Hartford's Deepa Soni says no to ChatGPT, yes to advanced AI and cloud

Deepa Soni is navigating the newer territory of advanced artificial intelligence carefully.

The chief innovation officer at The Hartford, who manages a team of 8,000 technologists, has decided, for instance, to bar employees from using OpenAI’s popular ChatGPT model, for fear that sensitive information about customers or the company itself might get leaked out. At the same time, her team is considering two dozen use cases for an enterprise version of a large language model.

In an interview with Digital Insurance, Soni shared some of her current thinking about advanced AI, new sources of data in risk analysis and her company’s shift to cloud computing.

Are ChatGPT and large language models on your radar? 

Totally. We’re leading the analysis, in that we’ve already done a couple of experimentations to understand hands-on experience. So what it means to us as a company, we see the potential, we see the disruption capabilities that this will provide, and we’re working to make sure it’s safe and it’s secure in an enterprise environment. 

What might be some use cases for you? 

We have about two dozen use cases that we’ve already thought about, and we’ve experimented with three already. The biggest use case that we’re working on right now is where we have knowledge workers that are going into various domains to understand our policies, our procedures, our products and pricing. I think we can bring a ChatGPT interface to them and an enterprise large language model. 

Some banks are banning the use of ChatGPT itself among their workers. 

We did, too. 

What was your reasoning? 

Anything you put in ChatGPT becomes data for it to train on. And we didn’t want personally identifiable information or our proprietary algorithms to get fed into it. This is what happened to Samsung. We have a policy that we issued. 

Did the Samsung case inspire you to ban it? 

We had already made the decision before that. We had anticipated that and we had actually published a policy a month ahead of that. 

Are there any other places where AI has become important to you? 

AI is very mature in Hartford. We’ve been using traditional AI and predictive AI models across all our value streams from sales to servicing to claims to underwriting. 

Is there anything on your tech roadmap that you could tell us about? 

There are two big things on our tech roadmaps. One is helping the businesses with step up change for data and analytics AI capabilities. I think now we’re in a position where we can reimagine our processing based on being digital data led, AI fielded processes. And I think within technology, to enable agility, we have to automate a whole lot of our own processes. So as you go to cloud, you can modernize more and more of your own processes within the IT shop. And I think that’s going to eventually fuel the business agility. We call it automation through technology and within technology. In technology, how we deploy, how we release, how we build our code, those are all processes that are dependent on whether you are on-prem or on your cloud. If you’re on cloud, then you are more automated. 

Have you gravitated toward a particular cloud partner like Google, Amazon or Microsoft? 

We are partners with AWS right now. A multi-cloud strategy is part of every company’s roadmap. So right now we’re focusing a lot on AWS, but we are multi-cloud. 

What percentage of your computing is in the cloud today? 

We’re about 25% in the cloud today. 

Are there certain things you’ll just keep on premise forever? 

Very few. We have a roadmap to leverage cloud modern technology more and more. Legacy technology is going to be an inhibitor to growth. You can’t innovate, you can’t blend with the ecosystems. You can’t really leverage a lot of capabilities. So we are, we have a pretty aggressive technology and data agenda to modernize. 

What are some examples of new data sources you are tapping into to better analyze risk? For instance, when it comes to auto insurance, what extra data is helping you get a better sense of how much risk a driver represents and what the price of the insurance should be? 

The latest trend in the auto industry is really around telematics. That’s your actual driving habits versus you saying, I drive X miles. But it’s not just the X miles, it’s where do you drive the X miles? So it’s a next level of thinking about the data. Another insurance trend is internet-of-things sensors. Telematics is one way of deploying sensors. But in the construction industry, we could put sensors on every part of the construction site and you can get way more data to analyze. 

Are you doing these things today, or are they on the roadmap? 

We’re starting. That’s an innovation where we are providing that service to some of our customers. We are starting to put sensors across construction sites, to get the data and we can watch it in real time. 

So you can see if workers are being sloppy or reckless on the site? 

Or even what the patterns are. If you have a heat sensor, there could be an alert that says, well, some part of the building is heating up, before a fire starts. So it’s more preventative, and I think that’s a big theme of the insurance industry: How do we become more preventative rather than just dealing with the calamities? 

You can’t prevent a wildfire necessarily, but safety practices you can have some say over. 

If construction workers are working on the construction site, we could have them put some sensors that would tell us whether they’re near a fire, near water. Those are all the preventative measures.