Underwriters or data scientists: who will help Brian Duperreault make AIG great again?

Brian Duperreault is focused on refreshing AIG's underwriting talent to execute a turnaround, but should he turn his attention to data science and machine learning instead?

In May 2017, the 97-year-old insurance behemoth AIG named Brian Duperreault as its new CEO in hopes of completing a comeback from its near collapse in 2008. [1] Since that announcement, Duperreault’s main focus has been on transforming AIG’s underwriting culture, hiring more than a dozen senior executives and 125 senior underwriters, many of whom formerly worked at AIG, to bring smart risk selection back to AIG. [2] However, in today’s machine learning (“ML”) age, is bringing in fresh underwriters the right step forward in transforming underwriting culture?

Insurance is no longer the same game it used to be. Data has always been an integral part of underwriting; however, insurance companies now have access to more granular and reliable data than they have ever had before, from social media to Internet-of-Things sensors. [3]  Using ML, algorithms can utilize the proliferation of such data to find correlations between characteristics of insureds and resulting losses, leading to continuously improving risk selection. Despite this, most insurers still process only 10 to 15 percent of the data they can access. [4] With the predictive power of ML being key to underwriting today’s evolving risks, Duperreault may be taking a step in the wrong direction by hiring traditional underwriters rather than data scientists to assist his turnaround.

However, Duperreault did bring with him a medium-term bet on ML—Hamilton USA, now rebranded to Blackboard. Hamilton USA, formerly part of Hamilton Insurance Group, which Duperreault founded and oversaw as CEO prior to his AIG appointment, has been focused on integrating data science and ML analytics into commercial insurance underwriting for many years. [5] Specifically, they are focused on providing commercial insurance to small- and medium-sized businesses, a target market which historically had extremely high customer acquisition costs. [6] Alongside Duperreault’s appointment as CEO, AIG acquired Hamilton USA for $110M and furthered their investment in Attune, a data-focused joint-venture launched in 2016 by Hamilton USA, AIG, and Two Sigma Insurance Quantified (TSIQ), a data science provider. [7] Thus, through Blackboard and Attune, Duperreault is admitting that ML has a role in the future of insurance underwriting; however, he is not putting it front and center as a way to turnaround AIG in the near term. Instead of augmenting AIG’s traditional human-centric underwriting workflow with ML, he is attempting to reinvent the underwriting “wheel,” developing a tech-enabled underwriting platform from the ground up. 

AIG is also relying upon TSIQ to experiment with a few shorter-term underwriting initiatives across AIG, outside of Attune, to get exposure to new types of data sets and predictive capabilities. [8] This shorter-term external partnership model is not unique within the commercial insurance industry. For example, AXA XL, a competitor of AIG, has the dedicated team XL Accelerate to curate insurance technology startups, often ML-focused, and to introduce them to business unit leaders to similarly pilot new data sources and analytics offerings. [9]

Thus, Duperreault’s technology strategy seems to be fixated on two parallel paths: utilizing an external partnership with TSIQ to explore incremental, nearer-term underwriting improvements while developing a separate data-driven underwriting platform within Blackboard, in isolation from the rest of AIG’s business. Perhaps most important is that Duperreault has made it clear that his short-term fix for AIG’s underwriting woes is not ML, but rather recruiting better underwriting talent quickly, as seen below. [10]

 

 

 

 

 

 

 

However, I believe to successfully turnaround AIG over the medium term, Brian Duppereault has to focus on building ML capabilities, which requires quickly hiring data scientists who can work alongside underwriters in the short term. Integrating data scientists into their organization is not only necessary to begin the cultural change that AIG requires to modernize their traditional insurance operation at scale but also to begin developing AIG’s necessary new risk selection “secret sauce.” Today, insurers’ “secret sauce” to select risks consists of rigid underwriting rules and the expertise of individual underwriters; however, with ML, this will be replaced with proprietary data models and algorithms that continuously improve as they get fed more data. By relying heavily on an outside partner in TSIQ, AIG may be compromising their ability to fully own the IP, or “secret sauce,” underlying their future underwriting platform. Meanwhile, by concentrating their technology development in TSIQ and a separate subsidiary, Blackboard, they are missing the opportunity to expose traditional insurance operators to how data and ML may be beneficial and to get underwriter buy-in for future technological advancements.

Brian Duperreault has the difficult challenge of overcoming short-term financial challenges while also preparing AIG for longer-term sustainability. Will his choice to double down on traditional underwriters to turnaround AIG in the short term make it difficult to integrate ML and data science later down the road? Additionally, at what point should he start preparing his underwriters for a potential change in their role due to ML – before the technology is ready or after? (795 words)

 

[1]  Barlyn, Suzanne. “CEO Duperreault Bets on Underwriting Culture, Greenberg Ways in Restoring AIG to Greatness.” Insurance Journal, September 11, 2018, https://www.insurancejournal.com/news/national/2018/09/11/500630.htm, accessed November 13, 2018.

[2] Ibid.

[3] Malhotra, Ravi and Sharma, Swati. “Machine Learning in Insurance.” Accenture, 2018.

[4] Ibid.

[5] Kandell, Jonathan. “Can This Man Return AIG to Glory?” Institutional Investor, September 29, 2017, https://www.institutionalinvestor.com/article/b15130qkmvywsz/can-this-man-return-aig-to-glory, accessed November 13, 2018.

[6] Ibid.

[7] Barlyn, Suzanne. “AIG Picks Hamilton’s Duperreault as CEO.” Insurance Journal, May 15, 2017, https://www.insurancejournal.com/news/national/2017/05/15/451078.htm,  accessed November 13, 2018.

[8] Hamilton Insurance Group. “AIG, Hamilton Insurance Group and Two Sigma Insurance Quantified Announce Expansion of Partnership.” Press release, May 15, 2017, http://hamiltongroup.com/insurance-group/press-release/aig-hamilton-insurance-group-and-two-sigma-insurance-quantified-announce-expansion-of-partnership/?cache=r33Zb, accessed November 13, 2018.

[9] “XL Catlin’s Accelerate Partners with UK Artificial Intelligence Startup Cytora.” Insurance Journal, October 02, 2017, https://www.insurancejournal.com/news/international/2017/10/02/466119.htm, accessed November 13, 2018.

[10] Davis, Gavin and Casapietra, Gianluca. “AIG turnaround: The four Rs.” The Insurance Insider, October 30, 2018, https://www.insuranceinsider.com/articles/122686/aig-turnaround-the-four-rs, accessed November 13, 2018.

 

Additional sources:

Buluswar, Murli and Reeves, Martin. “How AIG Moved Toward Evidence-Based Decision Making.” Harvard Business Review, October 01, 2014, https://hbr.org/2014/10/how-aig-moved-toward-evidence-based-decision-making, accessed November 13, 2018.

Lenihan, Rob. “Insurers to tap technology for future success.” Business Insurance, June 05, 2017, https://www.businessinsurance.com/article/20170605/NEWS06/912313741/Insurers-to-tap-technology-for-future-success-AIG-Attune-Hamilton-Duperreault, accessed November 13, 2018.

Scism, Leslie. “Insurance: Where Humans Still Rule Over Machines.” The Wall Street Journal, May 23, 2017, https://www.wsj.com/articles/insurance-a-place-where-humans-not-machines-rule-1495549740, accessed November 13, 2018.

 

Banner image:

Ralph, Oliver and Gray, Alistair. “AIG: The long struggle to repair its reputation.” Financial Times, June 17, 2018, https://www.ft.com/content/3b9f9384-634c-11e8-a39d-4df188287fff, accessed November 13, 2018.

Previous:

Vojd: filling the Void in Luxury Design with Additive Manufacturing

Next:

Leveraging Machine Learning at Spotify

Student comments on Underwriters or data scientists: who will help Brian Duperreault make AIG great again?

  1. @Rohanmal, thank you for sharing this article on AIG’s approach. I was intrigued by your second question, regarding how to prepare the underwriters. I imagine that credit-extending institutions are also facing similar questions about how to train their teams making credit decisions. A shift in process to a machine-learned model will require individuals with deep lending knowledge to help train and develop the model. But one of the limitations of machine learning, is that at a certain point you can’t ascertain why or how the model made it’s decision. My sense is that this will be a big obstacle for insurance and credit firms as they start deploying machine learning in their decision models. According to a Quartz article last month (link below), regulatory bodies are already starting to raise concerns over discrimination on the basis of gender and race. How do think this will affect AIG’s decision to invest in machine learning and AI?
    https://qz.com/1277305/ai-for-lending-decisions-us-bank-regulations-make-that-tough/

Leave a comment