This article is written by Sasha Haco, CEO and Co-Founder. unitary
The insurance industry is under increasing pressure to modernize. Recent industry data shows that 63% of claims professionals cite inefficient systems as a major barrier to effectiveness. At the same time, 56% of industry leaders believe the biggest impact of AI over the next five years will be the reduction of manual labor and admin, and 55% of insurers have already invested in AI or machine learning for claims.
The opportunities for automation are clearly vast, with the potential to improve cost, speed, accuracy, and customer experience. However, according to MIT's State of AI in Business 2025 report, 95% of generative AI pilots fail. Despite companies pouring billions of dollars into developing artificial intelligence, the vast majority of pilots never make it into production. This challenge can be particularly acute for insurance, a sector built on assessing risk. The risk is that the hype around “AI agents” in particular is starting to outweigh what these tools can safely deliver in a regulated, process-driven environment.
AI agents are now ubiquitous and emerging as a breakthrough technology that can automate entire workflows simply by giving large language models access to tools and natural language instructions. This is an attractive idea. No code, no integration work required, and promises end-to-end automation. But while AI agents are powerful, they are also widely misunderstood. Unless business leaders and policymakers distinguish between places that truly do work and places that essentially don't, companies risk making big financial bets on tools that are not up to the job.
The core of an AI agent works by interpreting instructions, predicting the most logical next step based on patterns, and executing that decision through a software interface or API. This makes it suitable for exploratory, unstructured tasks such as research, data collection, and summarization, where the goal is to produce useful rather than completely consistent answers.
However, AI agents have unique characteristics. This means that AI agents are probabilistic, meaning they generate new predictions every time they act. And while that's a strength for creativity, it's a weakness for operational work. “Probably true” is not enough, as 88% of insurance customers say their biggest concern about AI-driven processes is losing human oversight, and 70% report distrust in AI decision-making. Accuracy, auditability, and consistency support regulatory compliance and consumer protection.
Insurance workflows are defined by trust. Claims processing, underwriting support, policy management, and reporting all involve a long series of deterministic steps where the required actions are fixed rather than interpretive. A single mistake can have serious consequences. Incorrectly updated policy amendments, incorrectly routed claims, or incorrect documentation attached to case files can lead to non-compliance, customer disputes, or costly remediation efforts. Even if the AI agent shows 95% accuracy at each step, small error rates will add up in long workflows. A 20-step process with 95% accuracy at each step has only a 36% chance of completing correctly from beginning to end.
The problem isn't that AI agents aren't “ready” or need more training; it's more fundamental than that. AI agents make probabilistic predictions rather than definitive guarantees. If a task has one correct output every time, traditional software is always a more reliable option.
This distinction between probabilistic tasks that require judgment and deterministic tasks that require consistency is critical for automation leaders. Most business processes are a combination of both. Some steps require reasoning that humans have historically had to perform. Others are simple and rule-based. This is why the conversation around AI agents needs to evolve. These tools have a role in insurance, but not as a universal automation engine.
The future of automation in insurance is hybrid. The strongest solutions combine deterministic software for steps that need to be executed perfectly with AI that generates moments that require interpretation and expert-level understanding. Logging into the system, finding the appropriate policy, downloading documents, and updating records are critical tasks. However, analyzing medical reports and interpreting whether a loss falls within insurance terms is much more complex and would benefit from AI’s ability to process and structure the information. With a hybrid approach, AI is used only when its inferences add value, and the software otherwise ensures accuracy and compliance, delivering the speed and efficiency insurers desire without sacrificing reliability.
For insurers embarking on an automation journey, the first step is not choosing technology, but understanding workflows. Break down each process into steps and identify which processes must follow fixed rules and can be automated, and which require interpretation and require human escalation. When evaluating vendors, ask two simple questions. Can you clearly separate these step types and how do you handle errors? The goal is not to eliminate human expertise, but rather to eliminate repetitive tasks so teams can focus on making important decisions.
When deploying AI in environments where accuracy is non-negotiable, customers and regulators naturally demand predictability, auditability, and transparency. As the UK takes the lead in AI and accelerates productivity gains, we need to ensure that the tools we support are fit for purpose. AI can unlock enormous economic value. However, it is not blind optimism, but only if applied accurately.
