The role of AI in the insurance industry is expanding: AI tools can capture documents, assess risk, process claims, organize data, and reduce fraud.
These applications of AI can help insurers improve efficiency and accuracy in assessing risk and managing claims. However, AI technology also poses a range of risks and challenges. Insurers must act carefully when implementing AI tools to mitigate, detect, and manage artifacts and biases.
AI can bring significant benefits to the insurance industry if issues such as algorithmic decision-making and data transparency are properly addressed.
AI Use Cases in the Insurance Industry
AI is being used throughout the insurance industry for tasks such as underwriting, risk quantification, and fraud detection. Here are some specific ways insurers are incorporating AI into their workflows:
Contact center streamlining
Chris Raimondo, EY Americas Insurance Technology Leader, worked with a personal lines insurer to integrate generative AI into its customer contact center. The AI tool automatically summarized call notes, improving the accuracy and standardization of record-keeping, and empowering agents with clear case histories to better support customers in the future, enabling them to provide more effective, personalized support.
Importing paper documents
Despite significant digitalization efforts, the insurance industry still processes a lot of paper documents.
“Most insurers still deal with huge volumes of paper applications, payments and claims documents, and much of the content and data on those documents is handwritten,” says Franklin Manchester, global insurance strategy advisor at SAS. Manually reviewing documents takes time and effort, but AI can perform analysis at scale much faster than humans can.
Risk prediction and quantification
AI can also help insurers better predict and quantify risk. Suhas Sethi, global business leader of the insurance practice at professional services firm Genpact, explained that AI can reduce claims costs for property and casualty insurers, which are rising due to inflation, supply chain issues and extreme weather.
“By leveraging high-quality data from the claims process and integrating AI, insurers can better predict the impact of natural disasters and enhance risk management and disaster modeling,” Sethi said.
This allows insurers to select risks and price premiums more accurately. In the event of a major claim, generative AI tools can help insurers respond to claimants faster and ultimately manage the aftermath of a catastrophe more effectively.
Muddu Sudhakar, co-founder and CEO of generative AI platform Aisera, said insurers are also using generative AI to improve risk assessments. By analyzing demographic data and creating simulations, generative AI can identify potential claims and assist coders in developing statistical models. This approach can save insurers time and money in the long run.
In-house Virtual Assistant
Kjell Carlsson, head of AI strategy at Domino Data Labs, said the most common production use case among Domino customers in the insurance industry is in-house virtual assistants. These AI-powered assistants provide employees with customized information about customers' insurance policies and provide better context when employees answer customer questions.
The internal aspect is key: employees are in a better position than customers to refute or question illusions and can provide valuable feedback at an early stage.
“This has been a particularly successful use case because it is low risk, delivers value quickly and plays to the strengths of LLM,” Carlson said.
Data Preparation
Gene Linetsky, CTO of business insurance service Embroker, has been exploring the potential of AI to improve extract, transform, and load (ETL) processes using large language models (LLMs).
“Traditionally, ETL across multiple sources is costly and time-consuming because it requires custom-coded schema transformations,” he says.
Recent advances in LLM promise to significantly improve the ETL process. LLM excels at semantic mapping, translating data fields between sets at a fraction of the cost and time required to configure ETL adapters. Linetsky believes this will help insurers create fully autonomous data pipelines to process vast and highly diverse data streams, extract actionable insights, and integrate those insights into marketing, underwriting, and customer experience workflows.
Usage-Based Insurance
Joel Pepera, director of telematics product development at transportation analytics platform Arity, previously played a key role in Geico's usage-based insurance work, which leverages AI and machine learning to develop insights into individual driving behavior from sensor data generated by smartphones and other IoT devices.
Insurers argue that these insights allow them to set premiums more accurately and fairly, focusing on a customer's specific risk profile rather than demographic factors, but the practice has also drawn backlash from drivers who are concerned about data privacy.
Automating assessments
Insurers are increasingly turning to AI to improve damage assessments, which are based on photos submitted by claimants. Nikos Vekiarides, founder and CEO of deepfake detection platform Attestiv, said AI can help assess damages and settle claims through a self-service process for policyholders.
As a result, insurers are able to resolve more claims and claims professionals can focus on the more complex scenarios that require their attention. AI can learn from millions of photos of resolved claims and gain decades' worth of experience much faster than human adjusters.
“Of course, it doesn't eliminate the need for adjusters, as checks and balances are often necessary to ensure AI models don't make errors,” Bekiarides says, “but this type of hybrid verification is much quicker than resolving every claim manually.”
Automating fraud detection
AI is also essential for managing insurance fraud, Bekiarides says. Traditionally, efforts to root out fraud have required significant human review and analysis. But most insurers don't have the resources to do so, as typical special investigation teams are relatively small. As a result, even though a significant amount of fraud is uncovered, many others go unnoticed.
“It makes a lot of sense for insurance companies to use AI to analyze photos, videos, documents and circumstances of claims to find anomalies for their investigative teams to investigate further,” Bekiarides said.
But he also stressed the importance of checks and balances: human input is needed to prevent false positives and to take proactive action, such as having calibrators or investigators to verify the results of AI analytics.
Improved claims process
AI can help reduce data fragmentation in the claims process, says Sarah Bratschun, a senior scientist at agricultural risk assessment platform Ceres Imaging. Her company is tackling this issue in agriculture, where data used to manage risk, such as government databases, sensor readings, farm reports and satellite imagery, is often fragmented and siloed.
AI can help consolidate and standardize this data to better account for differences in soil and crop health, including yield potential that is influenced by factors such as soil type, climate, and agricultural practices. Improved data enhances risk management, giving growers access to reliable credit and fair, prompt compensation for crop losses and failures. This helps growers make more informed decisions, prioritizing the long-term health and productivity of their land over short-term profits.
On the insurance side, reducing data fragmentation improves the entire process: AI can extrapolate plant- and field-level data before and after severe weather events to quickly and accurately identify, analyze, and visualize affected areas. It also reduces claim resolution turnaround times, improving customer experience.
Improved underwriting accuracy
AI can also enable insurance companies to create products and policies that are better suited to customers' needs.
In agriculture, for example, AI is playing a key role in modeling farm data and effectively classifying land to accurately define productive and non-productive factors, Bratschun says. Growers benefit from faster and more accurate claims decisions and, with a true understanding of risk, can make informed decisions about crop selection and resource allocation.
At the portfolio level, lenders can better understand and rebalance their risks, tailoring loan terms and pricing to the policyholder’s particular circumstances, while land investors can actively manage the risk of their holdings through diversification.
EY's Raimondo also worked with a wide variety of insurers that used generative AI to consolidate disparate unstructured data sources into a unified system to provide underwriters and service center resources. This AI implementation reduced time-consuming manual research and helped the team get faster, more comprehensive answers to underwriting and quoting questions.
AI Challenges in the Insurance Industry
Insurance companies face the same AI challenges as many other industries: trust, integrity, bias, illusions, etc. In particular, the insurance industry is struggling with how to handle algorithmic decision-making and data transparency in AI implementations.
Algorithmic decision making
AI algorithms can learn the decision-making patterns of human experts and automate the process of making similar choices in the future. But these models can suffer from illusions and biases, leading them to impose poor decisions on the experts who should be making the decisions.
For example, UnitedHealth, one of the largest health insurers in the United States, has been criticized for imposing algorithmic decision-making on doctors: The company faced lawsuits alleging that its algorithms systematically rejected elderly patients' long-term medical needs against the advice of doctors.
Insurers looking to scale algorithmic decision-making should not do so at the expense of subject matter experts. Algorithmic decision-making is faster than human decision-making, but it requires oversight, especially when it does not match the judgment of human experts.
Data Transparency
Insurers are constantly increasing the data base they can use to estimate risks and price products, but the opportunities to use that data must be balanced with the interests of their customers.
For example, The New York Times reported earlier this year that many new subscribers to GM's Smart Car Data Service found their insurance premiums had increased significantly after their mileage, braking and acceleration data was shared with hundreds of third parties, including LexisNexis, a global data broker that compiles risk reports for insurance agents.
After the backlash, GM backed down and said it would no longer share such personal data. But insurance companies, which raised premiums, have yet to make a similar promise. Building trust with customers requires transparency about what data is used to train algorithms. If companies allow this, customers can make better decisions, not just for insurance purposes, but also regarding road safety.
Providing transparency to end users and getting their consent about what data is collected and how it is used is key, says Arity's Pepera. Investing in the infrastructure needed to manage large datasets, train AI models, and maintain an oversight and governance framework to track performance over time is also key.
George Lawton is a London-based journalist who has written more than 3,000 articles over the past 30 years on his areas of interest, including computers, communications, knowledge management, business and health.
