AIG’s chairman and CEO Peter Zaffino is overseeing a new era of stability for the insurance business after a tumultuous few years. In the aftermath of receiving one of the biggest U.S. government bailouts in history during the financial crisis, AIG saw a series of management changes and has sold off multiple assets to repay the Treasury. Zaffino, who joined AIG from Marsh McLennan in 2017 and took the helm in 2021, says AIG has shed more than $1 trillion of exposure, and he continues to streamline the business. That has included spinning off Corebridge Financial and selling AIG’s global personal travel insurance and assistance business in a $600 million deal.
[time-brightcove not-tgx=”true”]
As part of his growth mission, Zaffino is looking to artificial intelligence, partnering with Anthropic and Palantir to integrate the powerful technology into AIG’s processes.
TIME spoke with Zaffino in June about how AIG is using AI, navigating risk, and his approach to leadership.
This interview has been condensed and edited for clarity.
You joined AIG in 2017 and became CEO in 2021, and you’ve led something of a turnaround. Can you tell me about that?
I joined AIG primarily to see if I could bring a team together and work through a transformation that was going to reposition the business and the company as an industry leader. Pre-financial crisis, AIG was a big life insurance company, it had aircraft leasing, so the company today is much smaller. And comparisons are very hard. What I can say is that the AIG brand is very strong. Its desire to really create solutions for risk issues, for its clients, was very strong, but its strategic, financial, and operational performance was not strong. And so we needed to do a variety of things. The company from 2008 to 2018 lost over $30 billion in underwriting—that’s never been done before, just because it probably had excess capital coming out of the financial crisis. And so turning the company around financially, but also creating a culture of underwriting and operational excellence was critical for us, and we began the journey in 2018 with a vision for creating that excellence in underwriting, controlling volatility. The company over a period of time, when we were leading this transformation, shed over a trillion dollars of exposure. And then we repositioned the company by investing in our end-to-end process, our data, our technology, our cloud computing, how we actually service our business. And that journey proved to be incredibly successful. We have a great group of colleagues that wanted to take this on. It was hard work, it was deliberate work, but we came through the other end, and now we’re a top quartile performer. [We] have created that culture of underwriting excellence and operational excellence, and it’s enabled us to be very focused and having a sense of urgency on Gen [generative] AI, large language models, data ingestion and compute. If we didn’t do that foundational work, we would not be able to do what we’re doing today to accelerate the progress we’re making there.
How did you decide about which areas to step away from?
I spent a lot of time analyzing the performance of a portfolio, and what its expected outcomes are going to be, and what the volatility is around that, and felt that what really drove the exposure reduction was the company just took on too much volatility. So you didn’t really know what the results were going to be under certain scenarios.
Usually having $100 billion of natural catastrophes is a very active season. Now it’s every year. Last year we had $140 billion, and so we have to deal with this volatility in a world that’s changing and creating more volatility. That’s why I’m proud of the company in many ways—to be able to reduce volatility when the external environment is increasing. Volatility is exceptional. And now when we look at our catastrophes and the volatility around that we’ve become one of the most predictable and lowest-end volatility companies in our peer group. Reduced volatility, increased profitability, have more predictable outcomes, and have underwriting practitioners that are going to focus on underwriting profit and solving risk issues: That’s basically, at a high level, what we tried to do at AIG, and we were successful.
How does AI play a role in helping make those assessments?
Our strategy in AI has been built on five pillars, but the middle is the underwriter. And the underwriter is still making underwriting decisions and going to make risk and policy decisions. What we decided was to build an end-to-end process enabled by Gen AI and large language models to do a couple of things. One is to provide the underwriter with much more detailed and comprehensive information than what might be available today, and then also to decrease the cycle time dramatically. And then the corporation looks at how to optimize the portfolio in a way where the portfolio is having a better expected outcome relative to the classes of business that we write.
We took a philosophy that we want to document an end-to-end process enabled by Gen AI. So what does that look like? Setting up an agentic operating model. But how do we actually do this, and how do we get it to be part of the fabric of AIG, but also how are we going to create a cultural change? The strategy was, how do you ingest data when it comes in, and how can you get structured, unstructured text, PDFs, qualitative, quantitative data to be more accurate, and then information that may be missing for what would be really relevant to the underwriter. How do we create reliable sources that we can actually drive more information to make more informed decisions? What do you do with that? We already had started to digitize the workflow, absent large language models, but we began in the underwriting process to train large language models to go and extract that data. And so we use Palantir as a tremendous thought partner [to think about] about, how do we actually take this data and create much more value from the data? And then we adopted Anthropic Claude models. We have been working with each of their new releases to be able to enhance that data extraction, but also decrease in the cycle time.
One of our businesses that we threw in the deep end of the pool first, the results were far in excess of what we had for aspirations. And I thought we had high aspirations. With the adoption of large language models…we got a lot more data, a lot more detailed data, and we could do it 10 times faster. So now we are building out capabilities to scale this to our bigger businesses.
Has that affected the makeup of your workforce, as maybe you don’t need as many people crunching data because you’ve adopted this technology?
Here’s what we’re doing: Think about an underwriting assistant [who] would be the one putting in a lot of the data into models, extracting, making requests, pulling things. Those positions get simpler. However, as an underwriter at the center, able to do the cycle time now that, all of a sudden, they’ve got 10 times the amount of work, which is not quite true—I said we could do [it] 10 times faster. Do I believe we’ll do 10 times the amount of business? Not today, but I do think we will do multiples of what we do today. We are training that underwriting, sort of a system, to become a junior underwriter, so they can do 70% of the underwriting, and then the more experienced practitioners can do the last 30% or whatever that percentage ends up being over time, maybe it’s 80/20. Part of the cultural change is upskilling, retraining positions in a new world that enables them to be more productive than we were in the past. Now, if you ask me, What are you doing on shared services? Or what are you doing on finance? Or, what do you do in other places that you’re going to put in single agents? I would say yes, on a same store sales basis, implementation of large language models has to decrease costs for companies.
How are you navigating the current geopolitical instability as a business?
There’s so many different variables that one needs to look into. Tariffs have been hard to follow. But we’ve been very much focusing on the potential impact from our clients, or the effects of how we underwrite our business. That hasn’t really had a dramatic effect in the short run, but it’s something that we have been paying attention to, as to what are the implications. The geopolitical challenges are ones that create awareness for your organization, but also your ecosystem. Business resiliency is something that you focus on. Cyber attacks have already been heightened, but are going to a new level, and in terms of trying to stay ahead of what could be different ways in which companies can can be compromised, and I think that’s probably one that we spend a lot of time on almost daily, in terms of, what are we doing? Have we created the right risk variables to make sure that we’re not only monitoring but accelerating our evaluation on that?
And then I would say terrorism anywhere in the world is something that we think about as a potential issue for our businesses, and what could be the implications there. So I think risk issues have heightened. I believe that this acceleration of large language models and the ability to actually accelerate perhaps some of the things that were done on cyber could create another set of risks. So it’s a very complicated world, the systemic impact of that. How does that impact an organization? And while we’re assessing all that for AIG, we have to pivot real-time to be able to advise clients on what they should be thinking about. And so it’s been a very active start to the year to make certain that we are driving relevance with all of our stakeholders. It’s required more discussion and collaboration with regulators. Investors are very interested in the acceleration of AI, more so than I’ve ever seen in any time in my tenure at AIG, a single topic that continues to come up in so many of the different discussions we’re having. And then also, what are the implications of that in the future? If you look at the [capital expenditure] that the large hyperscalers and tech companies are investing in this particular calendar year, it is unbelievable. And then you should expect that acceleration to continue to happen, and companies are going to need to be able to operate in a world that’s changing in a very rapid way.
What have you’ve learned about leadership over the course of your career?
Leadership and management are two different skills, and leadership is really trying to set out a clear vision. We call it leading from the front. You have to have your executives actually out driving outcomes so that leadership [is worthy of being] emulated. I think you have to encourage debate. You can paralyze yourself with too much information. You can make decisions without enough information, but I think having an organization that does make very good decisions comes through repetition and also making sure that the decisions that are being made, every one is maybe not deliberated forever, but we constantly want to try to make quality decisions, communication is very important. And quite frankly, I think having a great team is the most important thing in leadership. There’s just no way that an individual can drive outcomes without having a very strong leadership team, and one that’s, I call it committed with a capital C, because a compliant culture will not drive outcomes like AIG has achieved. You have to have a committed culture. And the committed culture needs to believe, and if you’re going to believe in the journey, you need to outline it with transparency. These are professionals, they know there’s the good, bad and the ugly. I mean, in terms of how we are going to have to drive outcomes, and then they get inspired.
I’ve been very fortunate to lead an incredibly talented group of people that have made a difference, not only at AIG, all our stakeholders have made a massive difference in the industry, and that’s really the difference.