Contents

there’s varying degrees of sophistication and

preparedness at different organizations, be they

carriers, agents, brokers, vendors, to do that.

Certainly, there’s a bit of a transition period that I

think we’ll all go through, but eventually that will be

the norm, if not something newer and better.

How do you stay innovative while still

keeping the technology in house?

Mattern:

When we look at our internal staff, we

have to consider both our business

internal as well as our IT, and also, I

think, now our partners. We talked

a lot about partnership, [how] our

solution partners as well as our

sourcing partners are extremely key.

We need to think of our in-house

from all those perspectives. Then

when you think of the talent and you

want to keep them empowered and

keep them engaged, that’s critical—

that engagement, empowerment to

stay innovative. It’s not something

that you set up the office and you’re

ready to go.

Some of the events I talked about

earlier, the encouragement, senior

leadership involvement, and I think

[all] that’s key. Another part that’s key to it is the

career development. We’ve got to stay current ...

That’s something where you want to make sure [is]

in everybody’s career path and their training, that’s

across the board.

Leveraging the partners—cannot state that

enough. They’ve got the scale, especially in, say,

smaller or mid-tier carriers, to be able to take

advantage of that and keep that innovation flow and

make sure that you’re going through that life cycle—

those four stages, exploratory, experimenting, the

ability to then do tactical, and then do strategic. To

constantly do that with all new technology, and keep

people engaged, keep them empowered, and make

sure it’s part of their career and ongoing training.

That’s the way to stay innovative and keep it fresh.

O’Shaughnessy:

One interesting item that

resonates from what you said there is, at HSB, we’ve

had a lot of engineers at HSB over the last 158 years,

but we now have sensor engineers, IoT engineers.

These folks have the coolest job in insurance,

because all day long, they’re trying to break and

play with different devices around fire, water. ...

It’s pretty innovative. It’s different, but it’s now

something that’s necessary in the insurance space

that previously just didn’t even exist. To me, that’s an

example of some innovation that we probably would

not have envisioned a few years ago.

George:

I would add, as well, the retention of

talent and energizing and incentivizing because

we see so much movement of talent, and at times,

books of business for another shining little star. The

ability of that technology, and how

good the technology is, from where

they currently are to where they’re

going to go is a key component,

and what we see is a lot of the

underwriting and their books of

business can be moved from one to

another. Retaining that and also what

technology are you on, the familiarity

of that technology is going to be key

to retain key individuals within the

organization.

Is data being harvested or

leveraged as much as it should

be, and where is that data

being generated from?

O’Shaughnessy:

That’s probably

a topic for a few different episodes in the future.

It’s a big question. I would say that, yes, it’s being

harvested, but we’re probably still in the early

innings of where we can go. Traditional insurance

data, insurance to value, and all that good stuff,

most carriers have done a really nice job over the

past several years in keeping up with that and

properly leveraging it.

I would argue that, right now, in the marketplace,

we’re seeing a little bit of

A

Tale of Two Cities

of

carriers who have been really good at leveraging

and harvesting and embracing that data, and maybe

those who weren’t as successful in doing that in

terms of some of the results we’re seeing in the

marketplace.

As we’re looking at the future, IoT is one example

of new data elements that are coming in. These are

data elements that never existed before. The volume

of them blows away any other data element out

there. We’re talking about billions of bits of data that

we’re getting routinely. Over time, it allows carriers,

underwriters to make more informed decisions.

BEST’S REVIEW

•

NOVEMBER 

Kris George

29

How and where are insurers deploying

AI technologies that they’ve developed

themselves?

Wein:

Well, insurers are actually doing a lot

with large language models and in a couple of cases

building out capability. We see Zurich doing risk

modeling using large language models. Nationwide

has partnered with Digital Owl for medical records

in the context of life underwriting. And then Nassau

Re has done some stuff with SpyGlass to support

the call centers, basically generating

call center scripts in real time ...

In addition, we’ve seen Axa build

a centralized large language model

through the group and now are

distributing it across the world to

their various subsidiaries.

Are there best practices for

developing AI systems or

processes?

Naveed:

The best practice when

you’re developing an AI model is to

ensure that it will act as you intended

it to in production. So there are a few

key parts to this. Ensure that there’s

parity between the development

environment and the deployment

environment. Make sure that when you build and

test a model on your local machine, it works how

you think it works in production. And there [are]

principles you can adopt from software engineering

that can be leveraged here. The other piece is to

ensure that your service is stable. What if you get

100 requests at the same time? Is it going to fail?

How is it going to fail? How do you let the customer

know?

And also, how do you ensure that you know the

limitations of your technology before putting it out

there into production so you can set expectations

appropriately? The other piece is you have to ensure

that your AI system is bias-tested and monitored

according to both the best practices for the use case

and the regulatory conditions for the industry. So as

we all know, insurance is a heavily regulated industry

and it actually is the one industry where AI is being

regulated by the actual industry. So by the state

commissioners, they are developing the requirements

around documentation and filings around AI models

that are being used in production.

Wein:

Obviously, carriers need to make sure that

the models don’t hallucinate when they give results.

For example, I mentioned life underwriting. So if

it’s summarizing medical information and it gets it

wrong because maybe there’s bias in the model or

inaccuracy that could run into problems with the

regulators. What carriers are doing is they’re using

a technique called retrieval-augmented generation,

otherwise known as RAG. And the idea is to

essentially preconfigure a context window with

factual information that will prevent

the model from hallucinating vis-à-

vis a probabilistic guess at what the

model was trained at versus a more

deterministic outcome that carriers

may need. So the idea is to increase

or lower the temperature on the

model.

Lowering the temperature

means more [being] deterministic,

increasing the temperature more

probabilistic, but using a context

window to ground the outcomes so

that the underwriting, the medical

summary, call center, etc., will

come out with the right type of

information.

Are your carriers today developing and/or

deploying smaller language models?

Wein:

Actually, very, very few carriers are

building small language models. The vast majority

are subscribing to large language models, which are

provided by vendors and using retrieval-augmented

generation context windows to get outcomes

they want. However, there are cases where a small

language model is a better choice. Why? Because

they control the model themselves. They build it

using open-source tool kits, and it may, in fact,

be cheaper. It may allow the carrier to narrow the

model’s coverage to be much more specific. And

actually it may allow for training or retraining to

take place in a much more comprehensive and tested

way.

So responding to the regulators or responding to

very, very niche type of requirements may mean a

small language model is better. The reality, though,

is carriers, for the most part, are not doing that

unless they’re fairly large and can attract the kind of

talent you need to build out a small language model.

BEST’S REVIEW

•

NOVEMBER 

Mitch Wein

31

©2024 Conning, Inc. All Rights Reserved.

The right tools in the right hands.

You need the right tools to sketch out a customized vision, but it takes an expert team to construct a framework

that stands the test of time. With over a hundred years of serving the insurance industry, a high level of expertise

in strategic asset allocation and a partnership approach, Conning has earned a reputation for delivering

innovative and client-centered solutions across a wide range of asset classes. Our capabilities extend beyond







LEARN MORE ABOUT OUR ASSET CLASS CAPABILITIES AT CONNING.COM.

North America

|

Europe

|

Asia

“In a market that is very volatile and

challenging, it provides investors the

ability to set the return rates based on

the structure of the surplus note, and

maybe get them more comfortable with

putting capital in the market.”

Chris Draghi

AM Best

to claims handling. They are similar to mutual

insurers, with a few operational differences.

“If it’s a three-party reciprocal: Member 1 insures

Members 2 and 3; Member 2 insures Members 1

and 3; and Member 3 insures Members 1 and 2,”

Matz said. “That’s the technical difference, but

in reality, the money goes into a pot, claims and

expenses are paid out of that pot, and if there’s an

underwriting profit, those dividends can be shared

via a subscriber savings account.”

The exchanges have already shown success. Logan

said Pure’s member retention is very high and “we’re

really proud of that.”

“You have a lower cost of capital the more

you build up member capital so there is less need

to borrow,” Logan said. “When those funds are

allocated into SSAs, they are not accessible to the

member. They can’t take them unless they leave,

so there’s one incentive to leave. Members remain

if you treat them in an extraordinary way. An

important part of being a healthy reciprocal is that

you delight your members so they stay and that

capital does not leave you.”

The influx of new insurers—including the

reciprocals—occurred after the passage of historic

Florida legislative reforms that placed restrictions

on the use of assignment of benefits and, more

broadly, one-way attorney fees, both of which were

considered critical pieces in stabilizing the Florida

legal landscape for both carriers and policyholders.

Lawsuits, whether assignment-of-benefits-related

or first-party, drove up direct cost and containment

expenses, AM Best said in a Best’s Special Report,

Florida Property Insurance Market Improves;

Reinsurance Dependency Remains High

.

One reason reciprocals may specifically have

gained popularity in catastrophe-prone areas like

Florida is because of the capital structure, with

such entities typically relying on surplus notes for

their initial capitalization, AM Best Director Chris

Draghi said.

Because the companies are member-owned, they

don’t have to be concerned with maintaining a low

combined ratio in an area full of risk.

The model “also shields other affiliated parties

from financial obligations should the entity be under

financial duress,” AM Best said in its report.

“In a market that is very volatile and challenging,

it provides investors the ability to set the return

rates based on the structure of the surplus note, and

maybe get them more comfortable with putting

capital in the market. Of course, payment of

principal and interest for surplus notes are regulated

and need to be approved by the governing body—in

this way the reciprocal’s ability to maintain healthy

results remains important,” Draghi said.

Growing a surplus through the surplus

contributions also reduces the pressure on the need

to have a lower combined ratio for an underwriting

profit, Matz said. “Therefore, it’s a better model for

cat-exposed areas where the cost of reinsurance is a

very significant component of the equation. It builds

up surplus and it just seems to work a little bit

better, particularly in the catastrophe-exposed areas,”

he said.

Reciprocal Growth

Reciprocals are not new. Pure, which deals in

the P/C needs of high-net-worth individuals and

families, launched in Florida in 2006 and has grown

to about 115,000 members, Logan said.

Pure Specialty Exchange, the company’s non-

admitted carrier, is rated A by AM Best. The

company writes in all 50 states and recently

BEST’S REVIEW

•

NOVEMBER 

35

property company in the

state by direct premiums

written, according to AM

Best’s report. Orange was

one of five new reciprocals

and 20 new property/

casualty carriers overall that

were approved to operate in

Florida in the past two years.

The other reciprocals were

Condo Owners Reciprocal

Exchange, Manatee

Insurance Exchange,

Ovation Home Insurance

Exchange and Trident

Reciprocal Exchange.

“Reciprocals are a great

model for Florida, and that’s

evidenced by the number of

reciprocals that have been

formed or had conversions,

or other stock companies

that have also developed

reciprocals,” Matz said.

“Reciprocals aren’t the only

way to do it, but they are a

good way to do it.”

Draghi said the growing

number of insurers using

this nontraditional model

is “probably related to

the structure in terms of

insulating the capital at risk

because it is surplus-note related.”

“There is a level of expectation for returns from

the investors that are placing the surplus note,”

Draghi said. “This is all predescribed rather than

hoping to make a return through full ownership

and getting margins and being profitable. To pay

those surplus notes back, a reciprocal needs to

be profitable and in a position to pay that back,

so there’s no guarantee they will get that money

back. It’s all dependent on the performance of the

reciprocal itself.”

According to the Alirt report, 18 of the newer

reciprocals reported a total surplus of just over $1.0

billion at the end of 2023, which was supported by

$1.3 billion in surplus notes.

Overall, Alirt estimates that there were 64

companies operating as pure reciprocals at the

end of 2023, and that

they wrote $44 billion in

direct premiums that year,

representing about 5% of the

total U.S. DPW.

Alirt did express concern

about the sustainability of the

model over the concept of

“playing with other people’s

money.”

“In short, with less

investor/manager funds

tied up in these insurers,

there may be an incentive to

opportunistically generate

premium in difficult property

insurance markets,” the Alirt

report said. “This is especially

true where these interests are

generating fees based, in part,

on overall premium volume.”

“I think the sustainability

issue really comes down

to the same issue as the

sustainability of a stock

company,” Matz said. “It’s

the management of the

underwriting and doing all

the things that you need

to do to be successful.

Companies do fail from time

to time, but whether they

are a reciprocal or a stock

company is not the reason they fail.”

Reciprocals’ Participation in Citizens’

Takeouts

Orange launched on Dec. 1, 2023. It

subsequently came away from two rounds of

takeouts from the Citizens Property Insurance Corp.

of Florida with about 13,000 new policies, Matz

said. The exchange has been approved for another

15,000 takeouts in November, although Matz said

he realistically expects to get about 5,000.

In February, the Florida Office of Insurance

Regulation approved Condo Owners’ request to

assume up to 250 commercial residential multiperil

policies and 150 commercial residential wind-only

policies.

As part of the agreement, Condo Owners would

BEST’S REVIEW

•

NOVEMBER 

“It’s a better model for

cat-exposed areas where the

cost of reinsurance is a very

significant component of the

equation. It builds up surplus

and it just seems to work a

little bit better, particularly in

the catastrophe-exposed

areas.”

Don Matz

Orange Insurance Exchange

37