AI Bots in Service Organisations, Part 4: Insurance
From Today's Chatbots to Tomorrow's Autonomous Agents
By Sandra De Zoysa, Founder & Managing Director, CEx Consulting
This closes a four-part series on AI bots reshaping service organisations. Part 1 laid out the framework, banking and finance followed in Part 2, telecommunications in Part 3, and here, insurance; an industry I engage with closely as a board director at a leading insurance organization.
Insurance is, in many ways, the most interesting test case in this series. It is a trust product, people buy insurance for peace of mind, and when a family faces a genuine loss, they want a human being who understands their situation. Yet insurance is also where some of the most dramatic bot-driven results in any service industry are showing up. Holding both truths at once is the leadership challenge.
The Bots Already Working in Insurance
Chatbots now routinely handle First Notice of Loss (FNOL), capturing claim details and routing them 24/7, without a customer waiting on hold after an accident or loss. They also answer policy questions and provide coverage recommendations. The US insurance chatbot market alone is growing at a 22.9% compound annual rate.
RPA bots dominate policy administration, data entry, document validation, premium reconciliation, the repetitive, rules-based work that has historically consumed insurance back offices. This remains the largest and most mature bot deployment in the industry by volume, even as attention shifts to more autonomous systems.
Web scrapers/crawl bots feed underwriting directly, pulling public records, property data and cyber-exposure signals so underwriters get a dynamic, real-time view of risk rather than a static application form. Social bots support both marketing and early fraud-signal detection, monitoring public claims-related activity.
Trading bots are less central to insurers' core operations but relevant to the investment portfolios insurers hold against their liabilities, asset-liability matching still leans on algorithmic execution.
The honest industry picture though, is sobering; fewer than half of insurers have deployed AI in even a single function at production scale. Most AI use in insurance today is still customer service chatbots and document summarisation, end-to-end automation in underwriting or claims remains the exception, not the rule.
The Bots on the Horizon
Autonomous agent swarms are where insurance's real transformation is happening, and faster than the industry-wide averages suggest. Lemonade's "AI Jim" holds the current record for a full claim settlement: two seconds. As of late 2025, 55% of Lemonade's claims were fully automated start to finish, with 96% of first notices of loss taken without human intervention. More broadly, McKinsey finds insurers deploying AI-driven claims automation cut processing costs by 30–50% while improving customer satisfaction scores by over 20 points. Underwriting is following, more slowly, as the data sources are messier and the stakes higher, but end-to-end AI-automated underwriting platforms are now live.
Empathic AI companions matter more in insurance than almost anywhere else in this series, precisely because claims moments are frequently the hardest days in a customer's life. The consensus among industry leaders is clear: agentic AI should function like a capable junior claims processor, handling the process, but with a human supervisor holding authority over the emotionally and financially significant decisions. Full autonomous determination on claims, without human oversight, is not where the industry is heading, nor should it be.
Decentralised, blockchain-based bots have genuine relevance here, particularly in parametric insurance, where smart contracts can trigger and pay out automatically when a defined event (a flight delay, a rainfall threshold) occurs, with no claim filed at all.
Physical service robots remain the furthest horizon for insurance, though IoT-connected sensors already feed predictive-maintenance bots that flag equipment failures before they become claims, a preview of a more automated, prevention-first model of insurance.
What This Means for Boards
The questions I'm raising at board level right now:
· Where exactly are we on the pilot-to-production curve, honestly, given that most of the industry is still stuck in pilots with minimal P&L impact?
· Have we defined, explicitly, which claims and underwriting decisions an agent may execute autonomously versus which require human sign-off, before a regulator or a customer, forces that definition on us?
· Are we building the empathy layer into automated claims journeys as deliberately as we're building the efficiency layer?
Insurance has the sharpest version of the dilemma running through this entire series: the same technology that can settle a claim in two seconds must never lose sight of the fact that, for the customer on the other end, this may be the worst day of their year. The winners will be the insurers who engineer for both.
