A homegrown Hadoop-based business analytics solution along with proprietary BI tools is helping HomeShop 18 understand and serve its customers better. By Harshal Kallyanpur
A smooth customer experience is at the top of the priority list for any retail chain. True to this dictum, e-commerce portal, HomeShop18, is focused on ensuring that its customer gets the lowest price and the best possible shopping experience.
According to Vish Magapu, Chief Technology Officer, HomeShop18.com, the two primary events that the company wants to respond to as fast as possible are when the customer is waiting for a product that he has ordered or during a customer service issue. It wants to optimize product delivery and ensure that the customer is informed, as early as possible, about delays or issues at the company’s end.
To achieve these objectives, the company relies heavily on business analytics. This is primarily divided into three areas with the first being Web analytics where the e-commerce portal monitors customer activity online in an attempt to understand likes and dislikes, products that are being sought and those that are trending. This enables it to customize its catalog and pricing and engage with the customer better.
The next piece is campaign management and promotion wherein the company can design targeted campaigns for a particular type or set of customers. Magapu said, “We want to segment our customer database, such that we can design specific offers and promotions based on their inclination towards a particular genre of books.”
The final piece is internal, which is the fulfillment, supply chain and operational analytics piece wherein HomeShop18 looks at how these processes are performing and how certain metrics are being tracked and if process efficiencies can be optimized.
As a customer clicks on something, the e-commerce portal keeps capturing all the clicks. These clicks form a chain of events, which help the company determine buying patterns or customer preferences such that they can, for instance, target the customer with other products that he is likely to buy.
“There are millions of events and these events, in turn, generate data that requires powerful data crunching capabilities. We adopted a Hadoop-based analytics approach to process and analyze this data,” informed Magapu.
The organization, with its team of engineers, has developed an in-house business analytics solution running on a Hadoop architecture. Magapu attributes the development of a homegrown solution to proprietary products being expensive while the relatively inexpensive ones are quite primitive in terms of the features that they offer.
Each day, the portal sees up to 8 to 10 GB of data being generated. It maps all of the events from this data and filters out that which is not required. Around 10 to 15% of this data is used for dynamic merchandizing while the remainder is used for aggregation in order to study browsing and purchase behavior.
According to Magapu, given the volume of data, it is difficult to predict the traffic and the associated volume of data after six to eight months. Hadoop provides a linear scaling option wherein the company can add a few more nodes as its data processing requirements go up.
He added, “We do have other internal BI dashboarding and reporting tools, which help us track sales on a daily basis, the daily revenue for each category of products, daily orders and daily SLAs. To do so, we have deployed a traditional data warehouse-based BI solution in an in-memory environment. We have built a sophisticated data warehouse with a feed from the operational data store, with cubes and dashboards running on top of it.”
He explained that, in a business scenario, if a manager looking at a sales dashboard wanted to go into a category or a sub-category, or build a dimensional view of the data, either that cube could be built on-the-fly, or have a precomputed database. Both approaches had performance constraints such that you had to wait for the cube to be built. In comparison to this, in-memory offers a faster way of generating cubes on the fly.
HomeShop 18’s underlying data warehouse has been built in-house. The operational data store is built on MySQL but the data warehouse is built on SQL Server with integration between the two. The presentation and visualization layer is based on an commercial BI tool from Yellowfin.
“We started with our BI implementation a year ago but all of our systems have gone live and are working well today. A lot of work across departments used to be done manually using Excel sheets. Post the implementation of the BI tool, this efforts has reduced by 70 to 80%,” said Magapu.
Prior to the adoption of the in-memory approach it would take the company anywhere between two to three days to generate financial reports for auditors. After the implementation, it takes only a few minutes to generate the reports.
The company built fulfillment dashboards based on the BI tools. SLAs are specified within the dashboards for functions such as order received, order verified and order fulfilled. If there is any anomaly in this process it can immediately contact the customer about the delay in processing the order. This has improved the customer experience greatly.
