Shift Technology recently hosted a webinar on the impact of agentic AI within the insurance sector. Moderated by Patrice Amann, EMEA Regional Business Lead – WW Financial Services at Microsoft, the panel featured Robert Malan (Associate Partner, QuantumBlack, AI by McKinsey), Eric Sibony (Co-founder and CSO, Shift Technology), and Grady Behrens (Product Marketing Lead, Shift Technology). This article summarises the key themes covered during the English-language discussion.
Generative AI has dominated headlines — but that was only the beginning. Agentic AI now enables tangible transformation across claims operations, delivering measurable value. While generative AI excels at single-step tasks such as data extraction, document categorisation or conversational assistance, agent-based systems go significantly further. AI agents can reason over multiple steps, act on structured and unstructured data, and autonomously execute complex claims workflows. For insurers, this means automating many of today’s time-consuming tasks while supporting handlers in making better decisions when handling complex cases.
What agentic AI can do
Agentic AI represents the latest evolution of large language models (LLMs), specifically leveraging their reasoning, workflow orchestration and action-taking capabilities. In a claims automation scenario, it can replicate the actions of an experienced handler using historical outcomes, orchestrate multi-step processes — Interpretation → Contextualisation → Decisions → Actions — and interact with systems across the insurer and its service providers. As a result, agentic AI can fully automate processes end-to-end, from FNOL to payment. Crucially, agents can escalate to a human handler when required, providing all relevant information and recommendations to support optimal decision-making.
Why it matters
AI — and insurance AI in particular — cannot rely on a one-size-fits-all approach. Predictive AI excels at detecting anomalies at scale, making it ideal for fraud detection. Generative AI unlocked new capabilities for analysing and synthesising unstructured data. Agentic systems stand apart by being designed to perform multi-step actions and orchestrate workflows. At scale, agentic AI enables full automation of complex use cases that once required human judgement, significantly reducing effort in the most difficult parts of the claims process while maintaining oversight.
Tangible results
Early deployments are already delivering measurable improvements. Processing times have fallen from weeks to minutes in some workflows. Automation rates have exceeded 50% in certain implementations.
Beyond automation accuracy, agents support handlers directly in their day-to-day work, classifying, prioritising and recommending next-best actions to reduce claims costs and improve customer satisfaction. Early adopters report a 3% reduction in claims costs and a 30% increase in operational efficiency.
Key recommendations
Human involvement is essential
Technology alone is not enough. Agentic AI works best when embedded within the right operating model. Before deployment, insurers must capture tacit expertise via workshops with top handlers to encode specific rules and processes. Multidisciplinary teams must co-develop agents with end-users, rather than treating initiatives as isolated “lab experiments”. Data quality is critical: agents require reliable, real-time inputs to make accurate decisions, making data modernisation a common first step.
Governance — trust by design
Agents can deviate, making pragmatic governance essential. Define configurable checkpoints and human-in-the-loop thresholds, maintain full decision traceability, and conduct pre/post audits targeting the rules and data inputs used. Establish continuous monitoring and retraining to prevent drift and costly hallucinations. In short: treat agents like digital colleagues requiring supervision and continuous improvement.
KPIs to track
Focus on metrics aligned with business value: accuracy (document extraction, liability assessment), automation rates and processing times (moving from weeks to minutes). Also measure operational impacts such as onboarding time for new staff or surge response capacity during catastrophes, to capture the broader value of agentic automation.
A practical deployment roadmap
Start with high-impact pilots addressing complex, high-value processes rather than trivial tasks. Extract and structure historical files, and codify expert judgement to build training datasets rooted in real decisions. Establish a cross-functional “delivery factory” to develop, test, deploy and monitor agents. Define KPIs and guardrails early, beginning with human review and gradually increasing autonomy as reliability is proven. Finally, iterate continuously — agents require ongoing supervision, retraining and governance to remain effective.
Conclusion
Agentic AI does not replace human judgement — it redefines how it is applied. Claims handlers will shift from manual processing to supervision, exception handling and higher-value decision-making. Insurers must re-design operations so that humans and agents work as a team: agents providing scale and consistency; humans providing judgement, empathy and oversight.
To learn more about Shift Technology’s Agentic AI solution, consult our press release on Shift Claims.
