Software Helps Websites Predict Users' Tastes
New software watches users to figure out what products they might buy.
E-commerce websites have only a brief window of opportunity to win a customer’s attention and money. Many sites use personalization to appeal to particular customers, but this normally means making recommendations based on a person’s purchase history. A startup called Baynote, based in San Jose, California, has developed software that automatically reorganizes a website on the basis of much more information, including how users behave after they arrive.
“The best way to understand people is not to ask them or make them tell us what they’re like, but to watch them,” says Scott Brave, Baynote’s founder and chief technology officer.
Many websites offer purchase suggestions based on previous buys or what other, similar users have bought. Amazon, for example, is famous for its recommendation boxes, such as the one that notes “Customers Who Bought This Item Also Bought …” Amazon also tailors recommendations by taking into account what a user has looked at during a visit to a website. Brave says that Baynote’s technology goes further.
Companies including Expedia, AT&T, and the clothing company Anthropologie are already using Baynote’s software, called the Adaptive Web Suite. Anthropologie says that the software has increased customers’ average order value by 35 percent.
Baynote’s software gets installed on the machines that run a company’s website. It tracks purchases but also watches users’ behavior—where a user scrolls to on a page, how much time she spends on each page, and when she clicks. The software also looks at the search terms a person used to find a page on the site—by checking the referral data—and any search queries that the user enters on the site.
The software changes a page in real time, while the user is browsing. It returns search results that are refined using information gathered about the user and optimized to take into account the products that have appealed to similar users. The latest version includes new features such as the ability to adapt a landing page of a website in response to the search terms used to find that page.
The vision, Brave says, is to make the online experience adaptive, allowing a website to change immediately to suit the situation presented by each visitor. For example, AT&T’s site suddenly started getting a lot of searches for “Insight.” Baynote’s software was operating, and it was able to observe that a lot of customers were looking for a new model of phone that had just been released. The website was adapted accordingly in a matter of minutes.
Glen Urban, a professor of marketing at MIT’s Sloan School of Management, says that the type of personalization being done by many startups today could go further. Approaches based on behavioral rules have helped make advertising and marketing more effective, Urban says, but websites could also adapt by changing how that information is presented. For example, some users might prefer to see a lot of information in charts, while others might prefer a simplified comparison of products. His group researches ways to measure a user’s personality from online behavior. It also investigates ways to evaluate website adaptations to make sure they’re actually effective.
Keep up with the latest in machine learning at EmTech MIT.
Discover where tech, business, and culture converge.
September 11-14, 2018
MIT Media Lab