Insurance is a language-intensive industry, and now that technologists have really figured out language and unstructured content, the way insurance brokerage companies run will surely change. At Gaya, we're at the forefront of this revolution, and we're excited to share our perspective on the AI use cases we're seeing in the insurance brokerage sector, some adoption challenges, and how we are going about it.

Emerging AI Use Cases in Insurance

1. Quoting and re-quoting: That's where it all starts and that's where Gaya is leading the pack. We have developed what our users call “autofill on steroids” for insurance quoting. Agents are using Gaya to parse unstructured information coming from declaration pages, ACORD forms, comparative raters, carrier rating portals, policy summaries generated by Agency Management Systems, digital intake forms with embedded conditional logic, fillable PDFs, manually written PDFs, and (I kid you not), drafts on napkins. Our AI is tuned to accurately parse information from all these diverse sources. Once parsed, agents head to carrier portals and use Gaya to “superpaste” or autofill these portals in a few clicks. Information we autofill include customer details, household member details, vehicle details, property details, coverage details, etc. Our Beta users have reported productivity increases of 50% to 80% in their quoting activities, both for new business and remarketing. Interestingly, we've noticed a shift in agents' behavior: they're now more inclined to get exact prices rather than relying on comparative raters' estimates. This precision not only benefits the agents but has also caught the attention of carriers, who see potential in lowering their quote-to-bind ratios through more accurate initial quotes.

2. Proposal formations: Post-quoting, agents typically spend hours synthesizing and comparing quotes over Excel spreadsheets and drafting emails to clients with their recommendations. We are closely in touch with several agencies already uploading carrier generated quote sheets to GPT-4 or using services like Brightside AI or Parsure to generate these proposals. These AI systems can analyze and compare quotes, producing clear, concise summaries of the options. While valuable, we see this as just one piece of the puzzle. At Gaya, we're looking at ways to integrate this capability seamlessly into the broader quoting and client communication process, making it a natural extension of an agent's workflow rather than a standalone tool.

3. Policy checking for Commercial Lines: Companies like Powerbroker AI and Qumis are making strides in AI-powered policy review for commercial lines. We've also observed agencies creating custom GPTs for this purpose. However, the feedback we're getting suggests a need for more than just chatting with a PDF. Agents are asking for bespoke, verticalized SaaS solutions that can handle the complexities of commercial policies, from reviewing endorsements and exclusions to ensuring compliance with industry standards and client-specific needs.

4. Retrieving and synthesizing loss runs: Companies like Tailwind Insurance Systems are trying to improve on the huge value that Loss Run Pro is already providing by integrating Gen AI to automate carrier correspondence, loss run analysis, and next steps creation. As for parsing loss runs, foundational AI models haven't fully cracked it yet, as we are dealing with unstructured and context-specific data that looks structured. Loss Scan, however, is leading the pack with its more than 4,000 templates that seem to be effectively powering the extractions. We believe that foundational models will eventually figure it out however as they integrate human-like reasoning capabilities (i.e. thinking in steps aka chain of thoughts) in their outputs.

5. Generative Intake Forms: Horizontal intake forms today are great. Anyone can spin out  Cognito Forms, Jotform, Formstack, or Gravity Forms in a few minutes. Insurance-specific ones (RiskAdvisor, SALT, Quotamation) are also great and now have data enrichment capabilities thanks to Fenris, among others. The next generation of intake forms will be more dynamic. We see a lot of clients having 70% of the information required to quote sitting across different systems (emails, AMS, rater, pulled from carriers’ websites, county property records, etc.). They're asking for a way to extract what they already have and dynamically ask for the missing pieces required by the carriers they will most probably end up placing business with. There is no vendor out there today providing that flexibility. Meanwhile, Gaya’s early customers are getting creative (they can’t help it). To supercharge their new business quoting flow, agents at Nova Insurance have been using Gaya to super-copy dec pages pulled through Canopy Connect and then super-paste structured information with Gaya on their RiskAdvisor where they collect the missing pieces. Rob Bowen at Patriotic Insurance has been similarly super-copying with Gaya the Hawksoft generated policy overview of his existing clients and then super-pasting it again on RiskAdvisor where his producers are collecting the missing pieces. Similarly SALT is looking as we speak into reading from Gaya’s clipboard through a webhook all the extracted info to pre-fill some of the questions they ask for on the form. These are great workarounds today, we do see however a more cohesive and robust workflow in the future.

6. Voice agents: Several companies have taken a stab at this, with Gail, Sonant and Sonar being insurance-specific examples. Many other general voice AI systems can be tuned or scripted to handle insurance-related queries. One of our summer interns has fine-tuned an insurance-specific voice agent on top of Bland AI over a weekend and the results were highly similar to the rest of the crowd (but at $6/hour, 3x cheaper). However, while this technology is significantly better than a traditional answering machine, it's still far from feeling truly human. The current limitations are clear: most of these systems can't listen and speak simultaneously, leading to unnatural pauses in conversation. They often struggle with complex queries or unique situations that require nuanced understanding. At the end of the day, while LLMs can write great essays today, they still cannot spell words. Still, the potential is enormous. Imagine a voice agent that could handle routine queries 24/7, schedule appointments, provide basic policy information, and even start the claims process. While we're not there yet, the rapid advancements in AI suggest that truly human-like voice agents may not be far off. Actually, as we are about to release this post, OpenAI released their Realtime API where developers can now build fast speech-to-speech experiences into their applications. Things are moving very fast.

7. System of records with Generative AI at the core: Imagine an AgencyZoom 2.0 powered by generative AI. It wouldn't just manage pre-set communication pipelines; it would proactively create and manage interactions based on the context and history of each account. But let's take it a step further: if this system could also synthesize voice call transcripts, we'd be looking at a truly Guided AI Agent. This agent could have proactive outreach based on nuanced context from calls, understanding client needs at a deeper level based on their existing policy documents. It could analyze patterns in customer interactions to optimize communication timing and content, personalizing each interaction. This isn't just about automating correspondence like CRMs do today; it's about creating an intelligent system that helps insurance agents provide better service and identify new opportunities. The potential for integration across various Agency Management Systems could revolutionize how agents manage their book of business and help them better care for and round their accounts.

8. Web-Agents: 50% of an agency’s time is spent on servicing. And servicing is a series of activities that involve clicking on various carriers pages to achieve a certain transaction (mortgagee change, COI, etc.). Many progressive agents reverted to RPA (Repetitive / Robotic Process Automation) and built bots which act like a pre-recorded macro. Some like Mike Foosco are thrilled with thousands of hours saved. Others like Rob Martinek gave up as carriers kept changing their flows and after more than $20,000 in the hole, decided to wait for others to build so he could lease. Quandri and Adapt have been doing a great job and agencies swear by the time savings they generate for them. But for the most-part, we are now entering the era of “bots with brains” and we do imagine a world where every human insurance agent will be working with an AI-CSR that is doing repetitive work and guiding them through their workflows. Needless to say, we are building towards that vision. We are getting closer but still early in the journey. Stay tuned as we are dedicating an entire post to sharing more about this very topic.

AI Adoption Challenges

While the potential of AI in insurance brokerage is exciting, it's not without its challenges. As we've engaged with agents and brokers across the industry, four primary concerns consistently emerge:

Liability concerns: Who will be liable if AI makes a mistake? Will the AI vendor take over the E&O risk? This echoes the dilemma of liability in self-driving car accidents. However, for founders, this is mostly noise. Consider how sophisticated machines have been used in radiology for decades. We don't blame the machine for a missed diagnosis of a fractured bone. Physicians will always be in charge. The same principle applies here.

Adapting to probabilistic AI: This is going to be the era where humans are dealing with a non-deterministic system, not 100% working correctly 100% of the time. Empathy is needed at the core for folks to think of AI as a probabilistic contributor, similar to a human. Sure, it doesn't get sick, but it can be more generous on given days. This mindset won't be for everyone.

Data privacy and model training concerns: This is particularly challenging when approaching larger brokers. They worry about client data being used for training purposes - a valid concern with clear mitigation strategies focusing on integrity, transparency, and privacy. Interestingly, this fear is almost non-existent in the SMB sector, where users are now accustomed to sharing trade secrets while chatting with ChatGPT (often not even the Plus version).

Data Migration Issues: For AI services to be fruitful, good data needs to come in. For vendors adopting the integration route, the journey might be too long and drain the founders and the capital they raised. For vendors adopting the “replace and benefit from our AI+ services,” good luck fighting the whales and their data “immigration policies”.

User-level “intimidation”: Some users, particularly CSRs, feel as if their data-entry roles are being threatened by AI. However, we haven’t spoken to a single agency owner who wants to install Gaya in their agency to let go of some data-entry specialists/VAs. In an industry with a high turnover rate and steep learning curve, agency owners are just looking to repurpose their resources to sell more and service better. However, this initial user-resistance can slow the adoption but not really impede it. 

Looking Ahead

As we navigate this AI revolution in the insurance industry, it's clear that the next couple of years will be decisive in shaping the future landscape of insurance technology. We're at a pivotal moment, reminiscent of the era when relational databases were first introduced to the industry.

The adoption of relational databases led to the emergence of two major winners in the insurance tech space: Vertafore and Applied. These companies became the 800-pound gorillas of the industry, creating significant inertia in brokers' workflows. Their systems became deeply embedded in daily operations, making them indispensable to many agencies.

Today, we stand at a similar crossroads with Large Language Models (LLMs) and Transformers. These technologies hold a similar power to revolutionize workflows and create a new type of inertia in the industry. The potential for innovation is immense, and we're likely to see the emergence of new industry leaders in the coming years.

At Gaya, we have a unique vision for the future of insurance technology. We believe that a new category is emerging - an intelligence layer that operates on top of existing systems like Vertafore, Applied, other insurtechs, and carrier systems. This layer will enable every insurance agent and broker to work with an AI CSR for quoting and servicing workflows, dramatically enhancing productivity and service quality. We invite you to join us on this exciting journey of transformation and innovation in the insurance industry.