Soon, you will own a nice apartment or home. And you’ll want to furnish it, decorate it, and make it your own. What digital transformations might help you personalize your future abode?
If you recall from Angie’s List, the company decided to add a free tier to entice more user memberships, hoping this would entice service providers on the platform to continue paying for advertising . As a company that relies heavily on advertising revenues for nearly 80% of its revenues however, Angie’s List could use alternative ways of creating and capturing value.
One of its smaller and more recent competitors, Houzz, is seeking to diversify its revenue streams and be more relevant to millennials. Houzz, an online home improvement marketplace, is free for users and creates value for service providers (think architects, designers) by providing them a place to upload photos of their work to casual users who may be thinking about redecorating, redesigning or furnishing their apartments or homes. As a platform, Houzz captures value by charging for premium listings.
Fig 1: Houzz’s website, a Pinterest-like display of service providers’ photos of home interiors and exteriors. But what if you want some of the furniture or other items in the photo?
Unlike Angie’s List, though, Houzz also earns additional revenue streams by featuring advertisements from major merchants (like IKEA and Black&Decker) and through e-commerce commissions on products users see in the photos (Fig 2). In terms of operations, service providers and Houzz staff must explicitly and laboriously tag each product in the photo, and if the product is already in Houzz’s database (thanks to its partnerships with certain merchants), users can buy the item through a link to the product’s website.
Fig 2: After Houzz’s deep learning algorithm recognizes the products in the photo, Houzz’s Visual Match tool can also show you how products will appear in your own home . Neat!
For Houzz to achieve its customer promise of providing customers the ability to buy any item in its photos, it needs to overcome two limitations. The first is that many service providers do not know what products are in the photos they are taking – they are selling the interior design, or the porch, not necessarily the furniture or appliances. Having the service manually fill in the products in their photos when they are uploading is a nuisance. The second is that although Houzz’s marketplace contains 11 million photos, only 6 million products can be purchased through its database (that’s fewer than one product per photo) . Just looking at the photo below, 6 products appear in just one picture, so there seems to be a bottleneck in terms of tagging products in these photos.
To automate the task of marking items in a photo for sale, Houzz has begun to use software to scan service provides’ photos for products that look like those already in Houzz’s database. The software is based on an advanced method of data analysis known as deep learning. With deep learning, the goal is to get software to recognize things like facial features the same way that our human brain does – by first detecting simple patterns and then using them as building blocks to detect more complex patterns . Houzz’s engineers train their deep learning algorithm on pictures of products already in the database (e.g. tables, sofas, chairs) so that when the algorithm encounters a new photo, it can accurately identify what the product in the photo is and whether it exists in the database.
While automatically recognizing products in photos is an important new feature, an additional step that Houzz should be thinking about is how to get more items into its database, so that its deep learning algorithm can find virtually any product in any photo of a room, patio, or yard. By coincidence, just the other day I had a discussion with a fellow HBS CODE club member about a web crawling tool he is currently developing. Given a list of keywords (say, of various furniture) to be queried, and a way to look up the prices and websites for those keywords, this tool can autonomously query (or web scrape, as they call it) websites and automatically export into another file the prices and link information for thousands of products at a time. Such a tool, while still in testing phase, can conceivably be used to greatly expand Houzz’s database. With Ericsson predicting worldwide smartphone subscriptions projected to double to 6 billion by 2020, smartphone camera resolution increasing by a megapixel a year, and trillions of photos uploaded to the cloud annually, Houzz’s investments in deep learning should be a useful tool for future homeowners like you and me . (782 words)
 Dolan, Robert J., and Ayelet Israeli. “Angie’s List: Ratings Pioneer Turns 20.” Harvard Business School Case 517-016, September 2016.
 Lardinois, Frederic Houzz now uses deep learning to help you find and buy products in its photo https://techcrunch.com/2016/09/14/houzz-now-uses-deep-learning-to-help-you-find-and-buy-products-in-its-photos/ TechCrunch.com, Sept 14, 2016.
 http://www.houzz.com/ideabooks/61877912/list/inside-houzz-introducing-view-in-my-room Houzz.com, February 17, 2016.
 https://www.youtube.com/watch?v=CEv_0r5huTY Deep Learning, Simplified. Published Dec 11, 2015 by DeepLearning.TV