Digitizing Knowledge to Build Human-Centric Processes

We’re seeing applications of AI everywhere we turn—and rightfully so. Every day we’re having more and more natural conversations with ChatGPT and other generative engines. Clearly, we all sense that this is an unbelievable technological milestone that needs to be properly tapped into. The applications are nearly endless.

To me, one of the most interesting step-function changes we can bring with AI to different industries is the switch from deterministic technology models to human-centric ones. In a deterministic model—largely based on “if/then” decision trees—we see capabilities such as streamlined workflows, and value props like speed and removal of friction. These capabilities make execution smoother, but they cannot promise a better, higher quality end result systematically. With deterministic models, the hope is that if a person isn’t focusing on the workflow, they have more energy/time/mental space for “quality”—but again, there is no guarantee. In a human-centric model, we can build technology that doesn’t just improve process, but promises outcome. 

To better demonstrate this point, I’d like to take a look at our specific use case within the specialty insurance industry: an industry that is particularly language-based and dependent on back-and-forth inputs between wholesale brokers and retail brokers. More below on why this is so important.

Let me start by taking a step back.

In the world of Commercial Specialty Insurance, traditional wholesale models are broken. The purpose of wholesale insurance brokers is to offer product expertise and better placement of insurance coverage to retail brokers who have exhausted their own knowledge or network and need help. However, because sifting through insurance quotes and terms, gleaning insights, leveraging connections, and determining recommendations is incredibly time-consuming, traditional wholesalers are only incentivized (through commission structures) to invest their attention on large enterprise deals. 

Ultimately, the purpose of the wholesale insurance industry is to provide service and expertise—and yet, they can only promise that level of attention to a select few. When it comes to specialty insurance, the mid-market and small commercial segments are painfully underserved. They often receive no response to their inquiries or are given extremely complicated materials with no attached insights.

At Flow—a wholesale brokerage for specialty lines—we’re building an internal AI engine that enables our brokers to bring responsive, insightful service to every single submission, no matter the size. In other words: we’re building a human-centric model that can promise outcomes in the form of consistent service. 

From Data Mapping to Digitizing Knowledge

As I look at different technology players in the industry, I see two things over and over again. The first is the effort to disintermediate specialty insurance. In other words, the assumption that with the right technology, the wholesaler can be removed from the value chain. We’ve learned at Flow that this is not possible. Retail brokers that are looking for specialty coverage need an expert. They need the wholesale channel. Thinking we can solve the problem by simply removing the middleman is a complete misunderstanding of the industry and needs of the ecosystem. Moreover, it’s a bit defeatist. It assumes from the jump that meaningful service is not scalable. 

The second thing I see often is companies who approach AI implementation as an engineering-first problem. For example: “How do I best extract unstructured data and give it structure?” This is a reductive way of solving the actual problem, and leads to benefits like workflow optimization. In the world of insurance, it can look like taking an email with a bunch of PDFs and pre-populating data into a platform. Yes, this saves retailers and wholesalers time, but is it meaningful? Is it accurate? Can they actually do a better job? We don’t know.

Remember: our goal at the end of the day is to enable our Flow wholesale brokers to deliver consistently, with rich insights and responsiveness to every single submission in a scalable way. In order for AI to actually help, it needs to do more heavy lifting. Importing data is great. But what about knowing which carriers to send the quotes to based on understanding of risk appetite across different coverage lines and industries? Or writing up in-depth comparisons while considering the many trade-offs and subtle differences across multiple quotes? Or providing insights and expertise on new products and how they compare to more traditional ones? This is what dictates time input and quality from a wholesale broker. For AI to help, it needs a digital knowledge-base.

At Flow, we call this digitizing specialty insurance knowledge. For example, rather than simply feeding AI carrier brochures and forms to structure their risk appetite, we also track the decisions the carriers make. Sure, a brochure can say that a carrier has a healthy risk for healthcare coverage, but the underwriter might behave differently. How can we ensure that the AI accounts for that discrepancy? In a Quote Comparison, there are many parameters of coverage (cost, terms, exclusions, to name a few). Does AI know how to calculate the trade-offs against the goal of the coverage? Can it then communicate those deliberations and recommendations when preparing a comparison document? Can AI read the different visual formats of submissions, whether they be graphical check boxes, hand-written notes, or structured forms with data fields? 

In order to properly implement AI, we must take a business-first approach. In other words, we ask ourselves: what are all the explicit and implicit actions that need to happen to lead to a certain outcome, and train AI around those jobs to be done.

Incorporating Expertise Using Agentic AI  

This begs the question: how do we incorporate the right jobs to be done in a meaningful and impactful way that actually scales service? 

For this, we need Agentic AI. Agentic AI is a type of Artificial Intelligence that is trained as a specialist, and can perform tasks independently, make decisions, and learn from interactions to achieve goals within that domain. Agentic AI does not simply execute predetermined tasks; it adapts and evolves based on our broker feedback, learning from each interaction to become a more effective partner within a very well defined scope. At Flow, we are training a cohort of what we call Specialists, who learn to perform tasks within a limited scope of the wholesale brokering needs. 

For example, we have a Quote Insight (AI) Specialist, whose job is to draft rich insights for quotes delivered. An engineering-first approach would look something like: “when you receive a quote, please pull the following fields and populate the following email template.” A business-centric approach or human-centric process would ask: “what am I trying to accomplish? And what collection of actions and materials can give me the right answers here?” For a quote insight to get drafted in a way that provides true leverage for our brokers to deliver expertise at scale, we have to teach the AI Specialist to track the steps. A refined prompt might look something like this:

• You are a Quote Insight Specialist. Your task is to parse through quotes and pull out insights to help our client better understand the materials. In order to accomplish this, you have to do the following:some text
  • Review the terms in the quote letter that we received from the market.
  • Once you’ve found the terms, please review the policy forms that you have access to.
  • Review any standalone endorsement documents and override what you need to within the new terms. 
  • If you have any similar submissions from previous engagements, please surface any additional materials that have been requested in the past to help expedite the process for our end-client.
  • Please define the terms and conditions to the retail broker

Because AI knows English very very well, because we’ve been committing a lot of effort to a digital knowledge-base, and because our AI Specialists are trained by our expert wholesalers, we’re able to perform complicated actions that solve for the following business problem: generate insightful communications to retail brokers. 

Over time, as our wholesale brokers verify AI-generated outputs and supply additional insights, the AI becomes more adept at managing dynamic requests, gradually evolving into effective collaborators. This back-and-forth interaction ensures that the AI continuously learns from human expertise, sharpening its abilities with every task.

Scaling Service and Outcomes

This process has been a true thrill in my career. How do we build technology that enables our internal wholesale brokers to earn the partnership of retail brokers. How do we build technology that promises consistency, responsiveness, reachability, insightfulness without removing the interpersonal connection between the wholesaler and retailer? How do we scale service? 

In an ecosystem that needs expertise and connection in equal measure, traditional wholesalers have shown to either increase hiring or increase development of self-serve technology. The former is cumbersome. The latter has no adoption. At Flow, we’ve figured out how to scale human connection by scaling expertise, making our wholesale brokers more available for more retailers, with consistent white glove service.

PS

In a follow up article, I will share how an AI Specialist is born, and how we maintain the balance between our brokers and our AI to ensure that it is never at the cost of trust with our clients. Stay tuned!

‍