AM Best panel: AI usage increasing at DUAEs but not a ‘magic wand’

By Chris Munro

April 24 - (The Insurer) - Artificial intelligence has an important role to play in helping delegated underwriting authority enterprises (DUAEs) differentiate themselves, but it is not a panacea and must be used to support existing technical expertise within a business.

That was one of the views shared during a recent webinar hosted by AM Best that explored some of the key issues at play within the DUAE sector.

During the panel, AM Best senior director Sridhar Manyem said the MGA model has evolved from one that was historically based on years of relationship building between parties to one that is focused more on the superior technology that DUAEs can utilise.

Like elsewhere in the (re)insurance industry, the DUAE sector is looking at ways it can harness AI to support its operations.

FIT FOR PURPOSE

Delos Insurance CEO Kevin Stein said AI “is being used a lot more frequently” in the market.

However, he noted that “any AI model is only as good as its fit for the problem you're trying to solve”.

“You can't take an algorithm that identifies with something as a cat or a dog online, and turn that into a wildfire model, as an example. It just simply won't work.”

As Stein explained, a DUAE needs expertise to understand the technology and what the best algorithm would be for the platform’s specific needs.

By combining industry knowledge with technology, Stein said the DUAE will have “even more niche market specialisation”.

“You'll end up with a last mile solution for a single peril that will drive extra alpha and return over other models that are made for different perils,” said Stein.

“That technical expertise specific to data analytics is really what's allowing delegated underwriters to be much more effective than the rest of the industry at those perils.”

AI is a support tool though, and not something that can be relied upon to build a product.

“Just using AI doesn't mean folks without area expertise can move into a sector and create a solution,” said Stein.

The executive referenced the wildfire market in which Delos specialises as a case in point.

The wildfire sector is one of several industry segments that is “response variable starved” because only a few specific events have occurred that underwriters can correlate and base their risk assessments on.

Stein said there are about seven total wind-driven fires in California's history, and that limited number of events is not enough for an AI black box – a system in which users enter data inputs and then see the outputs, but do not view the internal workings – to create a model.

“(If) you stick a black box model on these seven fires, the response will tell you every single future fire will look exactly like the past ones,” said Stein.

“The problem is every single fire was different than the ones before,” he stated.

“So AI is enabling, and allows folks that understand the peril to take it further, (but) AI by itself isn't a sufficient amount of technology to be able to solve the problem.”

AI A TOOL

Eventual Weather CEO Mohit Chawla echoed those comments, stating that AI is something that should be used by DUAEs to support their operations rather than relied upon.

“The one trend that we have seen with our own company and our cohort of companies is essentially the top-notch underwriters which have extreme knowledge and specialisation in their niche are getting better using AI because they are able to offload a lot of data-cleaning tasks, which would otherwise take a lot of their time.

“But an underwriter with not much specialisation is actually getting worse because they are completely relying (on) and treating AI as a magic wand that's gonna come and solve all the loss ratio problems,” he said.

Chawla said technology “is existential for us” and “extremely critical” for any DUAE.

“By leveraging these emerging technologies like AI, (we) gain deeper insights from all of these diverse data sets that sometimes do not align with each other, and better correlate the risk matrix and the underwriting portfolios.

“This builds the strategic edge, or the extra alpha … that leads to better cost efficiency, lower combined ratios, which turns into better relationships with the risk partners.”

EVALUATION THE KEY AI CONSIDERATION

Chawla highlighted evaluation as a key challenge for the market regarding AI models.

There are two ways you can underwrite any risk, said Chawla. The first is using a deterministic, algorithmic approach, and the second a machine learning-based approach.

“The latest version of machine learning, or AI model approach … is still, I believe, in the early nascent stages.

“We are, of course, seeing some benefits on how you correlate data from different data sources, but the challenge is still reliability and trust on the models that are based on AI.”

Another major challenge that needs to be solved is how companies can create a feedback loop to test AI’s output alongside that of a deterministic algorithm, Chawla said.

Stein added that simply having more data is not helpful to DUAEs.

“What you need is data specific to a peril,” he said.

Chawla said there are further “exciting developments” on the use of AI in the weather sector.

“People are developing large physics models, which will be the ultimate test of AI – can you use AI to predict accurately how the weather is going to be 14 days from now?”

More broadly, Stein said the DUAE market is looking to harness insights provided by the greater availability of data, but he added that simply having more information will not necessarily lead to better underwriting.

“You have to match that (data) with folks that have deep understanding of the peril to be able to understand how to take that data and turn it into a risk model,” he noted.

“Simply having more data does not make you more accurate,” Stein stated.