Three Interfaces, One Workflow: Designing UX for Brokers, Underwriters, and AI Agents

The biggest mistake in enterprise AI is forcing everyone to use the same interface. A broker doesn't need a dashboard. An underwriter doesn't need a chatbot. An AI agent doesn't need a pretty UI at all — it needs clean, typed schemas.

A successful AI system in insurance looks completely different depending on who — or what — is using it.

The Broker Interface: Email

Brokers have been sending submissions by email since the 1990s. Every "modern" insurtech that forces brokers to use a portal is fighting human behavior. And losing.

In Oyyo, the broker's interface is their existing email client. They send submissions to a dedicated address (e.g., cyber@your-mga.oyyo.ai). They receive acknowledgments, missing document requests, and quotes — all via email. They never log into anything. They never learn a new tool.

This isn't a compromise. It's a design decision. The best interface for someone who sends documents and receives responses is the one they already use every day.

The AI Agent Interface: Typed Schemas

Under the hood, AI agents don't see emails or PDFs. They see structured, strictly-typed JSON schemas that define exactly what data needs to be populated.

When the Extraction Agent processes a document, it doesn't "read" it in the human sense. It maps unstructured content to a schema with defined field types, validation rules, and acceptable values. Each field has a type (text, money, date, oneof), constraints (min/max, required), and a description that guides the LLM's interpretation.

This separation — unstructured input, structured output — is what makes extraction reliable. The AI doesn't hallucinate fields because the schema defines exactly what exists. It fills in what it can find and flags what it can't.

The Underwriter Interface: The Workspace

The underwriter sees everything the AI and the broker don't. They see the full workspace: the Journal timeline of all actions, the extracted data with confidence scores, the rater with premium calculations, the quote ready for approval.

The workspace is designed for rapid scanning and decisive action:

• Today Board surfaces what needs attention now — urgent, today, this week
• Journal shows the complete history of a submission in one scrollable timeline
• Forms display extracted data with confidence-colored fields — green means "verified," yellow means "check this"
• Quote Builder shows the assembled document with one-click send

The underwriter's job isn't data entry anymore. It's review, judgment, and decision. The interface reflects that: minimal input, maximum information density.

The Design Principle: Decoupled Presentation, Shared Data

All three interfaces — email, schema, workspace — operate on the same underlying data. When the broker sends a document, the AI processes it into structured data, and the underwriter sees that data in their workspace. When the underwriter corrects a field, the AI incorporates the correction. When the AI generates a quote, the broker receives it by email.

The data layer is shared. The presentation layer is decoupled. Each participant gets exactly the interface optimized for their role.

This is what "human-AI collaboration" actually looks like in practice. Not a chatbot. Not a dashboard with an "AI" badge. Three purpose-built interfaces, connected by shared, real-time data, each optimized for the participant that uses it.