Buffett 2.0: How Hedge Funds Use AI to Crack the Insurance Code

Artificial intelligence is bringing transformative changes to the insurance industry across key areas like underwriting, claims, fraud detection, risk management, customer service, and more.

AI enables insurers to operate faster, more intelligently, and on a larger scale than ever before.

Some major ways AI is impacting insurance include:

• More accurate and personalized underwriting and pricing using predictive analytics. AI can isolate the variables that influence risk from massive datasets to support hyper-customized premiums aligned with specific policyholder risk profiles.?
• Faster claims processing through automation, computer vision, and natural language processing. AI speeds digitization, assessment, and resolution of claims to improve customer satisfaction.?
• Enhanced fraud detection leveraging AI pattern recognition across claims data, customer profiles, provider patterns, and other sources. This prevents costly fraudulent payouts.
• Holistic risk management powered by AI algorithms that uncover correlations, warn of emerging perils, model climate impacts, stress test strategies, and identify optimal mitigation approaches.
• Intelligent customer engagement and segmentation fueled by machine learning models that synthesize data for 360-degree customer profiles. This enables hyper-targeted products, pricing, marketing, and service.

While still early in adoption, innovative insurers are already using AI to transform legacy processes and leapfrog competitors. AI-driven insurance delivers faster, smarter, and more customized experiences while optimizing risks and costs. This is revolutionizing the competitive landscape across P&C, life, and health insurance segments.

At its core, AI allows insurers to shift from reactive, backward-looking statistical approaches to proactive, predictive data modeling.

For over a century, actuaries have relied on historical data and generalized linear models to assess risks and set premiums. But these methods have key limitations:

• Dependency on large historical datasets
• Difficulty modeling nonlinear relationships
• Static models that require periodic retraining
• Broad customer segmentation
• Lagging indicators of emerging risks

In contrast, machine learning and deep learning can synthesize wider, unstructured datasets covering customer profiles, behaviors, external risks, and more. Advanced algorithms model highly complex variable interactions and relationships missed by traditional models. AI models also update continuously by retraining on new data. And AI enables customer-level personalization versus broad segments.

Below is a comparison between traditional generalized linear models and modern AI techniques:

As shown, AI allows insurers to advance risk analytics across multiple dimensions for a strategic advantage. Below is an expanded comparison emphasizing some limitations of traditional actuarial analysis versus modern machine learning and data science approaches in insurance:

• Actuarial analysis typically relies on segmented risk pools versus individual-level modeling. AI enables hyper-personalization with models tailored to each customer's unique risk profile.?
• Actuarial models are static and backward-looking. AI models update continuously by retraining on new data, enabling more forward-looking risk analysis.
• Actuaries use lagging indicators to warn of emerging risks. AI pattern recognition can spot early signals of changing risk dynamics in diverse data.
• Actuarial datasets tend to be small and sparse for rare events. AI can generate synthetic risk scenarios through generative modelling for more robust statistical testing.
• Actuarial segmentation is coarse and reactive. AI enables proactive, extremely fine-grained micro-segmentation powered by deep learning feature extraction.

While actuaries still provide invaluable expertise, leading insurers are augmenting traditional actuarial science with modern AI to unlock richer insights across the insurance value chain. Combining actuarial knowledge with AI may enable the biggest performance leap for risk analytics and underwriting in decades.

But an over reliance on historical actuarial approaches in isolation, while complying with a rigid regulatory frameworks that does not reward (in fact, does not understand) innovation will make it difficult for insuretechs to keep pace with the rapid advances in tech that could give them and advantage.

This is even more difficult when they need to satisfy regulatory environments that favor "legacy methods", thus, artificially slowing down their growth.

Generative AI: A Game Changer for Risk Management

One of the most exciting emerging AI applications for insurance risk analytics is generative modeling. Generative algorithms can synthetically generate new, realistic data samples after learning patterns from real-world data.

For insurers, generative models open exciting new possibilities to enhance risk management:

• Augment limited data - Certain risks like natural disasters or unusually high jury awards ("nuclear verdicts") have little historical data. Generative AI models can simulate these events to create larger training datasets. This improves underwriting accuracy. And there is already a precedent for this approach in an insurance-like instrument: In the mid-2000s, I was the main quant developer part of a team of bankers that structured the first LSMFT ("Litigation Settlement Monetized Fee Trust"), where we used generative AI (not called generative AI then) to analyze an event yet to occur, from the legal and financial points of view. The $115 million transaction received an "A" investment grade rating from S&P by among other things, stress testing expected cash flows and loss levels beyond historical and simple multiple X scenarios.
• Fine-tune pricing segmentation - Generative algorithms can produce granular synthetic customer profiles. These can train AI models to advance hyper-personalized, usage-based insurance pricing versus relying on broad segments.
• Continuously retrain models - Generative AI allows insurers to continuously create fresh, realistic data to retrain machine learning algorithms. This ensures models stay up-to-date amid evolving risk dynamics.
• Sharpen loss reserving - Generative models can simulate claims scenarios tailored to insurer-specific claims patterns and exposure. This provides a superior statistical basis for loss reserving versus historical industry benchmarks.
• Model emerging risks - By assessing deviances between real and synthetically generated data, insurers can detect early signals of new risk exposures before they manifest at scale. This enhances dynamic risk monitoring.

Overall, generative modeling unlocks a paradigm shift for insurers and enables a proactive risk management instead of a reactive one. Scenario modeling is limited only by computational power, not constrained by finite historical data. As generative AI matures, leading insurers will embrace it to make risk management predictive, nuanced, and forward-looking as never before.

First movers leveraging generative AI for insurance risk modeling will gain a substantial competitive advantage. Those slow to adopt will lag in risk analytics capabilities.

Generative AI promises to be a true game changer for the insurance industry in managing risks and designing sustainable, innovative products.

Tradition Meets Innovation: Blending Actuarial "Black and White" Rigidity with Data Science "Many Shades of Colors" Agility

Historically, actuaries have dominated insurance risk modeling and underwriting, but data science teams are gaining influence as insurers embrace AI. This is creating natural friction between traditional actuarial culture and newer data-driven mindsets, specially in environments where one culture was already predominant.

Actuaries pride themselves on mathematical and theoretical rigor, statistical precision, and adherence to regulatory principles. Data scientists and quants tend to value exploration, flexibility, and openness to new techniques like machine learning. Actuaries optimize to avoid worst-case scenarios in a purely backwards looking sense. Data scientists accept some risk in pursuing predictive insights and establish risk-rewards tradeoffs.

These differences can lead to unhealthy turf wars within insurance firms. However, the most successful insurers will bridge their actuarial and data science capabilities. A collaborative, integrated approach is ideal.

In my view, actuaries should maintain leadership over core pricing, reserving, and capital modeling to leverage their risk management expertise. But Senior data scientists and quants should have free rein to innovate, enhancing predictive analytics, and deliver competitive advantages, not possible with actuarial science.

In a legacy industry and w/o much capital to grow, either you innovate or you die. Effective integration comes down to communication, mutual understanding, and true leadership.

Above all, insurers need senior executives in finance, insurance, risk management, computer science and actuarial analysis, who can synthesize the strengths of those disciplines in a forward looking approach.

I believe that with a patient leadership, insurers can reap the benefits of traditional actuarial science and modern data analytics working in harmony. The firms that can successfully bring their actuaries and data scientists together working under the tutelage of the most qualified professional(s) with relevant experience in some or all of the fields above will have an advantage in the age of AI.

But cultural obstacles require enlightened leadership with the know-how to visualize and understand the full potential. It also requires patience, and maybe capital beyond what is "normal" for a typical startup even at levels of an average Series B or beyond.

MGAs: The Trojan Horse for Hedge Fund, Family Offices and other challengers to Conquer Insurance.

A tech-driven managing general agent (MGA) can leverage ML and big data to capitalize on miss priced policies and cherry-pick the best risks, as indeed some of the smart capital out there is doing right now. An MGA is an intermediary that manages insurance programs and assumes underwriting authorities delegated by an insurer.

Using DNNs, the MGA can model complex variable interactions missed by traditional GLMs. The DNN continuously retrain on new data, ensuring risk assessments and pricing keep pace with evolving real-world conditions.

This enables the MGA to precisely segment risk levels and pinpoint optimal premiums for each client. The MGA can identify and capture safety-conscious clients with competitive rates - while avoiding poor risks prone to claims.

In effect, AI unlocks what I call a "slow motion arbitrage" whereby the MGA systematically undercuts rivals for good risks and avoids unfavorable risks, never taking the lion's share of the risk. This drives steady profitability over time as the MGA accumulates lower-risk policies in a disciplined, optimized approach.

The advanced analytics translate into a strategic edge, allowing the MGA to reshape industry dynamics through superior risk selection and pricing optimization. The MGA steadily accumulates quality risks without aggressively growing market share.

An actuarial-based insurer relying on static historical data and pricing models will struggle to match the ML-powered MGA's granular customer segmentation and continuous pricing optimization. Indeed, machine learning is far more than a tactical tool for the new MGAs coming to market. It is an asset that enables the MGA to disrupt the industry through superior risk analytics, disciplined underwriting, and sustained profitability.

InsurTechs, Hedge Funds, or Incumbents: Which Culture Will Dominate Insurance Disruption?

It is too early to tell, but there are some recent events that one can use to make inferences about the future.

Two Sigma, known for its data science and hedge fund trading strategies, is entering insurance technology by focusing first on commercial trucking insurance underwriting.

Through its new unit Two Sigma Insurance Quantified, effectively an MGA, the hedge fund aims to leverage its analytical capabilities to improve risk assessment and pricing in commercial trucking insurance. This represents a beachhead into the wider commercial P&C insurance market.?

Two Sigma IQ plans to organize trucking fleet data and enrich it with other sources to enable underwriters to make faster underwriting decisions, based on metadata in VIN numbers of trucks, private and public databases, etc. Rather than automating existing workflows, the priority is boosting underwriting insights through data modeling and AI.

With its hedge fund resources, financial quantitative analysis DNA, and proven capability to raise billions of USD in capital w/o the need of opaque VC firms in the middle, Two Sigma can take a patient, analytical approach to innovating in commercial trucking insurance underwriting. The firm's priorities align more with optimizing underwriting outcomes versus digitizing legacy processes.?

By concentrating first on trucking insurance, Two Sigma can refine its data-driven approach and predictive modeling capabilities in a focused segment before expanding into larger commercial P&C market. This initial focus demonstrates how Two Sigma is strategically leveraging its strengths in financial quantitative analytics to disrupt a niche insurance area first.

Two Sigma has over 20 years of experience in financial quantitive analysis and modeling to add to its new foray into insurance technology. With $60 billion in AUM, the hedge fund also has tremendous financial backing to support its insurance operations, like some of the pioneers of this area in the past: Lehman Brothers' Lehman Re, Goldman Sachs’s Arrow Re and Credit Suisse Re.

This deep war chest differentiates Two Sigma from many insurtech startups facing potential funding shakeouts for the rest of 2023 and maybe 2024. As some industry experts have noted, some insurtechs may run out of capital or might be acquired at deep discounts as the market consolidates. But Two Sigma's position is to take a long view in developing insurance underwriting capabilities.

Some compare Two Sigma's insurance tech ambitions to Warren Buffet's long-term investments in the insurance industry. With its data science expertise and patient capital, Two Sigma can strategically build out its insurance analytics platform and pursue acquisitions without relying on external funding.?

Given its financial prowess, Two Sigma is not just going to dip its toes in the insurtech sector like many others. The firm possesses the capacity to make a profound, long-term commitment necessary to revolutionize data-driven underwriting in a $6 trillion global industry. While some insurtechs may offer flashy yet shallow apps and/or overstate their technical abilities and expertise, Two Sigma brings substantial institutional weight and a tradition of analytical rigor cultivated over two decades.

If Two Sigma's initial venture into the insurance sector is just a beginning, it is evident that the dominant culture will be one of financial quantitative analysis and data science, surpassing the traditional and dominant actuarial mindset.

The dynamics in this sector are set to become highly engaging, evolving into a scenario where gains for a few players are counterbalanced by losses by many others.

#InsurTech #AI #HedgeFunds #Disruption #Analytics #Underwriting #quants #DataScience #ML #Algorithms #QuantitativeFinance Dan Butzbaugh Troy Barnes, CCXP Lenny Bellafiore Robert M. Forant Erran Carmel Natalia Díaz Rodríguez, PhD Arnold Aranez Ashim Sahu Timothy Lim Simeon Preston Kev?n H. Mason B Is Hiring Walter DE BROUWER Jo?o Abiul Menano