Kobo is cost-effectively growing its business and continuously using insights gained from its MapR solution to improve its offerings and website for customers and gain a competitive advantage in the market.
Toronto-based Kobo sells world-class E Ink eReaders, Google-certified Android tablets, and free eReading applications for the most popular smartphones and tablets. It also offers one of the largest eBookstores with more than four million titles to customers in 190 countries, across 71 languages. Kobo is owned by the Tokyo-based eCommerce company Rakuten
A competitor to Amazon’s Kindle, Kobo started small but has been growing rapidly and is now a global player in the eReading space. In about 1.5 years, Kobo’s data store expanded from the low-digit terabytes to 150 terabytes.
“The process of looking for books and reading generates a huge amount of data. As our customer base grew, we wanted to keep all the data about our customers so we could generate insights and serve them better. Expanding on our technology became increasingly important to us,” explains Dr. Inmar Givoni, Head of Research and Data Science, Big Data, Kobo.
“We were using SQL Server technology to do analytics, and knew we would soon require a speedier solution. We needed to make a change,” she says Kobo was looking for a platform that would serve as the central hub for data. The hub needed to integrate with their recommender system and provide an easy interface for BI analysts and data.
"We did a competitive analysis between Cloudera and MapR. We chose the MapR Distribution for Apache Hadoop because of its Network File System (NFS) capabilities, speed, support, and no single points of failure. It was the best Hadoop option."
|-- Dr. Inmar Givoni, Head of Research and Data Science,Big Data, Kobo|
Kobo loads structured and unstructured data into MapR from multiple sources including e-commerce websites, mobile phone apps that capture usage metrics, and raw text and metadata from books. The recommender system accesses MapR as network attached storage via NFS and performs clickstream analysis to track customer activity and create reports. Due to the POSIX-complaint NFS access of MapR, not a single line of code had to be changed in the recommender system. MapR also helps power Kobo’s website and eReader device search engines. Kobo can ingest unstructured data from a variety of sources and process it with the velocity they need to keep up with web/eReader traffic. They use Apache Solr with the MapR Distribution as the search engine for their website. On the business analytics front, Kobo leverages Apache Hive tables and makes them available via Tableau for data analytics and visualization. Data scientists also run MapReduce programs while applying machine-learning algorithms. All of these users are exporting and importing data using NFS.
"When you compare Kobo to Kindle, we’re a much smaller company. Through the power of MapR, we can act much bigger and be a more powerful entity than we could be otherwise."
|-- Dr. Inmar Givoni, Head of Research and Data Science, Big Data, Kobo|
Kobo is seeing multiple benefits to their business from their MapR solution.
Operational efficiency and performance NFS is the key feature that drove Kobo to select MapR. “It is the most commonly used feature on a dayto-day basis by the team. It makes the data directly accessible. Without NFS, it’s extremely inefficient to get data in and out of a Hadoop cluster,” says Inmar Givoni.
The shift from their SQL Server solution to a MapR solution has greatly increased performance as well. Kobo’s ability to process data has significantly improved—what used to take days can now be done in hours, and in some cases, minutes.
More revenue opportunities The ability to store virtually unlimited amounts of data opened new opportunities for their business. “Previously, we would only be able to store a limited amount of data. Since we no longer had restrictions, this alleviated the performance and capacity problems with SQL. We are now able to store more, capturing robust history and enhanced customer insight, and this allows us to generate more value”, she says.
With more storage and processing available, Kobo was also able to add value to its library of more than four million electronic books. “We analyze the text, themes and content. This enables us to handle the ‘cold start’ problem of new books so we can recommend these new books to customers who have liked similar books,” she says.
Changing how our business operates Kobo’s MapR solution is also yielding more insights and intelligence to help them continuously expand their business. “It has changed the way we run our business. We used to measure regular KPIs such as how many customers visited the website, or how many books or devices we sold,” she says. “We have moved from that into a testing model. We can measure the impact of changes we’ve made to individual features, and this is only possible with more data.”
Having access to more data makes it easier to understand what marketing tactics are most effective. “We do a lot of advertising—online, display, and through Facebook ads. Customers come in to our website in many different ways,” she explains, “and our customers are cross channel—they use our website, apps, and devices.”
“We need to bring in data from multiple sources. We can now keep a lot more data about where our customers came from, and what they did on our site. We can learn through A/B testing what ads are working for our high value customers and optimize our target ad spend”,she says.
A cost-effective solution like MapR enables Kobo to compete on a much larger scale than they could otherwise. “We don’t need that much hardware because it’s much more efficient. Kobo’s cluster manages several hundred terabytes of data. It’s very cost effective,”she says.
And it allows them to compete with major players like Amazon’s Kindle. “When you compare Kobo to Kindle, we’re a much smaller company. Through the power of MapR, we can act much bigger and be a more powerful entity than we could be otherwise,”says Inmar Givoni.