Get started with a free trial of Domino→

×
Industry
  • Insurance
Location
  • Headquarters: Seattle, Washington
Use Cases
  • Claims automation
Impact
  • 7 seconds to claims approval and resolution
  • 75,000 claims automated
  • 20% increase in adjuster efficiency when using model-driven claims “assistant” tool
Data Science Scale
  • Automated claims processing currently deployed at 4,000 veterinary practices with goal to reach all 25,000 veterinary clinics in the US.
Users
  • Data scientists and BI analysts
  • End users, such as claims specialists
Solution
  • Data Science Platform: Domino
  • Data Science Tools: DataRobot, Jupyter, Scikit-learn, TensorFlow
  • IT Infrastructure: Amazon Web Services (AWS), SQL database

Data Science at Trupanion

In 1999, Trupanion launched with a simple mission: to help pet owners in North America afford the best veterinary care possible for their pets. Without such coverage, pet owners often face considerable financial stress and difficult decisions when their beloved companions become injured or ill.

Trupanion set out to change this by not only providing comprehensive coverage for cats and dogs but also by applying algorithmic models to automate the entire claims process. Trupanion’s proprietary and patented software — currently used by 4,000 veterinary practices in the US — analyzes, processes and pays vet bills directly at the point of service in under seven seconds. Pet owners only pay their portion of the bill (typically 10 percent of covered services less their chosen deductible). They no longer have to submit claims or wait for approvals and reimbursement. Trupanion is the only company today in the pet medical insurance industry with this capability, which is dramatically improving the customer experience.

The system has processed more than 75,000 claims with 99 percent accuracy to date. “It’s a game-changer,” said Trupanion’s Director of Data Science David Jaw. “We’re cutting the uncertainty of whether or not you’re going to get paid. You can know the answer before you walk away from the reception desk.”

Challenge

In North America, pet owners spend more than $62 billion caring for their cats and dogs, yet less than two percent have medical insurance for their pets. (In contrast, approximately 25 percent of pet owners in the UK have medical insurance for their pets.)

There are many factors contributing to this low adoption in North America, and Jaw believes a cumbersome claims process adds to the challenge. “Pet owners pay the full cost of treatment at the time of their visit and then submit a claim to their insurer for payment,” Jaw explained. “The process can take several weeks. During that time you have no idea whether your claim will be covered and how much will be eligible,” said Jaw.

To improve the customer experience and grow adoption of medical insurance for pets, Jaw and his team began using DataRobot to prototype models that could predict whether an item on a vet invoice was covered. DataRobot was instrumental in testing the feasibility of new data and model features. However, once a model proved to be viable, the team couldn’t quickly put it into production, facing challenges in three key areas:

  • Integrating the models into the claims automation system using APIs

  • Tracking model performance to ensure the combined predictive accuracy of all models didn’t slip below 99 percent once in production

  • Gaining access to compute resources needed to test model changes

“We were falling behind schedule and would have spun for months or longer without a platform that allowed the data science team to manage model production,” said Jaw.

We can easily share our work with each other in a controlled and reproducible environment.

David Jaw
Director of Data Science

Solution

Trupanion’s data science team wanted to move fast to avoid further delays. As they evaluated possible solutions, Domino quickly rose to the top. “Domino fit our needs, and took only about two weeks to demonstrate a successful proof-of-concept,” said Jaw.

Today, the team uses Domino to support the entire model management lifecycle. “We now prototype, peer-review, and publish all of our projects using the Domino platform,” said Jaw. “We can easily share our work with each other in a controlled and reproducible environment.”

Moreover, the platform has improved collaboration between Data Science and IT. “Without the ability to provision resources ourselves and create APIs, we would have constantly been going back and forth,” added Jaw. “Instead, we can seamlessly push forward together on our shared mission.”

Use Case: Automating the Claims Process

In veterinary medicine, there aren’t standardized medical codes, and because pets can’t tell their owners or vets what hurts, medical diagnoses can often be vague, such as “not feeling well today.” As a result, to automate claims processing, Trupanion’s data science team had to create a series of 15 independent models, each designed to analyze a specific item or issue. For example, some models use simple text search to identify non-covered items like pet toys. Others use deep learning frameworks to uncover connections among words across one or more visits. Results from each of these independent models are then fed into a final decision model that correlates these predictions to decide whether to pay a claim, what deductibles to apply, and how to write a denial letter if necessary.

The team rigorously tests all models prior to putting them into production, matching thousands of predictions to actual adjuster decisions to ensure a combined predictive accuracy of 99 percent — the same standard the company sets for its claims adjusters. “We are only allowed to make one mistake out of every 100 claims, so we have two streams of models continually running: one in production making decisions and one in ‘test’ where we can continually work to improve prediction accuracy,” said Jaw. “Because of our modular approach, we can rapidly attribute any errors to one of the 15 models, and then using Domino, we can quickly spin up the 30-plus endpoints necessary to experiment on ways to correct the problem.”

We've been able to bring models to production and automate more than 75,000 claims with only a few expert data scientists.

David Jaw
Director of Data Science

The Domino Effect

Delivering answers in seconds, rather than weeks. Participating veterinary clinics that use Trupanion’s proprietary and patented software can process claims and receive payment in seconds. “Reducing the turnaround time for processing claims to seconds enables pet owners to focus on providing the best care for their pets,” said Jaw. “We believe this will pave the way for greater adoption of Trupanion and medical insurance for pets in general.”

Scaling operations. As of June 30, 2019, Trupanion insured 577,500 enrolled pets and delivered revenue growth in excess of 20 percent for 47 quarters in a row. By simplifying a laborious claims process, Trupanion’s claims team can apply their extensive medical knowledge on complex claims and support the company as it continues to scale.

Increasing data scientist productivity. With Domino and DataRobot, Trupanion is maximizing the productivity of its data science team while making data science more broadly accessible across the organization. “We’ve been able to bring models to production and automate more than 75,000 claims with only a few expert data scientists,” said Jaw.

Increasing claims adjuster efficiency. In addition its claims automation system, Trupanion has deployed a model-driven claims “assistant” tool to help guide it claims adjusters through particularly difficult and time-consuming decisions. Adjusters using this tool have realized a 20 percent increase in efficiency so far.