”
And while that is the flashiest consumer experience, image recognition technology has been showing up more passively across the digital ecosystems. Facebook uses it to flag inappropriate content, while Pinterest and Google let users perform visual searches for products.
“As this service becomes more ubiquitous, it becomes something people find more useful and less intrusive,” Rizzoli told Webster. But, he said, less skeptical and tolerant of error are two very different thoughts, and “the reality is, as soon as we break that trust, the first time the system fails, the customer very likely will not trust us again, particularly if it goes wrong in the transaction stage.”
So, how to deliver the right experience?
The Digital Market
The Poly system makes use of two layers of authentication. First, it looks for the phone – via a Bluetooth signal, either through its own app installed on the customer’s phone or embedded in the retailer’s app.
That, Rizzoli noted, allows an important sorting function to take place. In a city of a few million people, trying to facially ID every individual against every other possible inhabitant is not only impractical, it actually verges on being technologically impossible.
With the phone as a cue, the visual search technology can match the customer against a far slimmer range of images, so the process can work seamlessly. Consumers can walk into their local market and begin grabbing goods as they go.
“Once the consumer is identified, the virtual shopping cart is opened,” said Rizzoli. “Every item they remove from a shelf and keep – as in, don’t put down elsewhere in the store – ends up in their cart. When they have everything they came for, they bag it up and leave. They are then charged via Apple Pay or a card on file with the merchant app, and have a virtual receipt emailed to them.”
No stop, no loss of time for the consumer. And, Rizzoli noted, less lost time for the retailer, as the average merchant loses about 50 percent of their employees’ time to standing idle at a checkout waiting for a customer to go through.
“Now that employee can be helping a customer, or restocking shelves, or doing any number of administrative functions instead of standing at a cash wrap.”
Plus, he noted, it saves retailers money on inventory – because the system keeps a much more accurate count of what is and isn’t in stock on the shelves in real time than the store’s current method of having overnight workers visually scan shelves for missing inventory. That system, besides being prone toward human error, is by design always running behind consumers’ needs and wants.
“The visual ID stuff is a much more futuristic part of the system and the most attention-getting,” Rizzoli told Webster. “But this feature actually brings a ton of savings to retailers, who can literally loose thousands in a day to these kinds of errors. The systems pay for themselves with just that feature.”
And the feature does cost merchants some, as Poly is a SaaS provider that delivers the technology to perform the visual scan and the software tools to actionably use and leverage it for a further purpose.
The Store Of The Future (Is Like The Store Of The Old World)
As consumers are becoming increasing urban due to population shift, the massive suburban stores comprised of thousands and tens of thousands of square feet are a bit out of date. Consumers – particularly urban consumers – vastly prefer a smaller, more intimate shopping experience closer to home, even if they end up shopping more often, than one where they drive out weekly to a big-box retailer.
“Lifestyle preferences have shifted from an urban design perspective, such that people really value having something very local that they can easily access.”
And that enthusiasm, he said, is an opportunity – particularly as Poly can help those smaller market locations offer a more sophisticated digital experience that really defines what local commerce can mean.
“Our idea is that a great grocery store should feel like an extension of your kitchen or of your best friend’s kitchen. You can walk in, pick up what you want and go ahead and consume it.”