The reproducibility and agility offered by Domino impacts Allstate’s model management process in four key ways:
Accelerating research. Instead of re-inventing the wheel or rediscovering results, employees build on past experiments and test more ideas in parallel, so they deliver new models faster.
“Domino encourages us to experiment more,” said Amy Gu, data scientist at Allstate. “We measure ourselves based on the value we offer the organization,” noted Huls. “The more projects we can get through, the less time and effort we have to spend relearning or recreating results or doing the same thing twice. That is time we can spend creating additional value.”
Enabling open source. Data scientists have the agility to use their tool of choice, which is critical to staying on the cutting edge and can reduce costs. Rick Bischoff, director of Data Science at Allstate noted, “We had a project that used to run on a commercial solution and one of our employees was able to replicate it in R on Domino.
Ensuring regulatory compliance. “Insurance is a heavily regulated environment. We can go back to any project at any point in time, see what decision was made and recreate that model if needed,” said Bischoff. “Domino brings a software engineering discipline to data science… It’s the data science platform you should be investing all your time in.”
Accelerating employee onboarding. Data scientists spend very little time getting data or tools ready for experimentation in Domino, and they have the flexibility to use languages they’re most comfortable with.
Stephen Collins, a data science manager at Allstate explained, “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.”
Allstate set out to unlock innovations that would widen the gap between Allstate and its closest competitors.
“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 preventing those in the first place,” explained Huls.
On that path, the data science team noticed opportunities to drive innovation faster by preserving, sharing, and reusing more of their research. If they could improve reproducibility and collaboration, not only would they accelerate their research, but they would also reduce regulatory compliance risk.
Without historical systems of record or version controls, employees were constantly hypothesizing work of former employees. It took months to recreate models built in years past. Achieving reproducibility would make it easier and faster to build on past work.
Collaboration & Agility
Communication delays between data scientists and other stakeholders could be alleviated with a centralized platform facilitating collaboration. The team also sought an environment that would keep up with emerging technologies and trends, such as open source, to foster fast innovation. And Allstate wanted to empower data scientists to, for example, change their computing environment without needing to go through IT.
Non-technical users wasted time tracking down timely responses to regulators’ questions from a lack of institutional knowledge needed to access and understand current predictive models. Putting systems in place, improved reproducibility and collaboration and the ability to address regulator requests.
During a short pilot, Allstate quickly determined Domino 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 forthe productivity benefits. The original three projects accumulated more than 5,500 hours of compute time, which equates to 2.5 years of work.
Today, Domino serves as Allstate’s enterprise-wide data science workbench and collaboration hub, facilitating reproducible research by creating, quality-checking, and hosting data science artifacts. The platform brings together users from across the company, allowing data scientists to build models but also to deploy them in production and expose them directly to business stakeholders, such as adjusters in the field.
Domino manages the predictive models empowering business use cases across Allstate including loss models, claim insights, and marketing models to calculate customer lifetime value and churn predictions.
There were four main reasons why Allstate selected Domino as their data science platform:
Tool agnostic support allows the company to hire, ramp, and keep the best people who will innovate faster using the technologies that work best for them.
Built-in reproducibility solves a challenge Allstate has been struggling to address for years.
Intuitive interface is easy to use by varied employees across the organization, and encourages experimentation.
Quality people provide thorough, fast response. “They’ve bent over backwards for us. They’re probably the best vendor that I’ve worked with,” summarized Bischoff.