Customers are living in an era of greater convenience than perhaps ever before.

Big data analytics has been called the largest game-changing opportunity in sales and marketing since the advent of the internet, and it’s already transforming customer relationships across an array of industries. Researchers at Rakuten Institute of Technology, however, have developed a unique artificial intelligence agent that automatically identifies prospective customers with a high likelihood of engaging with a product or service.

Rakuten AIris (pronounced “iris”) uses deep learning technology to analyze over 900 e-commerce behaviors among users who bought certain products, and ranks others who have not purchased them, resulting in scoring for potential buyers. Compared to targeting methods based on suggesting products that are similar to ones in a user’s purchase history, AIris is up to five times more effective. Clients provide Rakuten with a list of their customers who use Rakuten services, but they do not need to have a store on Rakuten Ichiba to benefit from AIris.

Clients can also rest assured that their data is in good hands. Rakuten takes data protection seriously: It was the first Japanese company to implement stringent European privacy rules, ensuring a consistent, high standard of privacy protection in the Rakuten Group across the globe. Customer privacy is a top priority for Rakuten and AIris is no exception.

Using AI to find new customers

AIris launched last year and is showing great promise as a tool for generating new business. Some of the AIris developers recently sat down with Rakuten Today and gave us an inside view of how the platform works.

“The goal of AIris is to help internal and external clients find customers within the Rakuten universe,” says Tianyu Li, research scientist in the Applied Behavior Analysis Team, Intelligence Domain Group, Rakuten Institute of Technology. “It’s very easy for clients to use because the only thing they need to provide is a list of existing users. Through AIris, we can return a ranking of users, with top users having the highest potential of becoming customers.”

For example, a sports drinks maker using Rakuten as an advertising platform could benefit from using AIris to identify people who like sports goods and who might be interested in buying its beverages. Rakuten is then able to carry out a targeted marketing campaign on behalf of the client, without sharing the user data itself with external clients.

“At Rakuten, our AI efforts reach across many topics, from search results to recommendations on Rakuten websites and apps to personalization to marketing, advertising and more,” says Ashish Pandey, senior vice president, Data Science Products at Rakuten USA. “AIris is exciting for two reasons. One is it helps find new users that brands themselves didn’t know existed. The second is that rather than just being a transaction, AIris creates greater value in that Rakuten, shoppers and brands all benefit from customers gaining a more relevant shopping and advertising experience.”

 

Tianyu Li (left) from Rakuten Institute of Technology, Hiroki Mizokami (center), from Rakuten’s AIris Science Team, and Ashish Pandey (right) senior vice president of Data Science Products at Rakuten USA explain how AIris helps brands find new customers they never knew existed.

Tianyu Li (left) from Rakuten Institute of Technology, Hiroki Mizokami (center), from Rakuten’s AIris Science Team, and Ashish Pandey (right) senior vice president of Data Science Products at Rakuten USA explain how AIris helps brands find new customers they never knew existed.

How it works

How does AIris find potential shoppers? It uses a number of AI techniques including Word2vec, a type of neural network model, and topic modeling, a kind of statistical model. AIris crunches the demographic data, browsing history and purchase history from millions of users in the Rakuten ecosystem. After data processing, AIris comes up with feature sets that represent user behavior, including age,  gender and average spend on a certain class of product. If the feature sets are similar to those associated with a vendor’s existing clients, there’s a good chance that the highest-scoring users could become new customers. For more accurate identification of behaviors, the features are combined with high-level features extracted by unsupervised deep learning algorithms.

One advantage of AIris is that it makes use of data from across the Rakuten ecosystem, not just purchase histories from Rakuten Ichiba. Large search engines and social media platforms may have similar agents but they don’t have access to purchase history data, says Hiroki Mizokami, vice manager in Rakuten’s AIris Science Team. “Some competitors say they can present a similar audience, but we say potential customers because we know who purchased what,” says Mizokami.

Building on a solid foundation

Since its launch in Japan in May 2018, AIris has been used by brands in industries ranging from automakers to consumer products, drinks and healthcare companies. Client feedback has been very positive, and the AIris team is working to expand functionalities and service areas. Mizokami and his colleagues are keen to develop the service into a comprehensive platform that can automatically identify potential customers and launch marketing campaigns for them, making marketing all the more relevant and personal. Meanwhile, with an eye to future overseas expansion, AIris is being tested with more than 20 advertisers — including in USA and Europe.

“Based on the facts of what a user has purchased, we are working on inferring their life stage, whether they’re a student or married, and lifestyle, for instance are they music enthusiasts, as well as predict intent and affinity for brands,” says Pandey. “Can we predict that a customer wants to buy a car or replace its tires? Being able to correctly infer lifestyles and predict intent is only becoming possible now with better computer processing power. It’s very exciting times for AI at Rakuten.”