The insurance industry runs on documentation. Sales conversations, policy proposals, client choices, everything needs to be tracked, summarized, and handed over to implementation teams. But what if that handover could happen in seconds instead of hours? Our colleagues at Tensr, the AI experts within the Google at Cronos community, showed us exactly how agentic AI makes this possible, addressing challenges that every insurance broker faces.
The Challenge: When Growth Creates Friction
Here’s a paradox every growing business knows too well: your revenue increases, your team expands, but somehow everything takes longer. Meetings multiply. Emails pile up. Simple questions require hunting through fragmented systems.
For insurance brokers, this growth paradox hits especially hard. Sales teams excel at selling, not at meticulous CRM documentation. After closing a deal, they dump everything into their CRM system: emails, PDFs, proposals, sometimes even hastily scanned handwritten notes. Then comes the painful part: the handover.
Implementation specialists need to dig through this information to understand what was actually sold, what choices the client made, what proposals were rejected, and what’s next. The typical process? An hour or more of reading through disorganized documents, followed by a handover meeting that pulls valuable time away from both teams.
Even worse, incomplete documentation creates compliance risks. Insurance regulations require detailed records of all client communication. This documentation burden is universal across the industry.
The Solution: Agentic AI as Your Documentation Partner
This is where agentic AI changes everything. Not a chatbot that answers questions, but an AI agent that actually takes action.
Our colleagues at Tensr built a proof-of-concept that demonstrates the real power of agentic AI within existing CRM systems. The tech stack? Google’s open-source Agent Development Kit, integrated through custom Model Context Protocol (MCP) servers, deployed on Vertex AI Agent Engine.
Agentic AI Power
The AI doesn’t just read and summarize. It executes tasks.
Document Intelligence
The agent reads all documents, emails, and notes in a CRM deal, then creates chronological summaries of client communication with key decisions and proposal comparisons presented in implementation-ready format.
Automated CRM Data Entry
The agent extracts client information from policy documents and creates new deals, contacts, and organizations automatically. It populates fields with addresses, phone numbers, and policy details while attaching source documents for reference.
Compliance Automation
For regulatory documentation, the agent generates required forms documenting all client choices, proposals offered, and rejections, while flagging missing information that needs follow-up. What normally takes 30 minutes happens in seconds.
Human expertise
Crucially, human expertise remains central. The agent provides source links for every claim, flags data gaps explicitly, and requires expert review before finalization. This builds traceability for audits while respecting professional judgment.
We saw eyes light up when we demonstrated how the system could generate a complete mediation form from a messy deal. What normally takes half an hour happened in seconds. But the implementation specialist still reviews everything. We’re not replacing expertise, we’re eliminating busywork. — Jony Van Puymbroeck, Partner at Tensr
Technical stack
The demo ran entirely on Google’s AI infrastructure using the Agent Development Kit for building AI agents, Vertex AI Agent Engine for managed runtime with automatic scaling, custom MCP servers for CRM integration, and Gemini models powering document understanding. The Agent Development Kit works with any LLM provider, but the seamless integration with Google Cloud’s managed runtime makes deployment ridiculously simple.
Tensr built this entire proof-of-concept in about a week.
Lessons Learned: Four Pillars for AI Success
Tensr’s experience reveals several critical lessons for any business exploring agentic AI.
Talk to your users
Start with user pain points, not technology. Don’t automate processes that users don’t understand or value. Talk to end users first, identify their real frustrations, and build something that genuinely helps them.
Human in the loop
Keep humans in the loop for critical processes like contracts, compliance, and customer commitments. AI should prepare 80-90% of the work, not make final decisions. Include source links for validation and let experts do what they’re great at: judgment calls.
Context is king
Data context is king. Before building AI agents, audit your data sources. Is information structured consistently? Are naming conventions clear? Sometimes investing in data quality pays bigger dividends than fancy models.
Teamwork makes the dream work
Finally, build agent teams, not god systems. Instead of one massive AI that does everything, create specialized agents: a content agent that writes summaries, a compliance agent that enforces regulations, and a technical agent that orchestrates the workflow. Specialized agents are more reliable, easier to maintain, and scale better.
The Bigger Picture: AI as Your Back-Office Partner
Here’s what gets us excited about this case: it’s not science fiction. It’s practical AI solving real business problems today.
Every company dealing with fragmented data can benefit from this approach. Agentic AI acts as the connective tissue between your existing tools, not a replacement for them. It reads your documents, extracts what matters, takes action across systems, and presents results in formats humans can validate and trust.
The insurance broker’s challenge isn’t unique. It’s universal: How do we make our systems work together without forcing people to become data detectives? Agentic AI answers that question.
Want to explore what agentic AI could do for your business? Reach out to Dirk Vereycken at GC innovate and we’ll connect you with the right experts in the Cronos group community, like Jony Van Puymbroeck at Tensr. No generic pitches, just honest conversations about whether AI can genuinely solve your pain points.
Let’s make cool things happen fast, efficiently, and at your own pace.
