The History of AI & ML in Insurance: Part 1

For decades, insurance has been at the forefront of embracing new and innovative technologies — from actuarial science, mainframe computers, and generalized linear models (GLMs) in the 20th century, to cloud computing, telematics, and advanced machine learning (ML) models in the 21st century. As we now step into the AI era, we are confident that, just as it has in the past, the insurance industry will continue innovating. Consequently, we at Distributed Ventures believe it’s a fantastic time to invest in companies disrupting this $5T+ market.

In this piece, we aim to provide a brief history of AI and ML in insurance, highlighting key technological advancements and developments in the industry. The article will also serve as the first-part of a two-part series in which we will discuss the future of AI in insurance, focusing on deep learning, large language models (LLMs), and generative AI.

Artificial intelligence (AI) is an umbrella term for different strategies and techniques used to make machines more human-like.

Machine learning (ML) is the science of developing statistical models and algorithms that process large quantities of data in order to identify data patterns. ML is an application of AI.

Deep learning is a subset of ML that is based on artificial neural networks. The learning process is deep because the structure of artificial neural networks consists of multiple input, output, and hidden layers.

Large language models (LLMs) are very large deep learning models that are pre-trained on vast amounts of data. These models can extract meaning from a sequence of text and understand the relationships between words and phrases in it.

Generative AI is a subset of artificial intelligence that uses techniques (such as deep learning) to generate new content. For example, you can use generative AI to create images, text, or audio.

A History

Predictive modeling has been a key part of the insurance industry long before the emergence of modern machine learning. During the 19th century, actuaries (data scientists for insurance) relied on linear regression models and probability theory to calculate insurance premiums. As the 20th century unfolded and insurance expanded to various types such as health, property, and life insurance, the use of early linear regression models gave way to more sophisticated generalized linear models (GLMs) and decision trees. These advanced models incorporated multiple risk factors, enabling more accurate insurance premium calculations. Later, as computers became more widespread, access to these models increased, allowing firms to analyze large amounts of data in less time.

The 21st century saw a shift from relying solely on historical data for pricing decisions to utilizing real-time data as well. Smartphones and social media paved the way for usage-based insurance (UBI), in which insurance companies tapped into the continuous stream of real-time data from smartphone usage and social media to create products tailored to real-time behaviors. To cope with the computational demands associated with processing large quantities of real-time data, insurers capitalized on improved access to cloud computing platforms like AWS, Microsoft Azure, and Google Cloud Platform, coupled with advancements in ML such as Ensemble Learning — an approach to ML that combines predictions from multiple models.

All of this culminated in the Insurtech boom in the mid 2010s, when companies embraced cloud computing, ML, and robotic process automation (RPA) to disrupt the traditional insurance industry. It was during this period that Insurtech saw substantial investment from venture capitalists. According to CB Insights, the total global VC investment in Insurtech was estimated to be around $2.65 billion in 2015, and by 2020 it reached a record high of $7.1 billion.

Looking ahead, we’re confident that insurance incumbents will continue to pioneer new advancements in technology to further supercharge the growth of this $5T+ industry.

Stage 1.0. Computers, Software, and the Early Applications of AI

Before computers, insurers relied on manual processes for pricing. Actuaries and underwriters performed pricing calculations manually or used machines like slide rules, eventually transitioning to pocket calculators. Given the industry’s heavy reliance on data, it had been an early adopter of technology, with Travelers Insurance being one of the first companies to install an IBM mainframe computer in the 1960s. By the 1970s, other large insurance incumbents followed suit, using mini-computers for invoicing and accounting. Later, when personal computers became popular, insurers embraced them to automate processes and collect and analyze customer data.

Simultaneously, the insurance software space began to take shape in the 1970s and 1980s. Policy Management Systems Corporation (PMSC) was founded in 1974, marking the emergence of the first large-scale provider of software and information services in the industry. Focused primarily on brokers, Vertafore and Applied Systems were subsequently established. Today, Vertafore and Applied Systems are major players in the insurance brokerage software space with 11.17% and 14.75% market share respectively. On the carrier side, Guidewire and Duck Creek, which started around 2000, eventually became key players.

During the Dot Com boom, ML models — particularly decision trees and generalized linear models — were still in their infancy in insurance, with only a few pioneers exploring their potential. A notable example includes Progressive which launched the MyRate program, later renamed to Snapshot, marking one of the earliest implementations of telematics-based Usage-Based Insurance (UBI) in the US.

These early applications provided glimpses of the potential ML could have on insurance, foreshadowing its broader adoption in the coming years.

2.0. Telematics, the Shift to Real-time Data, and Usage Based Insurance (UBI)

In the past, risk assessment in insurance had relied on demographic and historical data, which provided only limited insights into individual behavior. However, a shift occurred in the 2010s when risk assessment expanded to include real-time customer-specific data. This gave rise to Usage-Based Insurance (UBI), initially within the realm of auto insurance. Leading the charge was Progressive’s 2008 Snapshot program, which motivated several other top-ten insurers to introduce similar programs by the end of 2012. As mobile adoption grew, smartphone-based data collection expanded the reach of telematics programs, growing the market at 30% per year in 2019 and 2020.

Number of policies sending telematics data to insurers

To process, train, and make predictions on large volumes of real-time data, ML models require substantial processing power. This need was addressed by the increased accessibility of cloud computing platforms like AWS, Microsoft Azure, and GCP. As of June 2021, according to Novarica, over 90% of insurance companies had embraced cloud solutions to power their core operations. These platforms offered the scalable and flexible computing resources necessary for powering UBI initiatives. Notable UBI programs that emerged during this period include:

• Allstate’s Drivewise (2011) allowed policyholders to install a device tracking their driving habits, providing discounts for safe drivers. In 2014, Allstate introduced the industry’s first mobile telematics app, Drivewise Mobile that participating drivers could use to earn discounts
• State Farm’s Drive Safe and Save (2011) measures the number of miles you drive to the times of day you take to the wheel as well as your vehicle’s acceleration, cornering, braking and average speed, providing up to 15% discounts for safe drivers
• Nationwide launched SmartRide (2011) and introduced the SmartRide mobile application in 2016 to help low-mileage drivers save on insurance premiums
• Metromile (2012) launched its personalized pay-per-mile auto insurance providing flexible insurance pricing based on mileage
• Liberty Mutual’s RightTrack (2017) tracks customers’ driving behavior and helps them save up to 30% on auto insurance premiums

Commercial fleets saw a boost in telematics adoption due to the 2015 ELD mandate, which required all commercial motor vehicles (CMVs) to use electronic logging devices (ELDs) to monitor driving behaviors such as speed and braking patterns. The mandate led to the widespread adoption of ELDs among commercial motor vehicle operators, a trend that continues today and has even expanded to Canada in 2023. The North American commercial vehicle telematics market is expected to grow 15.3% from 2022 to 2027, with key companies including Trimble Inc., Verizon Connect, Samsara, Zonar Systems Inc., and Geotab Inc.

On the ML front, there were significant developments in the 2010s, including the rise of ensemble learning, a technique that combines multiple models to produce more accurate and robust predictions. Ensemble learning became widely used in a variety of real-world applications in the 2010s, due to the development of new methods such as random forests and gradient boosting, as well as the availability of large datasets from the cloud. One key advantage of ensemble learning is its ability to outperform individual models on complex tasks like fraud detection, which is relevant to insurance. However, these performance benefits come with a trade-off of transparency. Ensemble models are complex, so it can be difficult to understand how the model arrived at a particular prediction, such as why a particular scenario is labeled as fraudulent or how a risk score for an insurance policy is computed.

3.0. Rise of Insurtech

As more and more insurers embraced technology and adopted cloud solutions, advanced ML models, and usage-based insurance (UBI), it culminated in the disruption of the traditional insurance industry. This marked the rise of the Insurtech era around 2013–2014, when a number of companies were formed at the intersection of insurance and technology.

These Insurtech startups often challenged the traditional insurance business model. Some offered new and specialized insurance products to compete with existing carriers, others attempted a direct-to-consumer insurance model that eliminated insurance brokers, while a few super-charged existing players like brokers. Notable examples of Insurtech companies that launched during this time include:

• Lemonade introduced a direct-to-consumer approach to streamline the underwriting and claims processes. This eliminated the need for insurance brokers and offered a fast, transparent, and digital insurance experience
• Hippo Insurance was one of the first homeowners insurance companies to offer a fully digital experience, with customers able to get a quote, purchase insurance, and file claims online. Hippo also integrated smart home technology into homeowners’ coverage like water leak detectors and security systems to proactively prevent potential issues
• Root Insurance improved the affordability, accessibility, and customer experience of auto insurance. Its customers could download an app to track their driving and receive a personalized quote. Root also offered telematics-based pricing that was fairer and more accurate than traditional insurance pricing. The company had a D2C approach with the goals of eliminating brokers
• Indio, one of our portfolio companies, was one of the first Insurtech startups to focus on commercial lines insurance, providing insurance brokers with a software platform to manage their businesses more efficiently. Indio was acquired by Applied Systems in 2020, demonstrating the willingness of insurance incumbents to bring innovative technology in-house.

Simultaneously, RPA (Robotic Process Automation) — the use of software robots or virtual assistants to automate repetitive and time-consuming tasks — started taking off in the insurance industry. Many of the RPA solutions used AI to improve efficiency and reduce operational costs by automating processes such as claims processing, policy administration, underwriting, and customer service. Some notable RPA players include UiPath, Automation Anywhere, SS&C Blue Prism, and Pegasystems. The COVID-19 pandemic accelerated their adoption in insurance as RPA helped automate routine and repetitive tasks while also enabling remote work and improving customer service. This coupled with rising labor costs led to an increase in demand for RPA solutions in insurance and is expected to drive further growth in the coming years. According to research conducted by Allied Market Research, the global RPA in insurance market was valued at $98.6 million in 2021, and is projected to reach $1.2 billion by 2031, growing at a CAGR of 28.3%.

Another type of insurance that benefited from technology and cloud computing is parametric insurance — in which a policyholder and insurer agree on a triggering event and the policyholder gets paid when the event occurs. For example, an insurer pays $20,000 to a policyholder if their zip code experiences an earthquake above 5.0 magnitude (the event), regardless of whether they incurred damages. Parametric insurance companies analyze a substantial amount of event-related data (floods, weather patterns, flight delays, etc.) and have therefore benefited from increased availability of granular historical data and cloud computing solutions. Some notable examples include: Floodflash, a parametric flood insurance company that was founded in 2016; Insur8, a typhoon warning insurance product that launched in 2017; Raincoat, a parametric weather insurance company founded in 2017; and travel delay insurances such as Blink Parametric and Wakam where policyholders receive payouts for a parametric trigger, such as a missed flight.

Looking Ahead

The history of AI in insurance has been nothing short of transformative. It began with basic linear regression models and probability theory in the 19th century, progressed to generalized linear models (GLMs) and decision trees in the 20th century, and saw the rise of cloud computing, advanced ML models, UBI, and RPA in the 21st century.

In the coming weeks, we will be publishing a piece that delves into how our firm, envisions the future of AI and ML in insurance, focusing on the applications of deep learning, large language models (LLMs), as well as generative AI.

If you are a company, investor, or simply someone interested in the intersection of AI and Insurance, please don’t hesitate to reach out to us at adam@distributedvc.com, alex@distributedvc.com, and shivani@distributedvc.com.

We look forward to connecting with you!