Interview with Nationwide's Guru Vasudeva: Keeping Humans in the Loop

The smartest AI leaders ask themselves a simple question: How can this tool help people do their jobs better? Or for students to learn better. Some roles will need fewer people, no question. But the leaders who really get it democratize work, giving people access to the kind of expert guidance once reserved for a few.

My guest today is Guru Vasudeva, Senior Vice President and Chief Technology Officer of Property and Casualty Technology at Nationwide. Nationwide is one of the largest insurance and financial services companies in the United States. As Senior V P and C T O, Grewal leads a workforce of three thousand associates and contractors. It’s their job to build and maintain technology solutions for Nationwide’s insurance lines. Guru is also accountable for driving transformational programming spanning simplification, digital enablement, data and advanced analytics. His tenure at Nationwide has been marked by numerous accomplishments in leadership roles spanning information technology strategy, cybersecurity, disaster recovery, and large program management. Guru was at IBM for six years prior to joining Nationwide. There, he was an executive architect working on large internet-based solutions for insurance, retail, and government. He's also worked at Tata Consultancy Services. 

Transcript

Interview with Nationwide’s Guru Vasudeva: Keeping Humans in the Loop (Lightly Edited Transcript)

00:00:00 Guru Vasudeva
Today, there is actually a great article that came out in Business Insider. This article really talks about companies that are leveraging AI to enhance their employees’ work and are really seeing growth in the company and also their headcount.

00:00:35 Peter Creticos
Welcome to the Institute for Work and the Economy’s podcast, “The Many Futures of Work.” I am Peter Creticos, President of the institute. Our podcast shares personal stories and experiences from work, business, the economy, and civil society. You’ll hear from people solving real problems and the lessons they picked up along the way. We organize the podcast into a few series that reflect what the institute cares about: artificial intelligence and its uses at work, accessibility and fairness in employment, and entrepreneurship. Today’s conversation is part of our AI series. Artificial intelligence is spreading faster than any technology before it. The people behind Harvard’s AI adoption tracker say it’s taken Gen AI about a third of the time it took personal computers to reach the same level of use. Why?

00:01:33 Peter Creticos
A few reasons stand out. It’s cheap or free, built right into search engines like Google’s Gemini. It’s everywhere. You get a result the moment you type a prompt, and maybe most interesting, it talks back. Business leaders have always driven the adoption of new technology. Today’s AI is still designed to stay within human control. The smartest AI leaders ask themselves a simple question: How can this tool help people do their jobs better? Or for students to learn better. Some roles will need fewer people, no question. But the leaders who really get it democratize work, giving people access to the kind of expert guidance once reserved for a few. My guest today is Guru Vasudeva, Senior Vice President and Chief Technology Officer of Property and Casualty Technology at Nationwide. Nationwide is one of the largest insurance and financial services companies in the United States. As Senior V P and C T O, Grewal leads a workforce of three thousand associates and contractors. It’s their job to build and maintain technology solutions for Nationwide’s insurance lines. Guru is also accountable for driving transformational programming spanning simplification, digital enablement, data and advanced analytics. His tenure at Nationwide has been marked by numerous accomplishments in leadership roles spanning information technology strategy, cybersecurity, disaster recovery, and large program management. Guru was at IBM for six years prior to joining Nationwide. There, he was an executive architect working on large internet-based solutions for insurance, retail, and government. And at Tata Consultancy Services. Welcome, Guru. Thank you for joining us this morning.

00:03:35 Guru Vasudeva
Thanks for having me, Peter. I joined Nationwide in 2004; that’s like twenty-two years ago. I joined as the chief architect, and at that time, the internet was still taking hold across commercial usage. We were in the early stages of insurance online and servicing making payments online. I was really hired to help bring that expertise from my work at IBM, and that really gave me a good insight into the breadth of the company. Nationwide is… many people think about Nationwide as an auto and home insurance company. That is a very big part of our business, but we are also the largest underwriter of agricultural properties, agricultural businesses in the country.

00:04:26 Guru Vasudeva
We also have a large presence in excess and surplus lines. These are high-risk lines of commercial insurance business as well. I really got to see the breadth of the company. Those experiences really gave me an opportunity to become a business unit CIO for some of our brands in the 2009 timeframe, and then later on lead our enterprise technology architecture, cybersecurity, data analytics, and so on. Most recently, I was running our infrastructure organization. That’s when I got to really transform the company to leverage the cloud and also deal with the pandemic. Three years ago, I took on this job of running the technology for all of the property and casualty lines of business.

00:05:15 Peter Creticos
I was wondering if we could conceptualize three roles of AI—three roles for AI at Nationwide—go through those and share your thinking about what those roles are.

00:05:25 Guru Vasudeva
At Nationwide, we think about AI in broadly three categories. First one is everyday AI. Second is transforming business processes, reimagining them to leverage AI. Third is reimagining software development using AI. And I’ll talk about each one of them a little bit.

By now, pretty much all of us are using tools like Chat GPT or Gemini or Claude for our personal purposes. People are using it to plan vacations or think about what clothes. In fact, I am leaving on a vacation here soon and actually leveraged it to really research my itinerary to give me suggestions on what my packing list should be. It was able to really look at the weather, all of the stuff that would have taken me.

00:06:18 Guru Vasudeva
Maybe an hour to really put together because we’re going to very different climates. And it did such a fantastic job on everyday AI. We want to bring that capability to all Nationwide associates, within our security, confidentiality, and privacy parameters, for their day-to-day knowledge work across the company. So pretty much everybody has access to it. We are making it increasingly available to our interns and contractors. Those tools are connected to employees’ email, their SharePoint databases, and their calendars. It is able to do the kind of synthesis and summarization and help you plan your next steps. Just like you can do at home with Claude or ChatGPT.

00:07:11 Guru Vasudeva
And it’s been so powerful when I do roundtable discussions with associates. I am trying to get a pulse of what’s happening in the company. One of the questions that I often lead with is, ‘What are you excited about? What’s really going well?’ Sixty to seventy percent of the feedback in the last three months has been, ‘We love what it is actually doing to help me do my job better.’ For example, one of the business unit CFOs shared how she uses that tool to really analyze financial aspects: what if I exclude this, what would happen to the numbers? Before, she had to go to multiple people to get those insights for her. It’s able to really connect to all of those databases and get that type of insight for her. We have associates who are really leveraging it in planning their projects.

00:08:08 Guru Vasudeva
Or thinking about requirements testing plans. So all those things. So the everyday AI, we really feel good about the progress that we have made, and we are happy that people are finding great benefit from it. So together, I think we have. We feel good about the progress we’re making, giving our associates power in their hands.

The second category is transforming our business process completely. When you think about AI, we always did predictive analytics, but they were done in a more of a large time scale timeframe. Where what we were able to achieve in 2009, 2010, is really doing machine learning to help make near real time predictive analytics. If we have ten people started a coding process on our website, whom should we call? And if they abandoned it halfway through.

00:09:06 Guru Vasudeva
If we had only one person to call them back, who should we call in what order? What is really remarkable about the generative AI, which is the second category that is actually what’s exploding in the last four years here, is that its ability to really read, synthesize, and summarize unstructured data. Property and casualty insurance is a perfect place to leverage this as an enhancer to the work that we do. In our field, we underwrite very complex businesses. They sometimes have operations in multiple states, they own multiple buildings, they have lots of cars or trucks in their fleet. And we need our underwriters if we’re really trying to renew that insurance or quote a new insurance for them. We need our underwriters to really read through all of that info and make judgment calls. What we are finding is most of their time is spent in some grunt work that is really not at their level of expertise. Not every business gives you information in the same format; they send it in different formats. They sometimes may not send you some information. If I am an underwriter starting to really review something and I need a hundred data points in a submission for insurance, there may only be seventy of them. Now they have to go back and ask for those thirty more fields or pieces of information, and all of that cycle time takes a long time. It just wastes time for everybody. They end up copying and pasting things from one system to the other in order to really get the info. What we are really looking to do is supercharge generative AI tools to really do it, and that is what we are really working on. Again, the key here is that it’s really human in the loop.

00:11:02 Guru Vasudeva
Human in the loop is the key here because we don’t want a probabilistic model like generative AI to really make decisions. We want it to do some of the grunt work so that we can actually free up the time so that our underwriters can really spend more time on the analysis, review and decision making. We also believe that if we can achieve this, we can focus on the most important risks that we should underwrite. We believe that we can actually double the territory that our underwriters can manage if we were to really deploy these kinds of tools. And those are not easy to really implement, and we are still working through these kinds of use cases. That’s the second category.

The third category is really I would call it the hero use case of the whole generative AI revolution that’s taking place. If you really take a look at where people are having the most success leveraging these tools, it’s to really increase the speed at which you can do software development. You can fix issues and so on. We spend over a billion dollars on technology because everything that we do, whether in financial services or insurance, is all IT-enabled. We are experimenting with tools like Microsoft GitHub Copilot and Claude Code, sometimes to really see how we can improve that speed of development. What is really interesting is it is definitely able to do autocomplete very, very well today. Our developers are adopting that very, very quickly.

00:12:55 Guru Vasudeva
You and I start typing; it is able to, in a document, suggest next words and so on. We have wide adoption. It’s the next step where the magic is: where agentic development is happening, where we are really letting it take a requirement and really go code. We’ve got use cases where we have done it end-to-end, and in other places, our teams are really struggling to make the shift. How do we really change the requirements? How do we verify what it has done is accurate? All those things that we are still working through.

00:13:30 Peter Creticos
One of the things that’s remarkable about what’s happening with artificial intelligence is that the adoption rate of artificial intelligence has been faster than most other technologies, electronic technologies. It achieved the same level of adoption three times faster than personal computers did. For you as a CTO, I would guess that the biggest challenge is just being able to keep up with what’s happening.

00:14:00 Guru Vasudeva
Totally agree. In 2022, ChatGPT launched, people really looked at it as something cute that you could write poems with and so on. Within six months, capabilities increase. Within another six months, you can do more analysis with it. It is able to connect to the internet, synthesize the new information as well. And then a year later, they introduce deep research capabilities to really look at a particular topic and do synthesis of that information. It’s been fascinating to really see the advancements. I was reading an article about the capability of these generative AI tools has advanced so much over the last four years, but the enterprise adoption is not even scratching ten to fifteen percent of that, and that is the gap that we are really trying to fill. My breadth of experience, having been at Nationwide for so long, having really played many different roles, I think, gives me a unique vantage point to be able to really think about what the company needs and what the capabilities of these tools are. And how do we really bridge that gap?

00:15:14 Peter Creticos
I’ve been intrigued by how AI has the potential to assemble data in the way you’re talking about and produce results that can support decision-making. It’s able to do this on a continuous basis, as new data come in, generating new conclusions. The ability to go from point to future kinds of predictions and allowing.

00:15:46 Peter Creticos
A more rapid reassessment of circumstances of risk in your case really moves from a very static kind of model to where you know maybe you are coming together every year to evaluate a situation. Now you can do this on a continuous basis.

00:16:05 Guru Vasudeva
It is technologically definitely possible. The question is, society-wise, norms-wise, how the contracts are structured. Are we really willing to change that much? Technology can do a lot of things, but I think we also have to go along with the pace at which businesses are willing to buy these kinds of services from us. For example, we actually do this today with our auto insurance. When the pandemic hit, some of our customers said, ‘Hey, you already have telematics deployed in partnership, why should I pay on a per month or six month basis? Why don’t I pay for the miles that I drive?’ And there is definitely a use case for that. We implemented it. We call it Smart Miles.

And what we are finding is people don’t like the unpredictability of the prices. People want a little bit of predictability in their lives, so they’re willing to really give up certain sometimes discounts, but in exchange, they also get a guarantee that the price won’t go up if they drove to Florida for our vacation. So I give that as an example, because I believe that human ability to really adapt is not at the same rate that technology is able to remove.

00:17:28 Peter Creticos
I think one of the great fears as people wrestle with AI is what’s going to happen to their jobs, massive layoffs. Companies like Microsoft and so forth, Claude generates for me a weekly report which is my augmentation, replacement and discovery report. I ask it to scan publicly available resources to tell me what’s the news on augmentation and what’s the news on replacements? Usually layoffs, things of that nature. On the replacement side, you are seeing very large numbers, and with great frequency now. Most of this comes out of Silicon Valley, but it happens. Your perspective, as I understood, was really about how this helps people do what they want, what they’re asked to do as an assignment. What do you think is the change in the role of the human in this process? Are there new opportunities that potentially come about because of these changes?

00:18:32 Guru Vasudeva
I think that’s a great question, and it’s something that we think about deeply at the company. But I also think about it as a citizen of this community. Let’s take a detour, like maybe a little jog through the history. By now, lots of people that are really studying about AI and its impacts have heard about the Jevons Paradox. So Jevons was an economist in England and the steam engine was becoming far more efficient. One of the predictions was that it would really result in a significant drop in the use of coal, and as a result, coal miners would lose jobs. Jevons really predicted the reverse would happen because now steam engines are cheaper to operate. People would use more of steam engines to do more work. People instead of having two pairs of shirts and pants. Now they have ten of them, right? So that means that you need more engines that are operated by the steam engines. He predicted use of coal to power. These new steam engines will surpass the efficiency gained by these steam engines. And it turned out he was right. And in fact the coal usage actually exploded. And I am sure I was in Chicago as mentioned earlier last week. And I saw the traffic. Oh my god, you guys have some traffic. And highways are expanding. But the traffic is increasing as long as the demand is elastic and the cost becomes lower. People’s use of it keeps going.

Let’s take a more recent example. Radiologists. In 2016, one of the godfathers of AI predicted that image recognition is getting so good that radiologists’ jobs will be obsolete in five to ten years. Here we are, ten years later. The number of radiologists in the US has actually increased by more than ten percent, and there is a severe shortage of radiologists. Even though use of AI has actually increased dramatically in the field of AI, it has brought the cost of doing radiology analysis down effectively, so the usage has gone up hugely. I really tend to take a very positive long-term view of what is possible with these technologies, and the same thing is true in the software engineering space. I would challenge some of the headlines because the number of things that our businesses want to really automate is huge.

00:21:24 Guru Vasudeva
And we cannot every year get to them. In fact, I am hearing from a variety of different universities in the area that the computer science enrollment is actually going down by a big number, ten, fifteen, sometimes even twenty-plus percent. That really concerns me because I think the need for software engineers is actually gonna go up dramatically over the next ten years. So I personally believe that as the cost of automation becomes lower, we will do more software engineering, and it’ll be easier for lots of people to do it. And we see it today. The way that the company is thinking about it is that we want to leverage these tools to empower our associates. We want humans in the loop. There was another dimension to the question that you asked.

00:22:21 Guru Vasudeva
As AI does more of this, what becomes the job of the human that’s actually directing it? What’ll happen is imagining and checking if the tool really generated what you wanted, doing auditing. But as a result, we are able to do more things like micropayments here and there and those kind of things. And how do we really keep up with it today? We use techniques like auditing to make sure that transactions are reconciling and so on. I believe that is going to become one of the shifts in the type of work people will be doing. They will be directing, checking, and making sure that the work is getting done the way you want. At Nationwide, we really talk about human-in-the-loop a lot because we want that at the end of the day. It doesn’t matter, but that I wrote that email or I got the help of a tool to write that email, it is my email. It is my responsibility to send that out. So it becomes mine. That means that I need to take full accountability for it, whether I got the help. It’s just like the calculator, right? That actually leads to a fascinating concept, which is a paradox: the more you can get knowledge easily, the more domain knowledge becomes important for you because without good domain knowledge you wouldn’t know what questions to ask. You wouldn’t know whether to assess if the tool AI is telling you it’s right, and you wouldn’t know which ones to really act on. So, paradoxically, people with deeper domain knowledge will become even more effective with these tools. I can only imagine if you are running a university, how do you really assess what the student wrote, whether how much they leverage. People are worried, does even learning matter anymore? And I tend to discount that idea because as I was sharing earlier, the deeper domain knowledge you have, then you can actually debate with the tool even better. It’s like references, right? So we need to give credit where the credit belongs. But we all build on each other’s ideas. My wife is a biologist and she can get into the depth of any research paper that comes out about it. I have no idea about any of that stuff, so I cannot use it for biology-related topics. But I can use it for investments. I can use it for technology, because I have the domain knowledge. So I think it’s also fascinating to really think about what other domains that I should go study.

00:25:12 Guru Vasudeva
I think the more you really leverage it to get better as an individual, the more effectively you can use that tool. It’s paradoxical because, in one level, knowledge is almost becoming free. I think it’s going to be different, but I think it’s going to be exciting.

00:25:36 Peter Creticos
I got into a debate yesterday over the phrase ‘critical thinking.’ That idea as a skill keeps coming up, and you know, time and time again. In terms of one of the essential skills in being able to work in an AI environment is to have good critical thinking skills. I don’t know that we all understand what that means. And maybe you don’t agree with that—that critical thinking is a skill—but I’d like to hear your reflection on that in terms of where critical thinking and what you think of when you say that.

00:26:13 Guru Vasudeva
When I think about critical thinking, it is the ability to really understand that topic and its implications, the ability to see what part of it is right, what part of it is not right or needs to be made better. So those are all the things that come to mind. Critical thinking is another one of those kinds of concepts that can mean different things to different people. But there is a very key kernel of truth in both of those: learning to learn, and then also critical thinking. It is the ability to really diagnose the issue at a more deeper level than the surface level, and I think those skills are key. But I do wonder about this: connecting the dots to the previous other topics that we talked about. I cannot critically think about biology topics like my wife can. I can think critically about financial topics, I can critically think about everyday life issues. I can critically think about software engineering topics. So you need the domain expertise to critically think. It’s going to be fascinating learning. What is the role of us in partnering with this tool is going to be something that we are all going to learn together.

 Not too long ago, we thought that maybe using calculators and tests is a bad idea. But in the US, we use it very regularly now because it is a tool. As long as you know how to do the math, it just helps you do the math faster. Interestingly, some of the schools, some of the tests in India even today don’t allow the use of computers. I mean use of calculators. At the end of the day, we need people to have that deep domain knowledge. I think this critical thinking is a very key tool as well in everyone’s belt.

00:28:08 Peter Creticos
What thought is being given to equipping your associates with new skills to deal with the changes in their roles?

00:28:19 Guru Vasudeva
One of the things that I am really proud of the way we do things at Nationwide is we believe in standard work concepts. We need certain standards. What is the job of a requirements analyst? What is the job of a tester? What is the job of a software engineer? And we have training around that. And we are rebooting all of that training to focus on what we call AI dot DLC. We are very much training. We have thousands of people, so we cannot train everybody. Instead, what we are doing is we’ve identified two or three key people from every team, and they are training them. They are the leads, and then their job is to train their teams. So it’s like a train the trainer approach.

00:29:04 Guru Vasudeva
And we are constantly revising this training because the tool landscape is also evolving. Do these roles come together? How do they come together? The last thing that I would say is we also have a great set of baseline metrics that we use to measure productivity, speed to market, cost of software development, and we are trying to see how those things are impacted by these tools. Absolutely. We are giving a lot of thought to it. And I tend to take a more optimistic view. It’s a tool that we can leverage. And at Nationwide, we really give a lot of focus on human-in-the-loop use of AI. It’s really an enhancer. If you start using it, it could lead to really differentiating yourself in the marketplace and lead to revenue growth or improving the customer satisfaction. And that is exactly how we think about it. It is really boost the capabilities so that we can do more, and serve our members better. That’s how we think about it. I just wanted to wrap by stressing that point.

00:30:23 Peter Creticos
That does it for today. Thank you to my guest Guru Vasudeva. Also, thanks to William Edwards, who directs and edits this podcast series. And to Evan Hughes, who produced today’s conversation. As always, thank you to Ronnie Malley for today’s music. Ronnie is a master of the oud and a multi-instrumentalist living in the Chicago area. You can learn more about Ronnie and his work at interculturalmusic.com.

Finally, support for today’s podcast is provided by the board of directors of the institute. You can learn more about the institute at www.workandeconomy.org. Your comments and financial support are welcome. And finally, you are welcome to contact me directly at creticos@workandeconomy.org. That’s spelled C-R-E-T-I-C-O-S at workandeconomy.org. Thank you for listening, and have a great day.