Can a giant be agile? Consumer credit reporting company Experian, a $4.7 billion enterprise organization, faced that challenge. Customers large and small were pursuing digital strategies and looking for real time insights. How could Experian better serve their needs?
About 8 years ago the company helped pioneer the use of Hadoop to store massive quantities of data for consumer credit scoring. Today Experian maintains something it calls the Ascend platform, a real-time data lake of the entire credit worthy population in the US -- the most recent 15 years of credit-related data on pretty much the entire US population with the exception of children and people who have just moved to the US or who have no credit history. Ascend lets customers do real time queries on the data rather than batch processing. For instance, a business customer could look at what the impact would be if they lowered the credit score threshold from 700 to 680 for their end customers. Based on the results they get, they can do another query right away. They don't have to wait to submit it for another round of batch processing.
This customer-facing service is just one aspect of the technology infrastructure at Experian, overseen by CIO Barry Libenson, whose career has included tours of duty at a startup he ran for 10 years, the database giant Oracle back when it was a $19 million company, and companies including consumer products producer Land O'Lakes and grocery chain Safeway. He knows a thing or two about running the technology at big organizations.
When he joined Experian in 2015, the first big project he set out to tackle was technology standardization.
"The strategy has been very much a decentralized approach," Libenson told InformationWeek in an interview. The infrastructure had been decentralized in a sort of federated approach. "That worked effectively for a while. When you are a $2 billion to $3 billion company, decentralized is good," he said. "But when you are $5 billion to $6 billion, decentralized is a little inefficient."
There was quite a bit of infrastructure to work on -- a dozen data centers around the world including two in Dallas, two outside of London and two in Brazil, plus primary and secondary failovers for those geographies. Experian also had six colocation facilities. Nearly all of its infrastructure has been made up of on-premises implementations, with some small penetration into AWS and some evaluation of Microsoft Azure. In 3 years, a lot has changed, Libenson said.
"We've done a massive consolidation and have driven rigorous standards into the data center in terms of hardware platforms," he said.
Here's what Experian has set as the standards: Dell/EMC for traditional storage (although Libenson said that it may deviate from that standard for secondary or tertiary commodity storage), Cisco for networking equipment, and HPE blades in the data centers. On the dev side, Libenson said the company has moved to OpenShift as its cloud-based development architecture. Experian has also standardized on Cloudera's Hadoop and data platform across the enterprise. The team also uses analytics engines including H20.ai, R, and Spark, and has based Ascend's visual user interface on Tableau. Libenson said that Experian's data storage is between 2 and 10 petabytes.
Did standardizing save a lot of money?
"Believe it or not, the goal of driving those standards wasn't so much financial as driving innovation and the speed of change," Libenson told InformationWeek, noting that Experian is saving about 10% to 15% from an infrastructure standpoint by reducing the number of suppliers.
"What is much more interesting is that as we moved to a more Agile development methodology we've been able to get a much higher degree of productivity out of our developers by having everybody using microservices, standardize on the same framework, use the same set of tools," he said. "We've gone from a traditional waterfall release cycle of 4 times a year with one major release every 12 months to a cadence that is much more in line with Agile where we are doing anywhere from 10 to 15 releases a month. That's something we'd never be able to do without having a common set of tools and common set of standards across the corporation."
Experian is enjoying some significant benefits in terms of reuse of the components that developers have built.
It's a lot different from how the company would have done it a few years ago. For instance, something like the Ascend platform might have been branded the same way around the world, but the developers in one country might be using different tools and create a different code base than the developers in another country. The standardization makes the code much more portable and has enabled Experian's developers to be able to build products more quickly, Libenson said. It has cut development time by 50% in different parts of the world.
"It's really more about being able to build products more quickly," he said.
Of course, every standardization effort is a process, and Libenson acknowledges that the enterprise is only 60% to 70% there. He said that they are 12 months away from having everyone aligned and on the same platform. Once you get there, maybe it's already time to look at updating the infrastructure and standards.
"People will say it never ends," he acknowledged. "But it's about putting the rigor and discipline in place to get everybody to use the same set of tools." That said, Libenson said they are always looking for opportunities to improve throughput. They are always evaluating tools.
Among the technologies Libenson said Experian is spending more time with is machine learning.
"We are using machine learning in both our application layers and in our data centers for analysis," he said. Libenson believes the industry is just beginning to ramp up with this technology, and it will start a wave that will continue for the next 12-plus years. He said it's no longer a buzzword or hype. It's reality.
"I get really frustrated now when I walk into a hotel room and I can't talk to Alexa," he said. "People's expectations are really starting to change."