XL Catlin Analytics Strategy: Quality Over Quantity
Kimberly Holmes, XL Catlin's SVP of strategic analytics, turns to an old joke to describe her pragmatic approach to big data: When you're chased by a bear, you don't have to outrun the bear—you just have to outrun your friends. That, Holmes says, is XL Catlin's approach to the ever-escalating technology possibilities of analytics: staying just ahead of the competition.
That pragmatism leads her to stress the “strategic” in her title above the “analytics.” She looks to solve problems in a way that will be useful to the organization. “You always have to start with that endgame in mind,” she says. “When we're choosing projects, we're already thinking about willingness of the businesses to change and adopt what we're doing.”
XL Catlin, the result of the recent merger of major property and casualty insurance rivals XL Group and Catlin Group, houses terabytes of information. But while possessing a big data store and using big data technologies such as SAS and R , it aims to find insight by analyzing the variety of its data, not its sheer volume. “A lot of people don't do analytics unless the data set is large,” she says. “I think they would be surprised at the insights you can find when you have 'medium data.'”
And that belief keeps the company focused not only on its technology execution, but on making sure key stakeholders are involved in the drive to find real solutions to real problems.
“I don't think it's really about the data or the tools you use,” she says. “Whether you do it in R or Python or SAS or whatever doesn't really matter. It comes down to adoption.”
Adoption as Endgame
Holmes was hired five years ago by XL Group to establish XL enterprise-wide strategic analytics center of excellence. It was a first in the XL Group's space, she says; other companies were still piloting proof-of-concept projects to win the executive buy-in Holmes already had. She grew her team, and with XL's January 2015 acquisition of Catlin Group, which had its own data initiatives, her team numbered about 20, spread among the United States, India and Europe. They're typically working on two or three projects at time, with Holmes leading the day-to-day, as well as the overall strategy.
Often the analytics work confirms and refines the way insurers do business. Sometimes, though, the models defy conventional wisdom.
“Those to me are the holy grails, where we now have an insight that's the opposite of what most people think to be true, and we can take advantage of that to create competitive advantage.” She adds that major contrarian insights are rare. “I've had it happen three times.”
Her team starts by identifying a question whose answer will deliver value. Her team asks about hunches, unanswered questions, and long-held assumptions that can be tested with data. The team works closely with the relevant business units, aiming to improve adoption by investing decision-makers in the data model. “The more it's their model in development,” she says, “the more it'll be their model after it's implemented.”
Her first deliverable is a data model, an algorithm, created in SAS and turned into a tool for end-users by developers using the .NET framework. But her work isn't done until she's trained underwriters and other stakeholders in the use of the tool.
Making the Model
Once a project's goals are identified—better efficiency, increased profitability, etc.—choosing the data is straightforward. “For all of our projects, we look for new data, from both internal and external sources,” she says. Just citing examples of government sources in the United States, she mentions the U.S. Labor Bureau, highway transportation agencies and the Food and Drug administration. “And of course, we as an insurer have our own data collected to draw from.”
Getting internally and externally sourced data ready for modeling is the bulk of the work, she explains. “Coming up with a cohesive data set probably takes four times longer than creating the model,” she says. “I would say we spend 45 percent of our time on the data, 10 percent on the model, and 45 percent on change management.”
The data is housed in SAS and SQL databases , linked through ODBC connections. The team writes code in the SAS programming language to run the data analysis, relying on a code library. SQL and R are also used, to a lesser extent; SQL to extract data from source systems, and R, a language and environment for statistical computing and graphics for exploratory data analysis.
In some cases, stakeholders can also explore data themselves, parallel to the Strategic Analytics Team's work on the data model. The complete data set is complete can be put into QlikView, a business intelligence application that lets users search and visualize data, creating reports, visualizations and dashboards. This allows underwriters to take a more analytical approach to decisions even before the full data model is complete.
As Holmes' team works on the model, it communicates preliminary findings with stakeholders through reports, which the team presents live, taking feedback that can help shape the model. Her team is developing reports in Shiny to show the relationship between dependent variables (what XL is trying to predict) and independent variables (possible predictors.) An open-source R package, Shiny is a web framework for building interactive data applications. The team presents the reports in live meetings, delivered as .PDFs.
Her team's direct work with the data ends in an algorithm. “The final deliverable is a mathematical formula,” she says.
From Formula to Working Tool
Once the algorithm is delivered, the project forks. One fork goes to developers in the IT department, who write code, using .NET, to turn the model into a working tool, automating the flow of data and creating a desktop interface for reports that can be updated daily. Stakeholders access the tool via their desktops, clicking an icon that takes them to reports that have been updated overnight.
While that tool is being developed, the project's second fork sees Holmes' team working with the tool's eventual end users.
“We work with the businesses so they can figure out how they're going to use the algorithm to improve business performance,” she says. “That's all done in scenarios and with Excel spreadsheets. You make some assumptions about how much business you're going to write, and what your actions will be—you don't need anything more sophisticated to do that.”
Then the Strategic Analytics team does training, and develops report templates that will allow the business to monitor how well the tool is being used. Some reports are in QlikView, some are in Excel sheets coded to pull in data from various systems. “All the reports we develop are close to realtime, refreshed nightly, available to everybody in the business,” she says.
A recent risk segmentation project, she says, took 16 months to complete and targeted a 4 percent reduction of a loss ratio—but is currently showing a 6 percent improvement. Holmes says risk segmentation models are her team's most common project, aimed at addressing the underwriter's decision process in assessing risks. This lets them write each individual piece of business more accurately, and it lets XL Catlin adjust its portfolio based on a better understanding of the most attractive risks to insure.
“Those small shifts in that profile drive enormous business benefit,” she says.
Running faster in the future
The bear joke talks about how you win one specific race, but business is a long haul—a sprint that never ends. Holmes talks about how XL Catlin plans to continue its goal of outpacing competitors for the long term. She says the team will continue as a center of excellence, serving all the company's business units. “As a specialty carrier, we have more than two dozen individual specialty businesses,” she notes. “Just from a practical standpoint, financially and finding the people, you couldn't replicate that within the businesses.”
She says her team's reach has been broad. She estimates that her team's projects have directly touched about 20 percent of XL Catlin's businesses, but says that these projects have driven other business units to launch effective data-driven efforts on their own.
Looking forward, Holmes says XL Catlin's strategic analytics team's pragmatic pursuit of competitive advantage will lead to investigations of machine learning and unstructured data. Her main focus is to keep pushing the envelope to do new things, to continue an overall cultural transformation at XL Catlin. That transformation will include a more analytical culture, and the tools, skills and data to empower underwriters and actuaries.
Being an analytical company, she says, isn't about having an advanced analytics team. “It's about the whole company becoming more analytical,” she says. “It's a shift from saying, 'I think' something to 'I know' something.”
(About the author:Brian McDonough is a freelance writer who covers technology).
(This article appears courtesy of our sister publication, Insurance Networking News).