Online shopping has the potential to be so much smarter. If you stumble across a rug or lamp you like in a photo, shouldn't it be easier to track down where to buy it? One home design website is aiming to do just that, with the help of artificial intelligence.
Palo Alto-based Houzz is a platform for people who want to remodel or redesign their homes and are looking for inspiration. Consumers and design professionals alike can upload photos of their completed projects, where they can then tag specific furniture and accessories offered by merchants Houzz partners with--letting other users easily buy any product they see and love.
The platform, which launched its online marketplace in 2014, currently offers about six million products through 15,000 merchants. And while that's a lot, the catch has always been that users can tag the products only if they're offered in Houzz's catalog.
Thanks to some advanced artificial intelligence, that's about to change. According to Tech Crunch, Houzz has developed a deep learning system capable of scanning images of rooms containing furniture and accessories, and then offering users the chance to buy similar items within its own database.
Called Visual Match, the feature has the potential to become a big revenue driver for a company that industry experts had already predicted would reach $1 billion in sales within the next five years, according to CNBC. Houzz partners with companies like Black and Decker, GE, and Keurig, offering their products and taking a 15 percent commission on all sales.
The technology could be game changing not just for Houzz but for the retail business in general. If identifying a specific product is as easy as having algorithms process a photo, it could make consumers that much more likely to make purchases: spot a jacket your Facebook friend or a celebrity is wearing, and the system could identify the product and some similar alternatives. Since the technology uses deep learning--a powerful branch of machine learning that can process data in sophisticated ways--it will presumably be far more accurate than standard image recognition and gain even more precision over time as it analyzes more photos and compares them with past results.
Facebook rolled out similar technology earlier this year with Automatic Alternative Text, which studies images and provides written descriptions for the sight impaired. Google also introduced PlaNet, which identifies where a photo was taken without using geotags--instead relying on landmarks and clues such as types of vegetation, languages on signs, architectural styles, and the side of the road cars are driving on.
And if the tech can be applied to photos, you can imagine a near future of e-commerce apps on mobile devices that can scan and identify items out in the real world, or something close to them. That's when shopping could get very smart--and retailers will have found their holy grail.