MUJI Finds Retail Mojo with Hybrid Analytics

(Nestor Rizhniak

Japanese retailer MUJI is enjoying the fruits of a hosted big data solution that delivers a more accurate view of its customers and their interests. The solution, which is split between Treasure Data and Amazon Redshift, is credited with driving customers from electronic channels into retail outlets, where margins are typically higher.

MUJI is renowned for its “no brands” mantra and minimalist design elements for a variety of products, including clothing, kitchenware, electronics, furniture, food, and even a MUJI car it developed with Nissan. The 35-year-old company doesn’t spend a lot on traditional advertising, instead relying on its simple shopping experience to drive word-of-mouth referrals.

While it avoids the glitz and glamor in favor of a low-key aesthetic, the multi-billion-dollar company does maintain demand-generation programs behind the scenes. To that end, it recently launched the MUJI passport mobile app to reward repeat customers with discounts and points.

One of the goals of MUJI passport is to get customers to visit one of the company’s 585 stores around the world. MUJI is in expansion mode at the moment, having just opened a new store in Santa Monica, California, this past week, with many more new store openings in the works in Europe and Asia.

The big data-related question for MUJI was this: How can it make the best use of the new clickstream data generated by the mobile app, as well as behavioral data from users registered to the MUJI ecommerce website? The company needed a way to analyze this new source of semi-structured data alongside the traditional data generated by transactional systems.

MUJI was a happy user of Amazon Redshift, the online data warehouse that’s powered by a powerful column-oriented database from Actian, as well as Omniture, the online marketing analytics software from Adobe. While these solutions are valuable in their own right, MUJI needed something that could handle massive amounts of semi-structured clickstream and behavioral data.

This is the type of big data problem that would get many organizations thinking about Hadoop. But in MUJI’s case, it was not interested in Hadoop, which can be time-consuming to implement and requires specialized skills. Instead, MUJI turned to Treasure Data, a hosted big data firm that specializes in helping customers make sense of large amounts of semi-structured data, including clickstream, sensor, and log data.

Treasure Data’s hosted offering starts with the agents that capture data from 250 sources and stream it into Treasure’s online repository. Next up is the powerful MapReduce-based batch analytic jobs that provide bulk transformation to fast-moving data–sometimes billions of rows a day. Lastly, Treasure mates a speedy column-oriented database called Plasma to the MapReduce output that enables users to query their data via standard SQL, or output the data into a data warehouse.

“We deliver very similar capabilities to Hadoop,” says Hannah Smalltree, director of marketing with Treasure Data. “We’re a schema-on-read database. That’s similar to Hadoop, in that you can collect everything–just shove everything in there and then do your processing and sort out the actual data you want afterwards. That’s different from many of the cloud data warehouse, which tend to be more reliant on data models and more upfront definitions.”

MUJI liked how Treasure Data would fit into its existing analytics workflow, particularly how it could work with Redshift. The initial implementation took several weeks, and today MUJI is using Treasure Data to do the initial ingest and processing on clickstream data for 4.3 million registered Web users and 1.4 million MUJI passport mobile app users.

“We’re really good at taking in large volumes quickly and processing it quickly,” Smalltree tells Datanami. “Our focus is the velocity and scale, and providing that brute force processing of enormous data volumes. MUJI saw the difference in technology and the difference in cost. From a total cost of ownership perspective, for large data values, where you don’t need advanced analytics or integration, it makes more sense to use a cloud-based service to do the raw processing.”

With Treasure Data providing the first-draft processing on the clickstream data and feeding more structured aggregates into Redshift, the company has a clearer picture of the linkage in-store purchases and online and mobile behavior. Now the company is getting better signals from analytic tools when to give coupons and points to its loyalty app users.

Currently, MUJI is using Treasure Data to collect and analyze log and clickstream data only for users in Japan. But the timely processing of data and delivery into the Redshift analytics platform is paying dividends for the company. MUJI was hesitant to provide Datanami with specific figures, but the company credits Treasure Data with helping to increase in-store visits in a matter of weeks.

Takashi Okutani, general manager of the retailer’s Web business division, indicates his satisfaction with the solution. “Treasure Data was the most cost-effective solution that allowed us to deploy our program quickly and leverage the resources and systems we already had, while capturing and processing new types of big data that were critical to our new programs,” he said in a statement.

Related Items:

Why Hadoop Won’t Replace Your Data Warehouse

Finding Big Data Treasure in the Cloud

Data Scientists–Who Needs Them Anyway?