Amazon is pushing into real-time recommendations and advertising, but a recently granted patent describes how the company uses "matrix factorization," which fills in missing values in algorithms using averages. This method becomes more exact in time as it learns.
The patent -- initially filed in May 2014, updated in June and granted last week -- shows how real-time recommendation may be made based on the user-specific update and item modeling.
To create effective item recommendations in real-time, the patent's inventors say the system must track and store "large amounts of past behavior of customers." While the platform analyzes customer data to make the recommendations, it also must have the smarts to fill in the blanks through predictions based on what others have purchased.
Combined with the data from Amazon's network of sites, which include Whole Foods, Twitch and Zappos, real-time recommendations would give the marketplace an edge. It's something the company has yet to explore.
Still, recommendations seem to attract shoppers. Analyst firm Raymond James, with help from Survey Monkey, conducted an online survey to gauge consumers' use of websites and applications. About 49% of the 1,000 consumers participating in the survey said they use Amazon for ecommerce search, compared with 31% who use search engines.
Some 40% of respondents said they would be likely to spend the most on Amazon, while 20% indicated big-box stores including Walmart, 10% indicated department stores and locally owned stores; 7% indicated outlet or off-price stores; and 7% indicated an online-only retailer excluding Amazon, according to the survey.
Ironically, Amazon has offered advertising longer than most people realize, but it's still a relatively young business for the company. It introduced display ads, 300 x 250 units, on its site about 10 years ago based on the business model's mantra to work backward from customers to company as a way to facilitate discovery of products and services.
Today there are several search products. Sponsored Products and Headline Search Ads are keyword targeted bought on a cost-per-click basis and feature one product each. Product Display Ads are more focused on brands and highlight three products rather than just one.
In short order, Amazon's jobs listing shows an increase in advertising-focused openings, many of which are international, in China, from account managers to Amazon Marketing Services experts. A post in early October went up looking for someone to support the new self-service advertising platform for brands selling on Amazon. Through cost-per-click search and display advertising products, AMS aims to help brands reach targeted audiences on Amazon.co.jp, according to the post.
Internally, advertising has been seen as a way to keep prices low for consumers, but that could change as the ad business model grows and becomes a major profit center. For example, the Fire tablet or Kindle that comes with ads cost about $25 less when consumers choose to buy the devices embedded with ads.
Today, ad teams span across the U.S. -- in Seattle and Los Angeles with recent expansions in New York and Chicago, among other locations -- as well as internationally in London, Munich, Paris, Toronto, Tokyo, Barcelona, and Milan.
There are agency teams of specialists to support brands along with advertising holding companies such as WPP in search, display, video, along with tech-focused services like programmatic and header bidding.