Recommendation engines are the norm on websites. The problem is that the basic, statistically driven, engines don’t understand an artistic view, especially when referring to images such as on home décor sites. While recommendation engines are great for more statistical analysis and relationships, something different is needed to improve success on sites that require analysis are more artistic concepts. Machine learning (ML), luckily, is a broad spectrum of solutions, and it is starting to be applied to design issues that can improve business, both on the internet and in retail.
Three years ago, I covered how a company was using visual ML and self-tagging to help movers provide more accurate estimates to their customers. However, that was only for identification of objects. The solution for moving firms had no need to be concerned with home design, but only with identify the object and the size for the purpose of logistics.
Home shopping has boomed in the last few years. Recommendation engines for direct sell, cross-sell and up-sell, are critical for helping customers quickly find objects to buy. That’s not a problem for items where the relationship is simple, such as plungers and drain unclogging liquid. However, there are many areas where design comes into play. For instance, people’s living rooms purchase decisions are more complex, with colors, styles and more having an impact on buying decisions.
The typical recommendation engine uses statistical analysis and clustering to suggest most customers who like A will like B. Art is more subtle. To add to the complexity, with many vendors for each product such as chair, couch, or lamp, and each site potentially using multiple vendors, trying to find enough data for accurate clustering is difficult.
One company looking at providing a solution the challenge is Renovai. “Online purchasing for home design is the perfect market for incorporating design knowledge within machine learning and AI based solutions”, said Avner Priel, Renovai’s Co-Founder & CTO. “It is not practical to infer a statistically meaningful solution for each shopper, so our system is based on industry best practices, rules, and trends, as if our algorithm attended design school”.
That is accomplished by design experts working closely with artificial intelligence (AI) experts to capture design knowledge, build a graph database, and have the system leverage that information for recommendation on sites.
On home design sites, one thing different than many other sites, is the creation of scenes. They can show an individual couch, but a more powerful way to display the couch is in a living room scene with other objects. The Renovai system uses visual ML to identify the objects and analyze the relationships between them.
When a visitor sees a couch, adds it to a shopping cart, and then searches on lamps because the one in the scene isn’t right for the person, the system uses the knowledge in the graphical database to provide a list of options ordered by the style relationship to the couch.
As more sites are added, and as vendor catalogs are updated, the system identifies outliers from its existing knowledge. As with most ML systems, outliers are the keys requiring the most human intervention. Whether it’s a new class of object that has been identified or a question about the category to which an object belongs, the system will make an initial decision and then flag the change. The design teams can then modify the graph and retrain the system. In addition, that knowledge is used to automatically tag new items with far more information than the basics of color and size. The captured knowledge from designers has more detail in order to build the system’s model of the artistic relations between objects. The potential buyer doesn’t need to know those concepts, only that the items look good together.
Another interesting component to the solution is support for “click and mortar.” There is an in-store component, provided through a browser, where sales teams in furniture stores can use the same functionality to help customers. Home design firms, especially those with both an online and a physical presence, can improve performance by addressing market needs through either source.
I’ve always viewed programming as a craft, not engineering or science, but a blend of those and art. Home design is also a craft. It is interesting to see one craft apply its trade to assist another. The blend of human and machine knowledge to address the sales challenge of home design is an interesting application of AI and other tools, a blend that can help both the businesses and the end customers.