The 5% of Internal GenAI Pilots That Succeed: What We’ve Learned from Insurance Claims Departments

Recently, MIT’s State of AI in Business report from its Project NANDA revealed a striking “GenAI Divide.” 95% of companies saw no return on their internally-developed generative AI tools.

This sobering statistic hits home for insurance executives: despite ambitious internal AI projects, measurable impact on claims operations remains elusive.

However, externally-partnered solutions perform better. 67% made it to deployment, according to the MIT report. 

I can personally attest to knowing of at least six carriers that have attempted to build what my company has built, accurate AI-generated claims letters.  

I’m sharing some reasons why, and what carriers can learn before investing too heavily in projects that may not pan out.

What Types of AI Tools Cross the GenAI Divide?

How are some companies crossing the GenAI Divide? The research suggests an answer: partnering with specialized vendors. In fact, enterprises that collaborate with external AI providers achieve successful outcomes at twice the rate of those building in-house.

In other words, not only do external tools make it to deployment significantly more often, those that do deliver on promised results or ROI at twice the rate.

Voltaire is one example of this approach in action. Instead of creating a claims-letter AI tool from scratch (an endeavor with a high failure rate), many insurers have chosen to partner with Voltaire.

We have even suggested that it would be possible to fund the development of an internal tool from the savings Voltaire produces.

As one MIT researcher noted to a reporter at Fortune, the AI startups seeing rapid success are those that laser-focus on a single problem and execute it really well. Voltaire fits this mold – it tackles only the claims letter process, but does so with deep expertise and co-development input from claims professionals.

How to Cross the Divide: The Winning Playbook

I can speak to the trends identified in the MIT report.

As many who have worked in large organizations have observed, “Preservation of dysfunction is a mission-critical priority.” Carriers are not alone in this, and it’s hard to change the course of large ships. Inertia is a major barrier for both internally- and externally-developed tools.

When some organizations try to develop internal AI tools, they are often mandated as part of efforts to do just about anything useful or impactful with AI. Projects can suffer from significant amounts of internal politics, restrictions on how the solution needs to fit into an existing ecosystem, and any number of roadblocks anyone who has spent time working at carriers knows about. This all creates layers of inertia that external vendors simply do not have to contend with as thoroughly.

Organizations that succeed in crossing the GenAI Divide approach AI procurement differently.

The most successful buyers drive adoption from the front lines, sourcing AI initiatives from team managers and individual contributors rather than central labs. In other words, the people who need it most, get the most say; it’s not all determined by committee or the AI “experts.”

Likewise, they act less like software customers and more like business service clients, holding vendors accountable for outcomes and demanding configurations aligned with their internal standards.

They look for tools that have a low setup burden, fast time-to-value, and provide immediate, visible benefits. They also prioritize vendors they can trust, who have a deep understanding of their workflows, and whose tools cause minimal disruption to current processes. These buyers understand that crossing the divide requires partnership and co-evolution with a vendor, not just a purchase. 

That’s exactly how we built the core components of Voltaire. We have been ceaselessly focused on the best way to solve the user’s problem, namely generating claim letters booth quickly and accurately. And we’ve iterated over and over based on user feedback and product performance to get where we are today.

A Case Study in Success: Voltaire’s Approach

MIT’s study observed that while flashy AI projects in marketing get the budgets, it’s the behind-the-scenes processes that often yield better ROI. In fact, organizations that crossed the GenAI Divide reported saving millions by automating internal tasks like document processing and customer support. Insurance claims letters fall into that category.

The GenAI Divide report highlights that successful vendors are those that aggressively solve for learning, memory, and workflow adaptation. Voltaire is a vendor-derived AI solution that directly addresses these issues. Voltaire is an AI claims letter tool designed and built by claims professionals. 

Voltaire tackles a specific, high-value, back-office task: accurately generating claim settlement, reservation of rights, and similar communications. Specifically, we eliminate sources of liability that legacy systems create in claims letters.

The traditional, templated process of creating these letters is a major source of “hidden costs” for carriers. Adjusters frequently resort to copying old letters to save time, a practice that is prone to errors and compliance issues.

Why?

It can take anywhere from 30 minutes to two hours to produce a single claim letter. Even if templates are up to date, the adjuster may not know which of dozens of exclusions or endorsements apply without reading dozens of policy documents.

This inefficiency inflates labor costs, increases litigation and regulatory risk, and contributes to claims leakage.

Voltaire’s AI solution transforms this process. It can draft an accurate and compliant claim letter in as little as 30 seconds. A mid-sized carrier that implemented Voltaire reported that experienced adjusters saved about two hours daily.  

The quality of the AI-generated first drafts please both adjusters and staff attorneys, making errors easier to spot and ensuring compliance. 

If you take a look at one of our case studies, the first-year ROI is over 200%. The case study demonstrates that a purpose-built, learning-capable vendor solution can move an organization to the “right side” of the GenAI Divide.

The Path to ROI: Focus on Impact, Focus on Back-Office

The “State of AI in Business 2025” report reveals a significant investment bias in GenAI. Executives tend to direct budgets toward visible, top-line functions like sales and marketing, which captured approximately 70% of the hypothetical AI budget allocation in a survey.

However, some of the most dramatic cost savings come from back-office automation, according to MIT.

Our results at Voltaire can confirm this. From a business case we put together with client-verified metrics, a mid-sized carrier might expect mid 6-figure to low 7-figure savings in about a year of usage. That’s net savings, not gross.

 Companies that get this formula right are already seeing measurable gains, from cost savings in back-office tasks to better customer outcomes.

The rest risk falling further behind in the GenAI Divide.

Voltaire’s story is just one example of crossing that divide. By zeroing in on claims letters, leveraging proven AI models, and delivering quick wins, it shows what a pragmatic GenAI investment can look like in practice. For forward-thinking executives, the path to value from AI may start with such targeted, high-ROI projects. And the time to act is now – start with a demo from Voltaire.