Data Science at Allstate
Allstate has a rich history of innovation when it comes to analytics, dating back more than 80 years. In 1939, it led the way in reducing rates for safe drivers. Customers embraced this more personalized approach to insurance pricing, and its popularity set the industry on a new course.
Today, Allstate continues to serve as a leader in insurance—putting models at the heart of its business to drive what Chief Data and Analytics Officer Eric Huls calls a “step-change” in capabilities for the more than 16 million households Allstate serves.
“There are opportunities through data, analytics, and technology to create more consistent, positive interactions with our customers, and shift from simply making things right when bad things happen to finding ways to prevent those adverse events in the first place,” Huls explained.
Allstate uses data and analytics solutions to support claims processing, help deliver quotes, and predict thousands of decision-making actions across products, sales, operations, marketing, and claims.
Helping 17,000 adjusters resolve claims faster; that’s just one business outcome resulting from Allstate’s model-driven approach. Allstate relies on the Domino data science platform to facilitate end-to-end model management in an agile environment, accelerating research while ensuring regulatory compliance.
For Huls, productivity of the company’s analytics organization (known as D3—Data, Discovery and Decision Science) is vital to the company’s transformation. This team, which in recent years has grown to more than 300, works closely with each of the company’s business units and functions, surfacing opportunities to apply models to business objectives that improve customers’ experiences with Allstate.
“The less time and effort we have to spend relearning or recreating results, the more time we can spend creating additional value, and the more projects we can support,” explained Huls.
However, as research and development efforts blossomed, time-consuming processes threatened to slow Allstate’s progress. Certain software tools and infrastructure resources were not readily available to data scientists, causing a delay in projects. Such delays put new research into a holding pattern.
Without historical systems of record or version controls, at times it took months to recreate existing models or obtain answers to regulatory questions. What’s more, with work scattered across different file systems and no easy way to maintain version control, team members couldn’t easily build on past research as they developed new models.
We are weaving fact-based decision making into the fabric of the organization to not just improve how we operate, but transform the industry itself.
Additionally, the team struggled when it came to testing ideas with business users and iterating quickly. It took substantial time and money to set up testing environments, limiting which ideas data scientists brought to end users.
During a short pilot, Allstate determined Domino’s data science platform addressed the needs described above. Two months into the pilot, the scope of projects running on Domino organically grew from three to 26, as data scientists migrated their work to Domino for the productivity benefits. The original three projects accumulated more than 5,500 hours of compute time, which equates to two and a half years of work.
One of the most important features is the ability to document work, maintaining a project’s artifacts and history for both research and auditing purposes.
Today, the team uses the Domino platform to reliably and securely develop, validate, deliver, and monitor new models for use cases like forecasting losses, uncovering new insights into claims, calculating customer lifetime value, and predicting potential customer churn.
Data scientists have seen benefits in three key areas:
Fast access to the environments and tools they need. With Domino, Allstate data scientists have one-click access to compute infrastructure and software tools including R, Python, and SAS. And new team members can take advantage of pre-built environments so they get started much more quickly. “By not dictating the tool set, we've found our people are more engaged in their work and more productive,” said Chief Data Scientist Rick Bischoff.
Data Science Manager Stephen Collins added, “Data science has become a multi-language environment, and if you have Domino with these pre-prepared Docker images that your team can load, it helps new hires get started much quicker.”
Improved collaboration. Allstate data scientists can seamlessly collaborate on code development, share feedback and iterate, and data science leaders have a comprehensive view of what’s in flight. “It's easy for team members to pass the same environment along, and every time they sync they can see what has changed in the project and if there are any new developments,” said Bischoff. “They can also see how their colleagues have developed their code over time, and learn from their work, which accelerates the pace of innovation substantially. At the same time, it empowers managers as they can track progress without having to disrupt their data scientists from their work.”
Institutionalizing knowledge, insights, and artifacts. Built-in reproducibility enables staff to understand and build on past knowledge. “One of the most important features is the ability to document work, maintaining a project’s artifacts and history for both research and auditing purposes,” said Bischoff.
Use Case: Faster Claim Resolution
For any insurer, managing claims is one of the most pressing areas to get right. Customers who have experienced an accident, storm damage, or other loss, are often already under a great deal of stress, and frustration can easily set in with any delay.
Allstate’s D3 team is developing models to help the company’s more than 17,000 claims adjusters gain new insight into each claim, better prioritize tasks, and tailor the process for their customers.
Testing models is crucial to ensuring end user adoption and optimizing model outputs. The team traditionally struggled with how to securely enable a test group of adjusters to access new app capabilities and provide their feedback. Now as data scientists develop new models, they can create a Shiny app in Domino using R and share it with targeted end users to gain immediate feedback from those who will ultimately use the model outputs.
“The testers in the field sign in through Domino and can play with the model to see if it would help them with their claims processing,” said Collins. “As they gave us feedback, we were able to roll that feedback into the app overnight. We got to test it really quickly and got to that learning a lot quicker.”
The Domino Effect
Better customer experiences. Allstate set out to unlock innovations that would transform insurance for consumers and widen the gap with its closest competitors. Today, data scientists can experiment more, driving more breakthrough product experiences. The claims process is faster and more seamless, increasing customer satisfaction and reducing costs. “We are weaving fact-based decision making into the fabric of the organization to not just improve how we operate, but transform the industry itself,” said Huls.
Domino encourages us to experiment more, and we can go from idea to testing much more quickly and securely.
Supercharged innovation. Instead of reinventing the wheel, employees build on past experiments and test more ideas in parallel, so innovations are delivered faster. “Domino encourages us to experiment more, and we can go from idea to testing much more quickly and securely,” said Collins.
Strengthening governance and regulatory compliance. Previously, non-technical users spent considerable time tracking down information for regulators. By putting in place a platform for creating, quality-checking, and hosting data science artifacts, staff can more quickly and easily respond to regulatory inquiries. “We can now go back to any project at any point in time and see what decision was made, and recreate that model if needed, which is huge in explaining to both management and regulators how we built a particular model—including what data went into it and what variables it identified,” said Bischoff.
Attracting new talent and accelerating onboarding. With Domino, Allstate data scientists spend very little time getting data or tools ready for experimentation and have the flexibility to use languages they’re most comfortable with—which is critical to both attracting the best talent and reducing the learning curve for new team members.
“We've had a large amount of growth over the last few years, and with Domino we can give new hires access to the tools they're using at universities or doing their own work on,” said Huls. “They’re able to get up to speed and begin adding value quickly.”