3 critical steps in designing a POC that will succeed

Generative AI (GenAI) is already transforming the way insurers operate. While many companies are eager to adopt the technology, it’s crucial to have a clear plan in place to maximize its business impact.

In an August 20 webinar, Keith Raymond, a principal analyst at research and advisory firm Celent, spoke with Dan Gremmell, chief data officer at Zinnia, and Ali Wilson, a senior data analyst at Zinnia, about what makes for a successful GenAI adoption and what insurance carriers need to consider before integrating GenAI solutions.

Identify the business problem GenAI can solve

One key to successfully deploying GenAI is understanding why you’re using it in the first place — knowing where GenAI can deliver the highest value. 

“The first thing that you want to do is scope out what problem you’re actually trying to solve and what type of solution would be most beneficial to you, and is it really an AI-based solution?” Gremmell said. “Once you’ve identified that problem, then you could start identifying what features you’re looking for from a solution.” 

One area where Gremmell said Zinnia has seen success in implementing AI is in processing contact center calls to reduce the time agents spend doing follow-up work after each call. 

“We’ve seen a 53% reduction in average time spent on after-call work,” Gremmell said. “We’ve seen a 43% reduction in [the] number of calls requiring after-call work, and we’ve seen a 16% reduction in overall call duration.”

As Wilson explained, the design of the AI tool is crucial to its efficiency and effectiveness: “We chose to package up an application that pushes the call center manager towards specific tasks and outcomes. However, we also have internal data products that dig deeper on the data part for strategic leaders of the call center that make actioning more convenient.”

According to Raymond, customer service is one of the biggest areas where insurance carriers are seeing clear success with GenAI solutions, and an industry survey conducted by Celent showed that 75% of the respondents were reporting moderate or significant usage of AI for customer service.

“You just gotta make sure you’re solving a business problem,” Raymond emphasized. “Make sure that you’re measuring efficiency and gain and return on investment.”

Ensure your data is high-quality

Successful GenAI implementation requires a steady pipeline of reliable, valuable data. “GenAI models require high-quality, diverse, and comprehensive data to make accurate responses and predictions,” Raymond said. “Insurance companies may not always have access to such data. Similarly, integrating generative AI models with existing insurance systems and scaling them can be challenging.”

In the Celent survey, insurers ranked “improved data quality” as the most important success factor to successfully using AI. “Insurers are recognizing that without high quality data, the effectiveness of GenAI is significantly compromised,” Raymond said. “The importance of clean, accurate, and well-integrated data cannot be overstated, as it underpins all aspects of successful GenAI deployment.”

According to Gremmell, it’s important to establish a solid data collection process even before implementing AI. “Having a mature data ecosystem with those built-in quality practices will help you ensure data is ready for usage…you really want to try to figure out the data sources that exist with your organization, centralize those into a single location like a data lake or a data warehouse, and then work to clean and normalize that data so it’s in a usable and useful format.”

Gremmell explained that Zinnia then uses a human-in-the-loop model for quality checks before and after putting new GenAI tools into production. 

“While quantitative measures are helpful to measure at scale, they often will miss nuance of things or they don’t always test for illogical responses. So human-in-the-loop testing is essential to ensure that you’re getting logical responses out of your AI system, and it’s accomplishing the goal that you want it to accomplish.”

Build in legal and compliance processes early on

In the insurance industry, it’s crucial to understand the regulatory landscape and legal guardrails that could impact your business. This is especially important when it comes to GenAI solutions, since, as Gremmell explained, “the regulatory landscape is really getting ever more complex as time goes on and AI becomes more prevalent.”

Gremmell cited the state of Colorado as an example: a 2023 law requires life insurance companies to take steps to prevent any discrimination or bias when using AI to evaluate customer data. 

“In Colorado, life insurers, if they’re implementing AI systems that are driving insurance-based outcomes and decisions or using sensitive data, those are regulated,” Gremmell said. “So you need to have documentation around the model you’re using and how you try to derisk it based on bias.”

This is one reason some organizations, especially larger ones, may be better off finding a vendor to partner with for AI solutions rather than building in-house tools themselves. Vendors that specialize in GenAI tools for insurance carriers have expertise and a thorough understanding of regulations, which can save carriers the time and effort needed to build internal AI solutions. 

According to Raymond, Celent’s survey found that 80% of “tier one” insurance companies said they were working with a solution provider or consultant that offers GenAI services, highlighting just how important trusted vendors have become in implementing this new technology across the industry.

“Most insurers undertaking GenAI initiatives are working with solution providers or consultants that offer GenAI or LLM services,” Raymond said. “I think everybody’s still in that learning phase.”