Introduction
At Versich, we work with insurers who are rethinking how data, analytics and reporting come together to support better decisions. Generative AI has moved from an experimental technology to a serious business consideration for our industry, and we believe insurers who understand both its promise and its limits will be the ones who benefit most in the years ahead.
Insurance has always been a data intensive business, built on decades of underwriting files, claims histories, policy wordings and regulatory filings. What has changed recently is our ability to make sense of all that unstructured information quickly, using models that can read, summarize and draft in natural language. This is a meaningful shift, and it touches nearly every function inside an insurance carrier, from underwriting and claims to customer service, finance and compliance.
In this blog, we walk through where generative AI is already changing insurance operations, the challenges we believe every insurer needs to plan for, and how we approach these projects with our clients. Our goal is to give you a grounded, practical view rather than a hype driven one, so you can make informed decisions about where to invest first and how to manage the risks that come with this technology.
What We Mean by Generative AI in Insurance
When we talk about generative AI in insurance, we are referring to large language models and related tools that can draft text, summarize documents, generate code, answer questions in natural language and, increasingly, reason over structured and unstructured data together. These systems are trained on very large volumes of text and can produce new content in response to a prompt, rather than simply classifying or scoring existing data.
This is different from the predictive machine learning models insurers have used for years in pricing and fraud scoring. Those models are typically built to output a number, such as a risk score or a probability of loss. Generative AI adds a layer on top of that world. It can explain a pricing decision in plain language, draft a claims summary from adjuster notes, or turn a raw dataset into a first draft report that a human can review and finalize.
We find this combination, prediction plus generation, is where the real value sits for our clients. A predictive model can tell an underwriter that a submission carries elevated risk. A generative model can then explain why, in language that a broker or customer can understand, and suggest which additional documents might be needed to complete the assessment. Used together, these two types of AI create a more complete picture than either one alone.
It is also worth being clear about what generative AI is not. It is not a replacement for actuarial judgment, and it does not remove the need for skilled underwriters and adjusters. We see it as a powerful assistant that handles repetitive drafting and summarization work, freeing up experienced staff to focus on the decisions that genuinely require their expertise.
Where We See the Biggest Opportunities
Across the engagements we have run, a few use cases consistently deliver measurable value early on, and we encourage our clients to start with these before expanding into more ambitious projects.
- Underwriting support: our teams have seen generative AI cut the time it takes to summarize submission packages, extract key data points from broker documents, and flag missing information before a file reaches an underwriter's desk.
- Claims triage and drafting: AI can pull key facts from claim documents, adjuster notes and even photos to draft an initial claim summary, which the adjuster then reviews, edits and finalizes rather than writing from scratch.
- Customer facing communication: policy explanations, renewal letters and FAQ responses can be personalized at a scale that was not practical before, using the customer's own policy details and plain language rather than dense legal wording.
- Internal knowledge access: employees can ask natural language questions against policy wordings, underwriting guidelines and compliance manuals instead of searching manually through PDFs and shared drives.
- Reporting and analytics narratives: we use generative AI to turn dashboard data into a written executive summary, which pairs well with the Power BI reporting work we do for our clients, since a chart is more useful when it comes with a short, clear explanation of what changed and why.
- Code and workflow automation: our engineering teams use generative AI to draft data transformation scripts and documentation, which speeds up the back end work that supports customer facing reporting tools.
We have found that insurers get the most value when they treat these as assistive tools that sit alongside experienced underwriters, adjusters and analysts, rather than as replacements for their judgment. The insurers who see the strongest early results are usually the ones who pick a narrow, well defined use case, measure the results carefully, and expand only once the first pilot has proven itself.
The Challenges We Help Our Clients Navigate
Every opportunity above comes with a matching set of risks, and we believe addressing them early is what separates a successful rollout from a stalled one. We spend as much time with our clients on these challenges as we do on the opportunities themselves, because a program that ignores them rarely survives contact with a live regulatory review or a customer complaint.
Data privacy is often the first concern our clients raise, and rightly so. Insurance data includes health information, financial details, driving records and other sensitive personal records, so any generative AI deployment needs strict controls on what data the model can access, how long outputs are retained, and who is allowed to review them. We help our clients build data access policies that follow the principle of least privilege, so a model only sees the information it genuinely needs for a given task.
Regulatory compliance is closely related and, in our experience, one of the most underestimated challenges. Insurance regulation varies by state and country, and regulators around the world are actively developing guidance on AI use in underwriting and claims decisions, including rules around explainability and non discrimination. We recommend involving legal and compliance colleagues from the earliest design conversations, not after a pilot is already built, since retrofitting compliance controls onto a finished system is far more costly than designing them in from the start.
Model accuracy and hallucination risk also matter a great deal in our industry. A generative model that confidently produces an incorrect policy interpretation, an inaccurate coverage summary, or a fabricated citation to a regulation can create real liability for an insurer. We build human review checkpoints into every workflow where AI output touches a customer or a regulatory filing, and we track accuracy metrics over time so we can catch drift before it becomes a problem.
Bias and fairness deserve particular attention in insurance more than in many other industries. Historical claims and underwriting data can reflect patterns tied to geography, demographics or past business practices that we would not want a model to repeat or amplify. We test model outputs across customer segments before we allow any AI assisted decision to go live, and we build in ongoing monitoring so that fairness testing is not a one time exercise but a continuous discipline.
Integration with legacy systems is another practical challenge we help our clients work through. Many insurers still run core policy administration and claims systems that were built decades ago, and connecting a modern generative AI layer to that environment takes careful planning around data extraction, security and system performance.
Finally, change management is frequently underestimated. Teams need practical training, clear guidelines on when to trust AI output and when to escalate to a human colleague, and a feedback loop so the model and the surrounding process improve over time. We treat this as a core part of our delivery, not an afterthought, because the best technology in the world will not deliver value if the people using it do not trust it or understand its limits.
A Summary of Opportunities and Challenges
The table below brings together the themes we discuss most often with our clients, alongside our own recommendations for how to approach each one.
| Opportunity / Challenge | What It Means | Our Take |
|---|---|---|
| Automated underwriting | Generative AI drafts risk summaries and pulls relevant data points from unstructured documents | We see faster quote turnaround as one of the clearest near term wins |
| Claims processing | AI reads claim forms, medical notes and images to accelerate first notice of loss | We recommend starting with high volume, low complexity claims |
| Personalized customer communication | AI drafts policy explanations and renewal notices tailored to each customer | We find this improves clarity without adding headcount |
| Fraud detection | Pattern recognition across claims history flags anomalies for review | We treat this as a support tool for investigators, not a replacement |
| Data privacy and governance | Generative models can inadvertently expose sensitive customer data | We build governance guardrails before we scale any pilot |
| Regulatory compliance | Insurance is a heavily regulated industry with jurisdiction specific rules | We involve compliance teams from day one of any AI initiative |
| Model bias and fairness | Training data can encode historical bias in pricing or claims decisions | We test outputs across customer segments before deployment |
| Talent and change management | Teams need new skills to work alongside AI tools | We pair every rollout with structured training and change support |
How We Approach Generative AI Projects
Our approach starts with a focused pilot on a single, well scoped use case rather than a broad rollout across every department at once. We find this lets our clients see measurable results quickly while keeping risk contained, and it gives everyone involved a concrete example to point to when discussing what a broader rollout might look like.
From there, we build the governance layer, including data access rules, audit trails and human review steps, before we expand the pilot to additional teams. We also make sure the reporting and analytics side of the business, where our Power BI work often sits, is connected to the same data governance framework so insights stay consistent across the organization and do not drift apart from each other over time.
We typically structure our engagements in three phases. The first phase is discovery, where we work with our client's teams to identify the use case with the clearest business case and the lowest regulatory complexity. The second phase is a controlled pilot, run with a small group of users and a tight feedback loop, so we can adjust the workflow quickly based on what we learn. The third phase is scaled rollout, where we expand access, formalize training materials, and hand over ongoing monitoring to our client's internal teams with our support in the background.
Throughout every engagement, we keep three questions in front of our clients. Does this use case have a clear owner. Does it have a measurable outcome that we can track over time. Does it have a human checkpoint before it affects a customer or a regulatory filing. When the answer to all three is yes, we move forward with confidence, and when it is no, we treat that as a signal to slow down and address the gap first.
Measuring Success
We encourage our clients to define success metrics before a project begins, not after. Depending on the use case, this might include the time saved per underwriting file, the reduction in claims cycle time, the accuracy rate of AI generated summaries when checked against a human review, or the improvement in customer satisfaction scores tied to clearer communication.
We also track adoption alongside performance. A tool that produces excellent results but that staff do not trust or use consistently will not deliver the return on investment our clients are looking for. We build simple dashboards, often using the same Power BI capabilities we bring to other reporting engagements, so leadership can see both the efficiency gains and the adoption trends in one place.
Looking Ahead
We expect generative AI to become a standard part of the insurance technology stack over the next few years, much as predictive analytics did over the last decade. The insurers who invest now in data quality, governance and staff training will be best positioned to adopt new capabilities as they mature, rather than scrambling to catch up once the technology becomes table stakes across the industry.
We also expect regulatory frameworks to continue evolving, and we plan to keep our clients informed as new guidance is published across the jurisdictions where they operate. Our view is that insurers who build flexible, well governed AI programs today will adapt more easily to whatever comes next, whether that means new disclosure requirements, new fairness testing standards, or new expectations around explainability.
Conclusion
Generative AI offers our industry a genuine opportunity to work faster, communicate more clearly with customers and free up skilled professionals for the judgment calls that matter most. We have seen firsthand how a well scoped pilot can reduce the time spent on repetitive drafting and summarization work within a matter of weeks, giving underwriters and adjusters more time to focus on complex cases that genuinely need their expertise.
At the same time, generative AI introduces real questions around privacy, fairness, accuracy and regulatory compliance that we believe deserve careful, deliberate planning rather than a rush to deployment. The insurers who succeed with this technology over the long run will be the ones who pair genuine ambition with equally genuine discipline around governance, testing and change management.
At Versich, we help insurers move through this journey step by step, starting with a focused pilot, building the governance and reporting foundation underneath it, and scaling only once results are proven. Our experience across underwriting, claims and reporting engagements means we can bring a practical, tested perspective to your generative AI plans rather than a purely theoretical one.
If you would like to talk through what generative AI could look like for your organization, whether that means a first pilot, a governance review, or a broader reporting and analytics strategy, we would be glad to hear from you and explore how our team can support your goals.You can reach our team through our Contact Us page.
