Razorsight Launches New Predictive Analytics Solutions for Telecoms Running on MapR Platform

Razorsight’s new Hadoop-based architecture on MapR has helped achieve horizontal scalability on commodity hardware, reduced storage and computing costs. This new technology stack allows Razorsight to continuously innovate and deliver value to its telecommunications customers by offering predictive insights from the cloud.

The Business

Razorsight’s cloud-based predictive analytics software specifically designed and developed for the telecommunications industry delivers insights to help communications service providers (CSPs) and media companies proactively improve customer experiences, reduce costs and increase margins. Current clients include AT&T, T-Mobile, Comcast, Telus, CenturyLink, Windstream, and Virgin Mobile.

Background and Challenge

Razorsight got their start a decade ago offering financial assurance analytics solutions to telecom companies to help them with financial compliance and controls, understand leverage points and optimize costs. Razorsight’s solution was based on traditional business intelligence and data warehouse technologies. Most of the structured data came from their customers’ operational systems, and Razorsight applied heuristic business rules, logic and workflow automation to validate and identify cost optimization opportunities.

“As we grew as a company and big data evolved, there was a lot more data sets available,” explains Razorsight CTO Suren Nathan. “Today’s telecom data has higher volumes, frequency and complex structures. There are new types of devices generating data for the Internet of Things, mobile phones using broadband for apps, and VoIP. Razorsight had to utilize this newly available data to generate predictive insights using data science and predictive analytics.”

To accomplish this, Razorsight had to evolve its technology stack to achieve scalability at the appropriate costs to deliver opportunities to communications service providers.

MapR Solution

Razorsight decided to move to an Apache Hadoop-based infrastructure to take advantage of the emerging trends in big data architecture and parallel computing. “We selected MapR for several reasons,” he says. “First, having the flexibility of the full Spark stack as part of the Hadoop distribution was very important. Second, MapR provided production-class Hadoop with enterprise support. And third, the NFS gateway was critical for us to integrate ingestion and data flow pipelines with HDFS for easy, high-speed access.”

Razorsight used MapR to build a central data lake as a primary data store for both online and archive data. Since the launch of this new stack in late 3Q2014, the production cluster has received, processed and analyzed more than 40 terabytes of data. Their customers send data in all shapes and formats from multiple sources. Razorsight leverages the MapR NFS gateway to move these data sets in and out of the cluster seamlessly, making it extremely easy and intuitive to integrate Hadoop into the overall data flow. Razorsight then uses Spark as an inmemory processing engine to enrich and transform the source data to prepare the analytical records for advanced modeling. Spark provides the required high performance to accomplish this function. Additionally they use ElasticSearch for search-based analysis. Meanwhile, the end-users and business analysts, continue to use existing business intelligence and visualization tools on their downstream data warehouse.

As Razorsight built their technology platform, new efficiencies and enhancements were incorporated into the Predictive Analytics solutions. “We saw that we could leverage big data and add to the data sets we already gather to be able to offer additional solutions to our customers. Since all data (raw, enriched, transformed and aggregated) is present in the data lake, it is easier for us to insert additional use cases to deliver newer insights for our customers quickly. An example of this are the operational use cases where we predict the repeat callers into the call center or predict network node failures before they occur,” he said.


Some of the MapR differentiating features include NFS support, multi-tenancy, remote mirroring, and Spark stack support. The combination of these features is leading to several business benefits.

Expanded customer solution opportunities

  • The new platform has enabled Razorsight to expand into new solution areas for their telecom clients.

    Their sales and marketing product is designed to improve the customer experience, reduce churn and identify the next best action. The marketing team at Virgin Mobile Latin America has deployed Razorsight’s sales and marketing solution in multiple countries to support its expansion there. Razorsight’s predictive analytics will help them tailor targeted marketing campaigns based on a particular customer’s propensity to churn.
  • Reliable enterprise-grade platform

  • “Our technology stack is one of our major differentiators. Our customers trust us with their data, and our ability to generate accurate and reliable predictive insights in the fastest possible timeframe,” says Nathan. “MapR doesn’t rush to market with every little thing the open source community puts out so I know when they deliver a release, I can be guaranteed they have done their due diligence and it’s ready for our cloud environment.”
  • Improved performance

  • With their previous architecture, Razorsight ran into multiple bottlenecks because data ingestion, processing, analytics, querying and visualization were all competing with each other for processing power. Now with the new MapR platform, they can completely separate these, making a huge impact on performance and scalability.
  • Cost savings

  • The MapR architecture also accounts for a huge cost savings. “The total cost of storage and processing for a traditional enterprise EDW platform is about $15,000-20,000 per terabyte. With the Hadoop ecosystem, this has dropped to about $2,500-3,000 per terabyte,” says Nathan.