AI in (Re)Insurance Operations: Why Agentic AI Requires an Operating Model Change
Insights from a panel conversation between Elliott Bundy, Chief Marketing Officer at mea Platform, and Subhasis Mukherjee, Senior Principal at Datos Insights, at the Datos Insights Insurance Leaders Technology Forum in Boston.
Key Takeaways
- Most AI in insurance today is pilots and fragments. Wins exist in isolation but rarely connect end-to-end.
- Back-office overhead consumes roughly 12 to 14 points of combined ratio at a typical (re)insurer.
- Existing core systems organize the work; they do not execute it. Agents do.
- (Re)insurers should buy pre-trained, insurance-native AI rather than rebuild commoditized operational infrastructure.
- AI agents handle the repeatable; people handle the consequential. Junior underwriters reach judgment work earlier in their careers, not later.
Why is Most AI in Insurance Still Stuck at the Pilot Stage?
Most AI in insurance today shows up as pilots and fragments; point wins that do not connect to anything else. According to Elliott Bundy, CMO of mea Platform, OCR and workflow automation have existed in insurance for years, yet underwriters still process the same volume of email, the same PDFs, and the same handoffs between systems. The end-to-end work has not fundamentally changed.
The two consequences Elliott highlighted on the panel:
- Efficiency gains plateau, because nothing connects the workflow end-to-end.
- Bigger commercial wins, including gross written premium uplift and capacity expansion, never materialize.
How Much does Back-Office Overhead Actually Cost a (Re)Insurer?
Roughly 12 to 14 points of combined ratio. When mea Platform models the entirety of a (re)insurer’s operations (system cost, people’s time, and general overhead) that is the impact consumed simply keeping the back office running.
“If you are competing on back office, you are probably losing. The idea that one carrier’s back office is more special than another’s is rarely the right place to apply judgment, client service, or specialist expertise.” — Elliott Bundy, CMO, mea Platform
Elliott ’s view is that judgment, client service, and specialist expertise are where (re)insurers should be placing their capacity; not on commoditized back-office processing.
What Measurable Results has Agentic AI Produced in Real Underwriting Deployments?
mea Platform’s deployment at The Hartford has produced the following outcomes, cited by Elliott Bundy on the panel:
- Approximately 20 minutes for full-lifecycle underwriting on new business cases, from ingestion to decision-ready.
- A 30 to 40 percent increase in underwriter capacity.
- Humans remain in the loop, with underwriters focused on exceptions, flagged items, and judgment calls rather than repeatable processing.
The mechanism behind these numbers: mea Operations runs on a domain-specific language model (dsLM) and a proprietary insurance knowledge graph, trained on real insurance documents and workflows. That is what allows agents to classify data, populate fields in core systems from unstructured inputs, and execute end-to-end operations.
How is Insurance Ingestion Different from OCR?
OCR digitizes a PDF. Ingestion classifies the data and populates fields in core systems from the unstructured inputs underwriters and brokers receive. The distinction matters because OCR alone does not move the work forward; ingestion does.
Ingestion is where mea Platform began as a company and remains the most common starting point for (re)insurers adopting AI into operations. From ingestion, deployments typically extend into adjacent areas — claims, finance, and beyond.
Will Agentic AI Replace Junior Underwriters?
No — Elliott took the opposite view on the panel. The insurance industry has a long-standing talent challenge, and watching documents move between departments was never a strong onboarding experience for new entrants.
With agents handling the repeatable work, junior underwriters spend more time on broker service, judgment calls, and portfolio thinking; the consequential work that historically came later in the career. Agentic AI moves that exposure earlier, allowing junior underwriters to grow and learn at a much faster rate.
“AI agents handle the repeatable; people handle the consequential.” — Elliott Bundy, CMO, mea Platform
How does Agentic AI Work Alongside Legacy Core Systems?
Existing core systems organize the work; they do not execute it. Agents do. According to Elliott some of mea Platform’s leading-edge clients now treat the platform as either a core-system adjustment or, in certain cases, a reason to deprioritize further core modernization altogether.
The reasoning is structural. Many (re)insurers have invested heavily in core modernization in the last three to five years, but workflows remain fragmented and data scattered. Agentic AI sits across that landscape, making transactional decisions on top of the existing core.
The commercial position from mea Platform is direct: (re)insurers should buy something pre-trained on insurance knowledge, integrate it quickly, and let it make their existing systems work together faster, smarter, and better. Rebuilding commoditized operational infrastructure in-house is rarely where differentiation sits.
“Compete where your ‘special sauce’ is, not the back office.” — Elliott Bundy, CMO, mea Platform
How Fast Should (Re)Insurers Expect to Scale Agentic AI Across Operations?
The methodology for proving out technology (use cases, ROI, staged rollouts) has not changed. What has changed, according to Elliott Bundy, is the speed at which execution has to happen and the speed at which (re)insurers reach conclusions about where else the technology applies.
In mea Platform’s experience, customers are often the ones pushing for expansion into adjacent use cases once the initial deployment is live. The starting point is typically one department, or one section of one department; the live question quickly becomes how fast the investment can spread.
The Bottom Line for Insurance Leaders Evaluating AI in Operations
The market has moved past the question of whether AI belongs in insurance operations. The live question is how deeply (re)insurers are willing to let it operate, and whether they are prepared to change the operating model around it rather than layer it on top of the one that already exists.
The economic case sits at the executive level. The metrics agentic AI moves (when it is deployed as operational infrastructure rather than as a feature) are combined ratio, underwriter capacity, cost per transaction, operating leverage, and where competitive differentiation actually sits.
About the Panel: Datos Insights & mea Platform
Elliott Bundy is Chief Marketing Officer at mea Platform. He spoke alongside moderator Subhasis Mukherjee, Senior Principal at Datos Insights, at the Datos Insights Insurance Leaders Technology Forum in Boston.
mea Platform builds insurance-native agentic AI for (re)insurance operations, including ingestion, underwriting, claims, and finance. mea Platform is pre-trained on real insurance documents and workflows via a domain-specific language model (dsLM) and proprietary insurance knowledge graph.
