Since the beginning of 2023, artificial intelligence (AI) tool ChatGPT has dominated the national dialogue. People in almost every industry have been using it, brokers included.
“I first signed up for the free version to give it a test run and see what it was all about,” says Corey Schatz, president of The Paladin Group, an insurance brokerage in Iowa that specializes in trucking. “At first, I thought it was kind of like Google. And then I dove a little bit farther into it. I was blown away by the technology, and it got us thinking: how can we apply the technology in our own world, in our industry.”
Since the success of ChatGPT, other tech companies have released similar applications, such as Google’s Bard and Meta’s LLaMA.
Across the insurance industry, generative AI is used to better collect, organize and analyze data.
Brokers can use large language models to review contracts and policies, and employers can apply the technology to improve benefits communication.
Since the success of ChatGPT, other tech companies have released similar applications, such as Google’s Bard and Meta’s LLaMA. What has followed is a new competitive corporate landscape where companies in many industries have started to incorporate chatbots into their workflows and products. According to a 2023 study by Accenture, more than 70% of companies now prioritize AI over other digital investments.
Now that ChatGPT has been open to the public for nearly a year, how have brokers and others in the industry been applying the technology? And what new risks do these tools pose to clients who are doing their own experimenting?
What Is Generative AI?
First released in November 2022 by OpenAI, ChatGPT (GPT standing for generative pretrained transformer) combined a large language model (LLM) with supervised learning AI functionality. According to TechCrunch, “Large language models are, generally speaking, tens of gigabytes in size and trained on enormous amounts of text data, sometimes at the petabyte scale. They’re also among the biggest models in terms of parameter count, where a ‘parameter’
refers to a value the model can change independently as it learns. Parameters are the parts of the model learned from historical training data and essentially define the skill of the model on a problem, such as generating text.”
According to TechTarget, the supervised learning functionality refers to the fact that the “computer algorithm is trained on input data that has been labeled for a particular output. The model is trained until it can detect the underlying patterns and relationships between the input data and the output labels, enabling it to yield accurate labeling results when presented with never-before-seen data.”
The result was a conversational chatbot that would answer dynamic questions from users, using information it pulled from the internet. Once launched, it became the first online service to reach one million users within five days. A little perspective: according to Statista, Twitter (recently renamed X), took two years and Netflix three and a half years to reach the same number of customers.
Since then, OpenAI has released an updated iteration of its AI chatbot. Named GPT-4, the latest chatbot by OpenAI contains updated facts and information and holds enhanced problem-solving comprehension when it comes to math problems or analyzing literature.
Various companies within the insurance value chain have started to invest in and produce products using this new technology in an effort to add value to their services. One of them is Coalition, which offers active cyber and related coverage using tech-based risk monitoring and response. Last April, Coalition announced the launch of its new product, CoalitionAI Broker Copilot. The service allows brokers access to a generative AI chatbot, which is equipped for answering questions about cybersecurity coverage and best practices. Using its own documentation and data to inform its users, the large language model can be used to help brokers and businesses protect themselves against cyber exposure.
“Coalition has embraced AI because it will empower us to assist our broker partners and customers in real time,” says Tiago Henriques, the company’s head of research. “CoalitionAI allows them near-immediate access to the information and resources they need to make informed decisions about cyber risk faster and with a higher degree of accuracy, giving everyone that works with and at Coalition a superhuman edge.”
CoalitionAI Broker Copilot uses LLM similar to ChatGPT, though it caters its output to how the company covers a specific risk. “We train an LLM on our own documents because, for certain topics, we want the answers to be Coalition-specific. One example of this would be if you ask ChatGPT how to fix a remote desktop protocol (RDP) it will give you generic industry suggestions,” Henriques says. “RDP is a technology that allows system administrators to access employees’ computers and control the system remotely and can be a major security concern if it falls into threat actors’ hands. This is one such technology that would be a Coalition contingency to fix. If you use our LLM, it will produce the specific set of instructions that we ask our policyholders to follow.”
Coalition’s new chatbot is meant not just to automate simple-to-answer questions but also to improve the time and resources of their human team’s capabilities. “Nearly 90% of the questions in our customer support queue are essentially the same, albeit with variances in phrasing,” Henriques says. “By having AI respond to these repeatable, non-critical questions, we can quickly remove them from the queue and optimize our resources better. This means that the human team members can focus on answering questions that the AI doesn’t have good answers for or providing services to customers with specific needs, such as those experiencing a cyber incident.”
Henriques says the feedback on the tool so far has been great. “Our Broker Copilot users, both brokers and cyber insurance policyholders, love being able to ask questions about coverage, security contingencies, and other topics that used to require email communication,” he says. “Now they can get an immediate, almost instantaneous response.”
While Coalition’s tool uses its already existing data to better serve brokers and policyholders, others in the industry are using generative AI to better collect, organize and analyze data. The Paladin Group is working on a new product called UnderwriteGPT. “It’s an AI learning machine,” Schatz says. “It will be able to assist us with such things as client design loss control recommendations specific to clients and the specific industry they are in.”
According to Schatz, Paladin sees this machine-learning tool as a vehicle for cultivating and processing data. “Right now, we’re looking at it more as an upfront underwriting tool and an assessment tool,” he says. “UnderwriteGPT leverages intake data. Specifically, I can select a line of business and account details, and that context prompts the insurance-specific UnderwriteGPT model. What this creates is amazing. I finally have the ability to scale loss control for lower-revenue accounts. Everybody needs loss control, and this democratizes it to all sizes of accounts.”
Similar to Coalition’s generative AI chatbot tool, Schatz says internal reception for The Paladin Group’s UnderwriteGPT has been exciting. “The early feedback on the new experience is positive. We now have an automated web and mobile interface. Our intake creates a full digital record of the customer. This provides great time savings in and of itself. Our second major deliverable is in a test environment right now and is a conversational application programming interface (API) that connects UnderwriteGPT to our new experience,” he says.
By implementing generative AI to survey and collect data, insurance industry stakeholders are building out their own data sets to improve customer experiences and internal workflows. They are increasing their ability to rely on internal data to drive business instead of relying so heavily on third-party data. “Instead of insurance carriers and brokers using third-party data to simply process business, we can use data to create unique recommendations for accounts on risk that help them better understand their exposure, manage risk proactively, and preempt losses,” Schatz says. “This is a game changer for the industry. Generative AI also allows me to find more opportunities to help the business run better. For example, if accounts become safer, it makes it easier to get loads, which means more top-line revenue for my customers. That is the holy grail of customer value, so it is a win-win for everyone.”
Further defining the potential for generative AI in insurance, Jeff Gurtcheff, vice president of Enterprise Comp Services, wrote in the white paper “Generative AI Is Rapidly Transforming the Claims Landscape”: “[G]enerative AI can be trained to analyze a large number of prior claims and identify patterns and trends that may be useful in estimating the outcome of current claims. Using machine learning algorithms, generative AI has the ability to identify similarities between prior and current claims, taking into account various factors such as the nature of the claim, the severity of the injury, and the geographic location where the incident occurred.”
While global third-party administrator Sedgwick is not yet using generative AI to predict emerging trends, that may be coming in the future. Sedgwick launched its OpenAI GPT-4 infused tool Sidekick in April. The tool, according to the company, is designed to help assist claims professionals and automate routine tasks. “Designed for internal use, Sidekick will allow Sedgwick colleagues to explore the impact of generative artificial intelligence performance and natural language processing on day-to-day tasks,” says Leah Cooper, managing director of global consumer technology at Sedgwick. According to the company, Sedgwick is first integrating Sidekick’s AI capabilities in existing platforms to promote claims document summarization, data classification, and analysis. Initial examples include scanning PDF documents to produce automated content summaries and adding the highlights to the appropriate claim file, as well as uncovering key data to help complete tasks and meaningfully impact claims. The company notes that future iterations of the application may be able to produce entire claim summaries, identify risk factors on individual claims and programs, explore emerging data trends, and more.
While Cooper notes that Sedgwick sees clients’ growing appetites for maximizing efficiency through automation, gaining their trust on these new technologies is no easy task. Cooper identifies two key elements to successfully integrating this new technology with clients. “One, data transparency and availability,” she says. “We must be able to powerfully demonstrate which aspects of automation are highly reliable and working well for a program and which need to be tweaked further. We cannot and should not expect savvy clients to simply take our word for it. And two, true partnership. The implementation of automation always works best when the client is heavily engaged in the experimentation process.”
While much of the industry’s use of generative AI has so far been largely internal, technology-focused brokerage Newfront has built a product for policyholders. Earlier this year, the company released its AI Benefits Assistant, a customizable chatbot for any client that can be integrated directly into an employee’s existing Slack channel and can answer questions they have regarding their benefits and coverages. According to Newfront, this new application allows employees who are disengaged or nervous to talk with HR about their employee benefits a direct and easy-to-access channel to learn and discover the coverages they are offered by their employers. Newfront claims its product saves a company’s human resources department four weeks of work annually. “One of the common pain points we hear from people leaders we work with is how to handle employee HR questions,” says Gordon Wintrob, Newfront’s co-founder and chief technology officer. “Eighty-five percent of employees say they don’t understand the benefits programs and plans a company is offering. We’ve built an AI-powered chatbot that lives in Slack to answer these questions 24/7. The chatbot is trained on the employer’s HR handbook, employee benefits guide, and any other resources and knows how to answer these questions and escalate to the HR team as needed.”
So how does Newfront train a client’s personal AI benefits assistant once it’s provided with a company’s HR handbook? “The training for the Newfront benefits assistant actually leans on something called ‘embeddings,’ which is the process of breaking a document up into relevant pieces and mapping that complex text into a series of numbers that can be rapidly queried,” Wintrob says. “We then use contextual prompts to pass relevant text to the large language model. With this technology, we deliver a secure, isolated chatbot for each of our clients, and that chatbot knows how to access the right HR information for any question.”
According to Wintrob, this new technology holds plenty of other use cases for future health and benefits needs,
especially in the realm of wellness. “One interesting example is extending the AI benefits assistant to directly message employees, helping them with well-being questions and boosting employee engagement,” Wintrob says. “We’ve also shipped several features to help our internal teams process messy documents, like claims information or carrier billing details, and extract important details. This helps us serve our clients faster and with a higher accuracy rate than legacy approaches that rely on pen and paper.”
A common first question about generative AI is will it take my job.
Indeed, according to a May report from executive coaching firm Challenger, Gray, and Christmas, AI contributed to nearly 4,000 job losses across most industries, roughly 5% of all jobs lost that month in the United States.
Simultaneously, however, AI has also contributed to job growth. For example, large banks have been posting job openings since the beginning of the year for AI-related positions. JPMorgan was searching to fill more than 3,600 AI-related jobs.
According to Marcus Daley, co-founder of AI-focused insurtech NeuralMetrics, people worried about AI replacing their occupation should be asking themselves how this technology can be used to enhance their role at a company, not replace it. “How can I take those features and go solve a problem rather than taking it and just using it to reproduce what I’m already doing,” Daley says. “What can I do that’s never been done? And how can I then do that as a way for me to make a living and be better, more unique and faster, and different than everybody else in the marketplace?”
The emergence and integration of this new technology has also brought with it new companies that wish to be a part of the insurance ecosystem. Monitaur is an AI governance software company that enables insurance companies to manage the entire life cycle of AI models, including development, deployment, monitoring and compliance, while keeping up with changing regulations and policies, including NAIC principles and NIST standards. According to Anthony Habayeb, co-founder and CEO of Monitaur, the insurance industry, due to how much data it collects, is a perfect place for this technology to be used. “It has so much opportunity to make our lives better,” Habayeb says. “Companies across insurance are using AI to improve products and customer experiences, and strong governance is fundamental to insurance. We want to help this industry. The more time we spend here, we’re really proud to be a part of this ecosystem.”
With the emergence of this new technology, brokers still have the most important asset that can’t be imitated—client relationships. No matter how effective or big LLMs can grow, they still won’t be able to relate to and connect with people. “AI will make more time for, but not replace, the desire many have for real human relationships,” Habayeb says. “As an employer, I need cyber, errors and omissions, general liability, and other key coverages for my business. Even as a founder of a software company in the AI space, I want to have somebody, an expert, listen to me and talk to me. The greatest competition to this new technology is building really great human-specific relationships.”