Ask the Experts: Kenny Holms, Chief Decision Scientist and Robert Hooley, Head of Machine Learning


As Chief Decision Scientist, Kenny Holms has led Accelerant’s work with data science, AI/ML, and advanced analytics for three years. Alongside him works Robert Hooley, Accelerant’s Head of Machine Learning.

We spoke to them about Accelerant’s AI and Machine Learning function to share insights on the initiatives the team is developing and the role of AI/ML in Accelerant’s ecosystem.

How and why did you join Accelerant?

Kenny Holms: I’ve worked with Accelerant’s executive team before – they’re tremendously sharp, and tremendously good. They had a clear, strong vision for leveraging AI and ML right from the beginning, and I was excited by the variety and scope of work: developing cutting-edge models across distribution, operations, risk selection, and portfolio management. When they said they wanted to launch an AI practice, I was in.

Robert Hooley: Kenny inspired me to join his team for the same reasons. Accelerant believes in innovation, and so do I. The company is young, and unburdened by legacy processes. I appreciated the freedom to develop processes from the ground up and add value through ML.

How does machine learning fit into the larger Accelerant ecosystem?

We embed machine learning into each business function to reduce siloing between departments and increase visibility within them. We employ both custom-developed solutions for specific insurance tasks — such as risk selection and claims management — and third-party tools for processes like data management and ingestion. Our tools act as a force multiplier, enabling our expert colleagues to automate routine tasks so that they can concentrate on more challenging priorities.

How does Accelerant’s machine learning team impact Member success?

We develop some tools that Members can see and others that operate behind the scenes.

Behind the scenes, our models identify patterns that support Members cross-functionally. They inform discussions between Members and our analytics and distribution teams, help manage the claims life cycle, and evaluate portfolios to ensure we’re securing the most efficient reinsurance deals for Members.

Simultaneously, Accelerant works directly with Members to inform their risk selection models, both in portfolio management and segmentation. The team tailors machine learning models to help Members stay within their underwriting pricing guidelines, especially when diving into new lines of business or increasing profitability. This is how Accelerant developed its Risk Index, a model that pulls data to provide a digestible risk score. We are currently in the process of rolling out access to Members via the Risk Exchange platform.

Increasingly, we are also working to supply Members with targeted risk signals and curated third-party data to support decision-making. As the team produces new models to help Members maneuver certain claim types, they then generalize the tools to share with other Members. These are deployed quickly and automatically through the Risk Exchange, rather than outdated distribution methods.

Can you share an example illustrating your collaboration with Members?

Managing claims is complex —  with so many spinning plates, it can be tough to keep track. Over the last year, we partnered with the claims team to build out an LLM  that helps them identify claim recoveries.

An example of a claim recovery is when a car crash faces split liability, i.e., neither party is solely responsible. Recovery occurs when one insurer claims back some of the payment, and the bill is split between multiple insurance companies.

Our LLM absorbs claims documents and suggests whether or not the claim is likely to have a recovery. The claims team then uses this insight to triage and litigate with other insurers. While it is not a replacement for claims assessors and professionals, it augments their efforts through an initial ranking that helps them manage workflows. With the tool, we’ve seen more recoveries come through, significantly reducing costs.

How would you say Accelerant’s approach to machine learning and data-driven decisions differs from the rest of the industry?

The P&C market has seen varied success in using data, technology, and ML/AI — largely due to the complex nature of insurance. Tools and models effective for tech giants like Amazon or Google often fall short in addressing the complex challenges of the insurance sector. Meanwhile, executive teams across the country lack realistic expectations: ML needs to be managed like any other business function —  it’s not magic and you need a plan!

We’re different in a few ways. From the outset, our leadership team has adopted a principled approach to leveraging ML, focusing not on experimental projects but on creating tangible tools and real value. It is not a side project; it is a core part of our mission and values.

We also believe that you can’t create value without strong technical AI/ML skills in combination with deep insurance expertise. Our team has both —  we’re insurance people who just happen to be experts in AI/ML. We speak the language, we understand our colleagues’ and Members’ challenges, and we’re razor-focused on helping them.

What industry trends will have a significant impact on your work and Accelerant in 2024?

2023 was the year that AI captured the public imagination thanks to tools like ChatGPT, Midjourney, and Copilot. We’ve been leveraging the technology underpinning these tools for years now —  but the increased focus from their newfound fame provides opportunities and challenges as we look to 2024.

For example, AI is making unstructured data —  like notes, a phone call, or a satellite image —  easier to use. We’re focusing on ensuring we have a strong return on investment from AI tools as we navigate whether to build out our own AI systems or rent them. The key is separating hype from reality and staying ahead of rapidly changing and geographically diverse regulatory challenges.

What will the use of machine learning in specialty insurance look like 10 years from now?

There will be more of it, and it will be more embedded throughout the value chain rather than locked in one place. The challenge is getting these models in front of people. Many industry processes don’t currently experience much ML use, but there will soon be better data ingestion and processing. We customize broad technological stacks, and then integrate them into operations.

We’re also seeing clear trends in the adoption of automated underwriting, i.e., automatically transacting business based on models with rules that learn on an ongoing basis, replacing handwritten code. We’ll see more adoption of these tools for simpler or personal lines — less so for niche specialty markets.

Could you share a moment of achievement from this past year that was particularly memorable for you and your team?

First, the adoption and success of our claims lifecycle LLM is having a big impact in helping our claims team better manage claims on behalf of our Members. It’s incredible to see the tangible value of our LLM models.

We’ve also made massive progress and excitement for our Risk Index, which uses first-party data (e.g., premium, loss, exposure), scalable third-party data of any kind, and Accelerant ML. We’re tremendously excited to roll these out over the coming year to help our Members continue to grow profitably.

What goals do you have for next year?

Our first is to operationalize our Risk Index signals to help Members select better risks. In other words, we are aiming to get more risk indexes in front of as many Members as possible. Our second goal is to further develop LLM technologies that enhance our ROI and provide pragmatic value. The third is to build upon our Portfolio Monitoring Machine, the web of ML that forecasts opportunities and challenges for Members. We can take its results to Members to discuss, for example, areas that are prime for growth or changing rates. We are aiming to automate insight as much as possible with a model that scans our portfolio 24/7 to flag items and help Members stay on top of them.

What is the most surprising thing about machine learning in insurance that people don’t know?

Building a machine learning process is simple. It’s properly validating it and aligning it with the problem at hand that’s difficult. This requires extensive expertise and experience, both in the business domain and in the math and technology behind it.

What would you like Accelerant employees to understand about what you do?

AI and ML can simplify many processes.  Consider the processes you do every day —  if they’re repetitive, but the task is too complex to write if-then statements, they can likely be solved with ML. Anytime you need to simplify immense amounts of data into bite-sized insights, ML can also help.

As experts in AI and ML, we act as a consultant to Members. For instance, if a Member is exploring an external vendor or data set, we are happy to advise on the best path forward.