The conversation has shifted. A year ago, life, annuity, and benefits (L/A/B) carriers were running pilots. Now that they’re in production, the gap between those moving and those still planning is widening fast.
That was the throughline at our recent Regional Forum on Agentic AI in Des Moines, where senior leaders from across the L/A/B space gathered to share what’s working, what isn’t, and what’s next. Datos Insights Advisor Jack Krantz set the tone early: Roughly three out of four L/A/B carriers now have AI deployed in production. For the holdouts still in planning mode, his message was direct, “You are falling behind.”
The ROI Problem Nobody Wants to Talk About
For carriers that measured results, the news is mostly good. Efficiency gains are real, and ROI is positive in nearly all cases where it could be measured. The catch? About 60% of carriers went into their AI deployments without baseline metrics or defined success criteria. They improved their workflows—they just can’t prove it.
That era of exploratory, metrics-free experimentation is over. As Jack put it, “Proofs of concept have to be tied to a return.” The window for “we think it’s good” has closed.
The Blockers Are Real
Talent is the number-one barrier to scaling agentic AI in the L/A/B space, not budget. Entry-level AI salaries in tech compete with senior executive compensation in insurance, and most carriers can’t win that fight directly. The prescription from multiple voices in the room: Bring insurance expertise to platforms that others have already built.
Legacy system integration ranked second. What looked manageable in a curated pilot environment gets much harder in production, where data is messy and system connections are fragile. Several presenters offered their own approaches to building an orchestration layer over legacy infrastructure rather than replacing it, a pattern emerging as the dominant architecture for agentic deployment by L/A/B carriers.
Drift: The Risk Most Carriers Aren’t Managing Yet
Another topic that drew visible recognition around the room was model drift. As underlying AI models update and as the data fed into them changes over time, the behavior of deployed agents drifts from what was originally tested and approved. In a heavily regulated industry where audit trails and decision consistency matter, this is a serious operational risk.
A few organizations have started building monitoring frameworks. Most haven’t. The carriers ahead on this issue will be better positioned when regulators come knocking.
What Early Adopters Are Doing Differently
Carriers that have successfully moved from concept to production distilled several lessons for their peers: Start small and be specific, design the process before applying AI, define outcomes upfront, and plan for production from day one.
Two insurer executives offered practitioner-level candor on what scaling actually looks like:
- Numerous change champions deployed for associate-facing AI tools—people who learn best practices, provide prompts, and help colleagues who want guidance rather than self-discovery
- Proof of value (not just proof of concept) small group rollouts with thumbs-up/thumbs-down feedback loops to validate accuracy and usefulness before scaling
- Productivity gains of 15% to 30% from associate-facing tools (Microsoft Copilot, Glean), self-reported by associates
- More than a 65% reduction in software delivery work time—what once took three days now takes one
- Governance model involving business operations, technology, risk, data, compliance, and legal in every evaluation to ensure time to value is shortened and not just time to an approval queue that’s longer than ever
- Continuous learning programs (LinkedIn Learning, AI courses) paired with tool rollouts, not just tool access
- KPIs and customer performance indicators defined upfront for each proof of value, then shared across teams to avoid starting from scratch on each use case
- Consistency, accuracy, and context named as the three priorities for customer-facing AI
Want to Go Deeper?
Some of the research behind what attendees heard in Des Moines, such as the “AI, the Implemented Reality” reports for P/C and L/A/B, is coming in July. Other research is available now. For example, check out our reports on Human-AI Supervision Models for Insurers and Model Context Protocol: Implications for Insurer Integrations. If you’d like to continue the conversation in person, join us for our Insurance Leaders Forum at InsureTech Connect in Las Vegas on September 29. It’s a gathering of senior leaders learning and sharing what’s real, not what’s aspirational.
