• Cerebral Valley
  • Posts
  • Sixfold - an AI solution revolutionizing insurance 🔑

Sixfold - an AI solution revolutionizing insurance 🔑

Plus: CTO Brian Moseley on expanding into generative AI and agentic systems...

October 08, 2024

CV Deep Dive

Today, we’re talking with Brian Moseley, CTO and Co-Founder of Sixfold.

Sixfold is an AI solution designed to revolutionize insurance underwriting by streamlining the risk assessment process. With Brian’s extensive experience in both high-tech startups and enterprise software—spanning nearly 30 years—he and his co-founders have built a product that leverages generative AI to help underwriters make more confident decisions, faster. Sixfold eliminates much of the manual work in underwriting, while providing transparency and traceability to ensure trust in its AI-driven recommendations.

Since its launch, Sixfold has expanded from commercial insurance to life and disability lines, working with companies to enhance underwriting efficiency. By automating repetitive tasks and highlighting key risk signals, Sixfold empowers underwriters to focus on the human side of their jobs, like relationship management and decision-making.

In this conversation, Brian discusses the evolution of Sixfold, the technical challenges of scaling a B2B SaaS platform, and the future of dynamic AI-powered underwriting.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Brian 💬

Brian - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to join Sixfold? 

Hey there! My name is Brian and I’m the CTO of Sixfold. I’ve been working in the high-tech startup scene, mostly in San Francisco but also some in New York, for almost 30 years now. My first professional experience as a software engineer was in the mid-nineties when my fraternity brother founded what we’d now call a social media startup out of our frat house. Eventually I dropped out of college, moving to San Francisco to dive into Web 1.0. I spent the next 15 years in the trenches at various B2B and consumer startups, with a notable detour at the Open Source Applications Foundation.

Eventually, in 2013, I co-founded a company in San Francisco called Hoodline. It was a hyperlocal journalism play, pretty popular at the time, but it will come as no surprise that we didn’t figure out a business model for local news. I ultimately made my way to American Express, my only real big company experience. I was the head of developer experience there for five years, looking after the quality of life for 10,000 engineers worldwide.

I was happy at Amex until my old friend Alex Schmelkin, who’s the CEO and founder of Sixfold, pitched me on this idea. He and Jane, our other co-founder, had previously co-founded Unqork, a no-code/low-code platform selling into financial services and insurance. They knew how to sell enterprise software to insurance companies. With generative AI building momentum and their expertise in this vertical, we saw the opportunity to create something new. That’s how we got here.

Give us a top level overview of Sixfold - how would you describe the startup to those who are maybe less familiar with you? 

Insurance underwriting, like any kind of underwriting, is the process of assessing risk. There's a lot of manual work involved, from gathering all the data to understanding it in the context of your business's risk appetite to deciding whether to take a risk on a particular policy application. What we do is help eliminate that manual work, and beyond that, we simulate the reasoning an underwriter goes through when sifting through piles of data to find the risk signals that help them understand the nature of the risk.

On a typical underwriter's day, they might have 30 submissions coming into their inbox, but only the capacity to handle five. We help them focus on the right five so they can ignore the noise and spend their time on the risks that are most meaningful to their business. We use generative AI to emulate human reasoning and drastically reduce the time and effort it takes for an underwriter to process all that information and get to a decision, with a level of accuracy comparable to that of an experienced underwriter.

Talk to us about the journey of Sixfold’s user base - who are your users today, and who’s finding the most value in what you’re building with Sixfold? 

I’d say we’re seeing value for underwriters across the spectrum, from junior to senior levels of experience. Junior underwriters typically spend a lot of their time doing manual tasks, while senior underwriters might focus more on reviewing work or business development. By using Sixfold, they can create more capacity in their day to focus on the human aspects of their job, like relationship management or pursuing new business, while deferring the manual work and risk assessment to our AI.

In terms of segments, we started in the commercial insurance space, working with property and casualty, and expanding into specialty lines like cyber. We’ve since moved into life and disability insurance, which are very different from the others. With commercial insurance, you’re underwriting a company and assessing the risks around it, whereas with life and disability, you’re underwriting an individual, looking at their lifestyle, and health, and making decisions based on that.

How does Sixfold use generative AI to assess risk across different customer segments, especially in a way that builds trust with your customers? 

Our process starts with understanding the risk appetite of the customer we're working with, whether it's the company or a specific line of business. We need to know what types of risks they’re comfortable taking and what they want to avoid. For example—and this is a bit of a crude example—a carrier might be fine underwriting farms, but they may want to avoid cannabis farms. We help by identifying those distinctions and then either disqualify a case or adjust the score based on the risk signals we pick up in the data we gather. 

We do this through semantic analysis and question answering. From the data submitted in an insurance application, we extract specific facts, deduplicate, consolidate, and classify them into a taxonomy and, where appropriate,  build a chronology. This is particularly important in the life and disability space, where over time, we track diagnoses, medications, lifestyle factors, and how they interrelate. This allows us to tell a more comprehensive story about the person, which is key for effective risk analysis. There’s a lot of summarization and fact extraction involved.

What do you think sets Sixfold apart from other players in the space? 

I’d say a big part of our value proposition is not just regurgitating a bunch of data scraped from the web or third-party sources. Instead, it’s about deeply understanding that information in the context of the company’s specific risk appetite. We focus on highlighting relevant risk signals for the underwriter rather than overwhelming them with large amounts of summarized data that they still have to sift through and analyze.

We use AI to streamline the process and give them clear answers. However, the underwriter remains in the loop—it's ultimately their decision to make. We’re not going to get it right 100% of the time, especially with generative AI and potentially flawed data (garbage in, garbage out). The underwriter reviews the recommendations, and one of the things that sets us apart is our transparency. We clearly explain how we arrive at our conclusions and link back to the original sources, allowing the underwriter to verify or correct the information as needed. This feedback loop helps improve our models over time.

Using AI to assess risk does come with its own set of considerations. How do you establish trust in AI that has been in the news for miscalculating and misassessing? 

It’s about giving underwriters visibility into the steps we're taking. Our AI isn’t a black box; it doesn’t just spit out a decision that the underwriter has to follow. They can build trust in our system by digging into the details. For every fact we find, we show the source. If the fact came from a PDF provided by the applicant, we’ll link directly to the page or paragraph where we found that information. If there are conflicting facts, we’ll link to both and let the underwriter decide which is correct.

We also provide underwriters with the ability to give us feedback—whether we got a fact right or wrong, whether a summary was accurate, and whether the risk score matched what they would have come up with on their own. Did they quote the policy or not? All of this feedback comes back to us to help refine and improve the model.

That transparency, and the ability to extract all of that data—a trace log of everything the system did for compliance, archiving, or further analysis—means that underwriters can trust our recommendations because we make it easy for them to see how we arrived at our conclusions.

What has been the biggest technical challenge that you’ve had to overcome as CTO, to get Sixfold into the position it’s in today? 

At the start, it was about understanding the toolset and learning, just like everyone else. As the model providers released more capable models, we were iterating on our prompts and refining the processes that tie all the AI components together to get things done. Early challenges were about building experience and intuition around how to use these models effectively. 

Nowadays, it's as much a distributed systems problem as it is an AI problem because we're building a B2B SaaS at scale that needs to be reliable, performant, and accurate by multiple definitions. The system has to provide consistent service, 24/7, and what works in a notebook setting can break in countless ways when you're running constant traffic through it—especially when that traffic is spiky, like when an underwriter runs a batch of 200 cases at once while the system is tuned for 10 concurrent cases.

About 80% of the time, things work as expected, but the other 20% reveals all kinds of edge cases. With the complex underwriting workflows we handle, each unit of work needs to be resilient. We've had to break them down into small, independent units that can be retried when something goes wrong—whether it's an LLM taking too long to respond or returning unexpected errors, or a gateway between us and a model provider going down. Then there are other issues, like chunking a document into pieces that are too large for an embedding model or vector database to handle.

All these edge cases create new challenges in making the system self-healing, so underwriters always get a result, no matter what breaks in the backend. With non-deterministic systems in play, there’s a different mindset around error management—it's more about how much error we can tolerate, rather than eliminating every single error. It's really about fusing AI techniques, running them at scale, and ensuring the system stays reliable and performant.

How do you plan on Sixfold progressing over the next 6-12 months? Anything specific on your roadmap that new or existing customers should be excited for? 

I'd say number one is continuing to expand the capacity of the team to keep delivering on our product roadmap while creating more space to innovate. One of the ways we’re going to stay ahead of competitors is by creating new kinds of experiences or meeting underwriters in the tool sets where they already live, instead of forcing them into our app. That’s going to involve things like figuring out how to present the Sixfold experience within someone’s Outlook mailbox, their underwriting workbench, or CRM, where they spend all their time. We don’t want to give them a 16th tab to open—we want to be exactly where they need us.

A lot of that will come from UX experimentation and building out our AI team so we can have a robust research process. That way, we’re staying on top of the unique needs of our customers, and that research can feed into production, allowing us to build these features and host them at scale. So it’s about expanding the team, continuing research and experimentation at both the AI and UX levels.

We’re also keeping a close eye on how the community is building agentic tools. Our underwriting workflows are complex and encode a deep knowledge of underwriting for different businesses and insurance products. We’re really interested in the idea that someday these won’t be static workflows, but more dynamic—where agents could optimize workflows for a specific case or line of business, or even find new ways to get to a risk decision that we hadn’t anticipated.

That’s the industry trend we’re trying to stay on top of, but it’s a challenge given the level of growth we’re experiencing. We’re piloting and onboarding more customers, and that puts pressure on the engineering team to deliver for customers, meet the product vision, drive innovation, and maintain operational efficiency. We don’t want to have to hire 50 customer success managers to onboard new clients. Managing all these priorities—across innovation, execution, and growth—is what we’re focused on right now.

Lastly, tell us a little bit about the team and culture at Sixfold. How big is the company now, and what do you look for in prospective team members that are joining?

We’re always hiring. Right now, we’re particularly focused on building out our AI engineering and product engineering teams. We need a lot of velocity and creativity in both areas. Like any startup, we look for people who are curious, adaptable, and able to cope with change because it happens every day. We need people who can work autonomously and motivate themselves by solving customer or business problems, not just ticking tasks off a backlog. We also look for people with leadership qualities, even if they don’t have leadership experience yet. We want to help build that experience so they can take on bigger and more impactful projects.

Another key trait we look for is versatility. Many of our engineers came on board  knowing one programming language in our stack—maybe Python, Ruby, or JavaScript—but now most can work across the whole stack. They can prompt an LLM, put a button on a webpage, or add a new step to our workflow orchestration system. This flexibility lets us put small teams together to focus on customer or business problems quickly.

Culturally, we focus on being plainspoken and direct, but in a kind way. We avoid business politeness and prefer clarity and support. Our team is distributed, with senior managers and contributors not based in New York City, but we value face-to-face human connection. Remote-friendly, yet believing in the power of in-person collaboration, there's a strong nucleus being built in the New York office, working seamlessly with talented specialists across the U.S. and even in Latin America.

We try to bring the company together when we can, and in October we’re doing an offsite in Austin for a few days to build those personal relationships and create moments where we can inspire each other in person.

Conclusion

To stay up to date on the latest with Sixfold, learn more about them here.

Read our past few Deep Dives below:

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email (newsletter@cerebralvalley.ai) or DM us on Twitter or LinkedIn.