The insurance industry is uniquely positioned to benefit from machine learning and AI. Insurance data is typically unstructured: phone agents collect information with open-ended questions, photos document a highly variable range of incidents, and large chunks of text pass from person to person.
Machine learning and AI can structure this information and introduce automation into the claims process. Working with high-quality structured data, insurance teams can transform manual, time-consuming processes into an efficient data-driven operation.
A leading Insurtech SaaS platform selected Cake to manage its AI infrastructure. Powered by Cake, this white-labeled Insurtech product digitizes and automates all the steps of the insurance claims process.
Key takeaways
-
Cake saves a leading Insurtech vendor “at least half a million dollars purely on headcount.”
-
Cake sped up the time to launch a new AI-powered product by several months.
-
“Cake is like DevOps on steroids.”
The Insurtech team started working with Cake early, with only a few initial services deployed as Python applications.
“We were a super lean team when we started working with Cake,” stated their CTO. “We wanted to scale our work without building a large MLOps practice.”
As a starting point, the Insurtech team needed a stable and highly scalable way to deploy a machine learning classifier model and expose it via an HTTP endpoint.
Given the need for high uptime, scalability, and the sensitivity of insurance data, this service needed to be secure, stable, and able to handle variable request volume. It also needed to be in production as soon as possible. The Insurtech team approached Cake as a partner to drive the project to production quickly with confidence.
Cake enables AI/ML teams to configure the right stack to move their AI use cases into production with confidence. The Insurtech team deployed their classifier using an open source stack:
-
RayServe - Used to deploy the classifier model into production
-
XGBoost - Optimized distributed gradient boosting library
-
SKLearn - Machine learning library
-
Evidently - Used to track feature drift and prediction drift over time. This information is used to determine when to retrain models.
-
CatBoost - High-performance open source library for gradient boosting on decision trees; used for the majority of boosting within this use case.
-
LightGBM - Gradient boosting framework that uses tree-based learning algorithms
-
Pandas - Open source data analysis and manipulation framework
-
Grafana - Monitoring data for the team’s RayServe implementation, showing how much memory is used, latency, and other associated system performance metrics.
-
OpenCost - Used to examine costs per category and conveniently optimize costs as needed.
The AI landscape is crowded, complex, and changes quickly. The Cake team offered guidance on which technologies would best fit the requirements at each step.
By providing expert recommendations on common challenges, Cake enabled the Insurtech team to focus on more strategic problems. As their CTO described, “One of the effective accelerators of working with Cake – especially in the early days – was following the pattern. Let’s work on the pieces that are distinct and meaningful, and let’s not reinvent the wheel where we don't need to.”
“Cake is like DevOps on steroids.”
- Founding ML Engineer, Insurtech Company
The Insurtech team has used Cake to deploy various ML models, Kubeflow pipelines, and even as a serving mechanism for a generative AI chatbot the team is developing. As their founding ML engineer outlined, “Cake is like a data platform combined with a serving platform, all-in-one.”
Their CTO continued, “Cake is a real force multiplier for us on MLOps. Beyond that, they’re an effective strategic partner and collaborator on our new initiatives as we push into new spaces.”
“With Cake, we are conservatively saving at least half a million dollars purely on headcount.”
- CTO, Insurtech company
By providing platform capabilities and strategic guidance, Cake has enabled the Insurtech team to do more with less – and most importantly, to deploy their products faster.
Working with Cake saved the team months, according to their CTO. “We didn’t have to hire specialized MLOps engineers. That recruiting would take six weeks optimistically, but more conservatively, it could take three months to stand up that initial team and then another couple of months to stand up the actual platform.”
For example, the Insurtech team uses Ray Serve and Evidently to monitor drift. Their Evidently server reads drift statistics on a periodic basis, a process managed by a Kubeflow pipeline also deployed with Cake.
“We would otherwise have had to build a whole serving platform ourselves – that’s at least a month of work,” their founding ML engineer described, “and then a whole job scheduling system and a whole data platform – that’s another month at least. We got that all from the get-go with Cake.”
“Without Cake, we would have had to spend months of engineering to figure out how to build all this stuff ourselves.”
- Founding ML Engineer, Insurtech Company
The Insurtech AI/ML team works both in a cloud environment as well as locally. Cake enables this work pattern, allowing teams to benefit from cloud resources when needed. As their founding ML engineer explained, “Whenever I was fine-tuning the vision model, we ran it on the Cake platform. Having 64 gigabytes of memory (or more) on-demand was super convenient.”
As the Insurtech team has grown, they’ve also used Cake to better collaborate with one another. “We can all look at each other’s notebooks,” their founding ML engineer said. “It’s multi-user and infinitely better than a local Jupyter server.”
Since partnering with Cake, the Insurtech company has scaled considerably. For their insurance customers, better user experiences lead to better customer retention. Additionally, more efficiency for end users leads to cost savings for the business. Excellent incentive alignment exists to transform an inefficient manual industry into a sector that is far more automated and efficient.
Their CTO detailed the road ahead: "We’re very appreciative of the partnership with Cake. They’ve clearly been effective partners through our initial phase, and we have a ton of confidence that we’ll be able to evolve well together.”
