AI spans the divide between
technological and human capital
By Joseph S. Harrington, CPCU
“We treat AI like a team-mate, not a tool,” says Aman Gour, founder and CEO of FurtherAI, a leading provider of artificial intelligence (AI) applications to managing general agents (MGAs) and insurance program administrators.
In an interview on the Today’s Markets podcast, Gour explains that AI, like a human team member, is deployed and evaluated by the outcomes it produces, rather by inputs plugged into it by designers. You ask questions of AI and you get answers plus prompts to additional lines of inquiry.
(Note: Comments from the interview have been edited for context and brevity. Other comments are derived from a subsequent exchange with the author.)
AI has now become “the starting point for research” into a problem or situation, Gour says, and can dynamically explore new dimensions and new relationships, as if it were a supercharged librarian or researcher. When it comes to executing research, Gour adds, “AI is more consistent than humans and has unlimited memory.”
Playing on a phrase commonly used in discussions of AI—about having a “human in the loop”—Gour comments that “the right architecture is not having a human ‘in the loop,’ it’s having humans ‘on the loop.’” In other words, humans can—and should—guide AI applications from the outside, directing them on the users’ terms much as they would a staff member.
This contrasts with earlier generations of software, even sophisticated platforms, that required users to structure inputs and analysis to conform to the software’s features and limitations.
“AI does not replace underwriting judgment, governance, or capital,” Gour says, “but it makes all three more scalable.”
Underwriting capacity
Gour emphasizes that “MGAs are not buying underwriting software.” Rather, “they are buying underwriting capacity by AI agents.” In other words, AI dramatically expands an intermediary’s ability to identify new product opportunities.
That entails “a fundamental shift in insurance,” he adds, as MGA platforms go beyond testing and implementing pre-designed program features to dynamically identifying previously undetected risk relationships, as well as coverage gaps and vulnerabilities that can be corrected or exploited.
Gour says that with these capabilities, AI will allow MGAs to assume more underwriting risk on their own books, expanding and enhancing their connections with carriers.
“AI will accelerate what is already happening in the market,” he says. “The best-run MGAs will earn more underwriting authority and retain more risk as a way to demonstrate their confidence in their programs. In turn, carriers will delegate more authority in lines where their MGAs have an edge.
“AI will help MGAs act more like carriers, while carriers, in an economic sense, will act more like reinsurers,” Gour adds.
“The best-run MGAs will earn more underwriting authority and
retain more risk as a way to demonstrate their confidence in their programs.
In turn, carriers will delegate more authority in lines where their MGAs have an edge.”
—Aman Gour
Founder and CEO
FurtherAI
Audit support
Gour has seen real results among the MGAs FurtherAI serves.
Policy audits, once a manual task applied to only a sample of a book of business, can now be carried out on all policies almost instantaneously, allowing for comprehensive review for adherence to underwriting guidelines, regulatory requirements, and company policies concerning underwriting authority.
“Previously, a carrier could audit only a portion of the policies,” Gour says.” With AI, you are not limited by the need for human time. You can actually run checks across all policies.”
Beyond that, Gour finds that “an AI agent can ingest policies, triage them, run checks, check against underwriting guidelines, put data into ACORD format, and produce bind letters.” With these capabilities, Gour says he has seen the percentage of quotes resulting in policies issued increase from 15% to 45%.
According to Gour, some MGAs balk at embracing AI because they believe their own organization is not “data ready.” Aware of the “garbage in, garbage out” maxim, they believe their data needs to be “cleaned up” before implementing AI.
Data optimization is always a good idea, but Gour advises against allowing data quality issues to impede AI implementation. AI is itself an “accelerator” of data quality, he says, capable of detecting data errors and consistencies.
Three key considerations
For those preparing to engage with an AI developer, Gour emphasizes three considerations:
- Make sure you retain ownership of your enterprise data. “The customer’s data should not be used for training AI models,” Gour says. The vendor should be working for you; you should not be enabling the vendor.
- Make sure the vendor can fully explain the processes, results, and output of the AI, as you will have to when dealing with regulators and business partners.
- Make sure that “change management” within your organization, something that AI will necessarily require, is a core consideration, and not an afterthought, in your vendor’s approach to development and implementation.
Things are happening fast in the AI world, as attested to in a light-hearted exchange in Gour’s interview with moderator John Willemsen, an advisor to MGAs and program managers with extensive experience in the area during tenures with Selective, Munich RE, Arch, Accelerant, and other organizations.
During their conversation, Willemsen asked: “At some point will I be interviewing an AI agent?” Gour responded: “Are you sure I am a human?” (He is.)
The author
Joseph S. Harrington, CPCU, is an independent business writer specializing in property and casualty insurance coverages and operations. For 21 years, Joe was the communications director for the American Association of Insurance Services (AAIS). Prior to that, Joe worked in journalism and as a reporter and editor in financial services.
