If you've attended an insurance conference recently, you've likely heard two radically different conversations about AI.
On stage, brilliant presenters share countless examples of insurers doing genuinely impressive things, leaving you imagining all the potential AI use cases that could apply to your organization.
Off stage, however, the conversation is different. Insurers share their frustration with attempting to scale AI, blaming inaccurate or disconnected data, but several less obvious cultural and operational barriers also stand in the way. As a result, many insurers are left unable to achieve the outcomes promised by their AI ambitions.
With a clearer understanding and the right foundation, insurers can more fully embrace AI and realize the benefits it brings to operations and scale.
It all starts with demystifying what's holding things back.
AI resistance remains
AI success stories typically focus on the larger carriers, but often insurance is written by smaller carriers and MGAs. These players have one or two lines of business they know well. Their teams deeply understand their customers, market and the data. Their user journey, whether through intermediaries or direct sales, flows adequately. But, due to a variety of factors, most often budgetary, resistance remains amid this sector.
The AI that would genuinely help them is not simply used for predictable considerations where deterministic rule-based automation would be sufficient and cheaper. It's the AI that supports the person who comes back to the system every few months, such as the broker managing a renewal or the MGA operations lead making a mid-term adjustment. These practical use cases demonstrating a relatable transformed user experience can help reduce resistance and encourage smaller insurers and MGAs to deepen AI deployment.
Innovation outpaces understanding
AI has widened the gap between what technology companies promise and what many insurers are currently equipped to implement. While insurance executives are understanding more about the tech available to them, discussing fully autonomous AI with an insurer running on a Policy Administration System (PAS) built 30 years ago can be like handing someone the keys to a rocket ship when their experience is limited to a manual transmission car.
Insurers should seek solution providers who can demystify AI and help them understand what it can realistically accomplish. The goal should be to introduce new systems in intuitive steps that fit into existing workflows, rather than starting with the most advanced use cases.
Security and reputational risks loom large
As professionals who live and breathe risk management, insurers are well aware of AI's inherent risks. They worry about what will happen to their customer and underwriting data once it is loaded into an AI platform. How can they be sure it stays safe, and how do they know whether the data is clean enough to trust?
This is where choosing the right tools and keeping humans in the loop matter most. Any AI platform an insurer chooses should run in a protected environment backed by recognized data-security standards, such as ISO 27001:2022 certification. Additionally, insurers should assign humans to review the outputs and ensure they provide answers as expected. Any hallucinations may indicate a problem with the data or logic the AI engine is pulling from.
The foundation is not ready
Once insurers and MGAs get off the AI starting blocks, they sometimes layer in additional automated tools, adopting an AI agent for underwriting, then another for ratings, and another after that. But every AI agent performing an underwriting task is only as accurate as the data it has ingested, the product logic, appetite rules and underwriting guidelines it acts on. If that foundation is not current and consistent, those AI agents risk producing incorrect decisions at a greater speed and scale.
Additionally, adding AI layer by layer runs afoul of Amdahl's law, a well-respected computer science formula that the overall speed of a system will always be constrained by the parts that have not been improved. For most insurers, that "part" is the PAS, and when it lags, it erodes AI's potential benefits.
To see how, consider a risk ingested under one set of underwriting rules. An agent quotes it correctly and the policy is bound. Three months later, the customer comes back with a mid-term adjustment. By then, the ingestion layer has been updated with a new appetite and tighter criteria, but the PAS has not caught up, so the adjustment is quoted against a different set of rules than the original policy. At renewal, a third set of rules applies. The customer now has a policy that means something different at inception, at mid-term adjustment, and at renewal, all because the layers fell out of sync.
When the foundation is wrong, the answer, "our AI agent made that decision," will not satisfy anyone. But when the foundation is right, insurers can confidently say every decision their AI made was built on rules they controlled.
Build on rock, not sand
Overcoming AI scalability challenges is possible for insurers and MGAs of all sizes, but it requires more than simply adding AI tools to existing processes. Success depends on creating the operational discipline, governance and technological foundation that allows AI to deliver consistent results at scale.
The insurers that achieve the greatest value with AI will not necessarily be the early adopters; they'll be the ones that treat it as an acceleration layer on top of clean data, well-defined rules and modern, adaptable core systems. Nailing those fundamentals can help AI deliver faster decisions, greater efficiency and better customer experiences. Get them wrong? AI may simply amplify existing inconsistencies and inefficiencies.
In the end, successful AI adoption isn't a question of ambition or speed, it's a question of foundation. Build on sand, scale becomes much harder to sustain. Build on rock, AI becomes a catalyst for transformation.
