How Ping Established ML-Based Leadership in Commercial Property Insurance

Ping, a data intelligence platform, has used ML capabilities to power state-of-the-art insurance underwriting since 2021. Ping applies advanced machine learning to cleanse data and calculate accurate premiums on property insurance quotes. With ML at its core, Ping uses Cake as its MLOps platform to build sophisticated products with a small team and stay at the cutting edge of machine learning.

Bringing structure to highly variable data 

Ping uses machine learning primarily to transform unstructured data into structured data. As a service for insurance companies, Ping’s models typically combine several qualitative attributes about buildings as inputs — for example, occupancy, construction materials, and location.

This information usually arrives in email bodies, PDFs, and other unstructured and highly variable formats. Structured spreadsheets and forms also have variability from year to year (e.g., shifting locations of a specific checkbox or poor data entry), and even individual data points can have inconsistent formatting that can complicate the modeling process (e.g., non-standard address patterns).

“We started this project building a purely heuristic approach,” shared Scott Stafford, Chief Enterprise Architect at Ping. “Well-designed heuristics can be effective in many scenarios, but the current crop of ML tools are an absolutely essential component to solve the long tail of problematic inputs.”

Off-the-shelf software “didn’t meet our needs”

Ping needed to scale its training data collection process but did not have the necessary ML infrastructure internally—“we just had an S3 drive and a dream,” as Stafford described it.

Off-the-shelf commercial ML tools either did not meet Ping’s requirements, were too inflexible, or were too expensive to consider. Due to PII concerns, Ping could not send data to LLM vendors or other managed data extraction tools. High source data variability led to regular quality concerns with standard tools. Ping had also considered using a collection of cloud-hosted SaaS tools; however, as Machine Learning Engineering Lead Bill Granfield described, “we'd end up investing significantly in a range of hosted services.”

Building an open source ML stack  

With a lean team and a need to build its own services, Ping selected Cake to manage its ML infrastructure. Working together, Cake helped select the appropriate stack of components from the open source ML ecosystem and offered a simple unified approach to ML application development.

As Stafford outlined the challenge, “We knew what we needed at a high level, but we also knew we didn't want to build out a team of 10 to do all this work. Cake came along, and we’ve been really happy with the whole engagement so far.”

Composed of AI/ML infrastructure experts, the Cake team served as a resource for Ping during the initial implementation. Cake begins engagements with consultative recommendations on which open source systems and tools might be most applicable to a particular challenge.

As Stafford described, “The expertise that Cake provided was extremely beneficial. We described our situation and what we were struggling with, and the Cake team had a comprehensive knowledge of the entire open source space in order to make tailored recommendations."

Accelerating production ML with Cake

“We've got a list of ML desires a mile long - and with Cake, those moved instead of standing still.” 

- Scott Stafford, Chief Enterprise Architect at Ping

Among other projects, the Ping team has used Cake to build a PDF data extraction pipeline, and core parts of its algorithm use annotated data created with a Cake-managed version of an open source data labeling solution. Ping uses these image annotation capabilities to annotate the different versions of PDFs and build an algorithmic OpenCV-style extraction library. Ping’s system can now automatically parse PDFs and successfully extract the needed data.

“With Cake, we're collecting this data several times faster than we were before,” Granfield explained, “in a way that makes it easy to analyze what we've got, understand the provenance of annotations, and measure the performance of different annotators. It’s enabled us to release a new model much faster than we would have otherwise.”

Central management of MLOps 

Ping unlocked additional efficiency by consolidating tools and services into one Kubernetes environment. The ‘before’ state was a “mishmash” of vendor-managed cloud services and SageMaker models that lacked central management.

Cake streamlined the environment for the Ping team. Bill Granfield outlined the benefits of this new environment, stating, “We know everything is managed with GitOps, and there is one single MLOps repo with the config for our entire environment. Not only do we have version control history for all the changes to our environment, but it’s much faster to make changes over time.”

Doing more with less 

“Our partnership with Cake has been a clear strategic choice – we're achieving the impact of two to three technical hires with the equivalent investment of half an FTE.

- Scott Stafford, Chief Enterprise Architect at Ping

For Ping, a significant benefit of working with Cake has been resource savings. The current market for hiring MLOps talent is highly competitive, and it is challenging to find people who understand machine learning theory, can implement it, and are creative enough to help tackle new problems. If not for Cake, the Ping team estimates they would have likely hired at least two more full-time senior level engineers for its current challenges.  

“Cake empowers us to achieve more with our existing resources,” Stafford explained. “While expanding the team would be ideal, finding, onboarding, and integrating skilled hires can be both time-consuming and costly. With Cake, our current team can focus on high-impact work, as they’re no longer tasked with building and maintaining infrastructure—effectively doubling our operational efficiency.”

Ping’s engineers have been able to focus on higher level ML and business problems now that the underlying infrastructure is being handled reliably by Cake.

Deploying open source ML with enterprise-grade security 

The open source approach initially presented concerns for Ping due to the sensitive nature of its customer data. Most open source tools lack security and privacy features such as role-based access controls and other features only available in cost-prohibitive enterprise versions.

Cake allowed the Ping team to deploy an open source stack safely. As Stafford described, “Security is a top priority for Ping, and as we developed this system, Cake’s enhancements to the basic security measures found in open-source tools were crucial. Their work added an essential layer of protection for client data, providing the level of security any client-focused company would demand.”

Staying modern with Cake

In contrast with the slow-moving insurance industry, cutting-edge AI/ML continues to evolve with breathtaking speed. Ping currently offers a best-in-class product for the insurance ecosystem by leveraging the latest ML technologies. Maintaining its leadership position requires matching the pace of innovation in the AI/ML ecosystem.

Cake offers Ping an easy route to maintaining a modern ML infrastructure stack. Cake continuously integrates new popular open source technologies (and new versions of existing technologies) for customers. Cake is a strategic partner for Ping due to its ability to support Ping’s AI-focused competitive differentiation.

Scott Stafford summarized the long-term benefit of the partnership: “Staying at the forefront of ML advancements is essential for us, even as a young company, to remain competitive and agile. Having Cake as a key partner on this journey provides invaluable confidence that we’re equipped to evolve alongside these rapid changes.”