Ambac exec talks about operationalizing AI for insurtech

With hands in a variety of insurtech functions, Ambac thinks about how to leverage AI

April 23, 2024 5:00 PM

Robotic hand and human hand point to holographic AI icon

Ambac Financial Group is a holding company with a stable including Everspan, a hybrid fronting carrier, Cirrata, an insurance distribution platform, broker Capacity Marine, and multiple MGAs and MGUs. Ambac supplies technology to the companies it owns. Digital Insurance spoke with Randy Paez, chief information officer of Ambac, about the technology and data management demands it helps address.

What are the biggest issues for Ambac?

Randy Paez of Ambac Financial — Randy Paez, chief information officer, Ambac Financial

From a technology standpoint, one is data, always. In our role as either the owner of a fronting carrier managing programs and retaining part of the risk, or certainly as the acquirer of MGAs, pulling in data is critical for us because, at the end of the day, our value proposition to our subsidiaries includes bringing them advanced technology that they might not have access to on their own, giving them operational efficiency through back-office optimization, and then also being able to bring data intelligence to these companies.

One of our core philosophies is to produce top-quartile underwriting results. To do that, you need data, and you need to be able to provide back insights. We think that the principals of our companies have all proven themselves to be very successful in their own right. They could only be aided by having data intelligence to help with their natural ability to underwrite and price insurance.

What are the biggest risks in underwriting at the moment?

One of the greatest opportunities we have is also one of the greatest risks – AI. Every company should be using AI. If you're not – and doing it ethically and safely, and with the right controls in place -- you're really far behind. However, if you don't have the right controls in place, and you don't have the right education in place, the AI is up for interpretation. It could lead you down the wrong path, to making the wrong decision. If someone's not doing research, if they're not owning that output that AI produces, they risk following the wrong path.

What do insurers need to operationalize AI?

You need programs that actually bring AI out. Last March, companies would have said, "We're closing down AI. No one can use ChatGPT." Then, in the second half of the year, companies were more bullish, saying, "You've got to be using it." In reality, all they're doing is giving people the right to go out and play with ChatGPT on their own.

Microsoft Copilot and ChatGPT get a lot of press. Copilot is really the top, and is going to be the top, simply because it's integrated into all the products you naturally use, whether it's Word or Excel. This is where a lot of underwriting data comes from, if not your own platforms, which in a lot of cases are built on Microsoft stacks. So those plays are going to be very prevalent and very important to the underwriter making decisions.

To the extent that you are able to get people using AI materially and in ways that enter the business decisioning process and/or the business document authoring process, then you need to also educate them on how to confirm they are not encountering a hallucination or, worse, taking a valid position albeit one that counters their own or the company's philosophies or positions.

Said another way, the user has to own the output, which means following through to the source of information to the extent they can get to it. Copilot and Gemini (ex-Bard), for example, show you links that you can follow to where the supporting information was sourced from. You can then decide if you trust this source and are willing to accept the writings or guidance from the AI. This is not a welcome concept, given just how easy it is to copy/paste the output from an AI, but the minor impact to efficiency, which is still net positive if you did all the research yourself, will pay dividends in the final AI-supported artifact or decision.

How do insurers need to manage data to make the most of AI?

In insurance, we've talked about data for a long time. What will be a real game changer is the enrichment of data. There's no insurance company that does not have all its data in a lake. But it's only as good as the data you put in, as good as the data you get from your trading partners, whether they be programs or MGAs or TPAs.

We're going to see many more insurtechs that have data for you. There are going to be lots of different kinds. It won't be one clearinghouse. One company will have flood data for you, another company will have property loss data, and others crime data or weather data. You need to pull that together into one view so your underwriters understand all the different vectors that help them underwrite risk, through the AI.

Then you enrich and analyze the data. You have that AI layered in. You have your data scientists and your analysts looking at it. They're working with the AI, pouring through data. They're asking questions. And then you enrich that same dataset with that much more. The AI now is able to give you better answers. What is the propensity for hail damage in Kentucky or Kansas? We know it's pretty high, just naturally, because we as humans have seen the weather data. The AI, when you pipe weather data into it, gives you a prediction with much higher probability. Once you start mixing that together, you end up with some really, really powerful outcomes.