Aviza Copilot — Bilingual Insurance Support Assistant

Compliance-Guarded French/Dutch Copilot Studio Agent for a Belgian Insurance Broker

Stack: Microsoft Copilot Studio · GPT-4o · Adaptive Cards · Conversational Booking · Microsoft Bookings

The Challenge

Aviza.be, a Belgian insurance broker, had a chatbot that stayed silent until a visitor typed first, then interrogated them for personal details before offering even basic information. Its fallback logic was broken, and in a regulated insurance context it carried real compliance risk — with no controls in place to prevent the bot from discussing quotes, pricing, medical conditions, or competitors.

The Solution

I rebuilt the assistant in Microsoft Copilot Studio on GPT-4o, combining Generative Answers with deterministic, rule-based topic controls. The bot now greets visitors proactively in both French and Dutch as soon as chat loads, and leads with immediate value — phone number and hours — before ever asking for personal information. Rule-based guardrails redirect any attempt to discuss quotes, medical conditions, or competitors to a compliant response, and Adaptive Cards replaced raw URLs for Microsoft Bookings, with routing based on the visitor’s language.

Implementation

  • Microsoft Copilot Studio on GPT-4o, combining Generative Answers with deterministic controlled topics
  • Proactive bilingual (FR/NL) greeting on chat load, replacing the previous silent start
  • Restructured conversation flow: immediate value (phone + hours) before any data collection
  • Rule-based compliance guardrails with topic-level redirection for quotes, medical conditions, and competitors
  • Adaptive Cards replacing raw URLs for Microsoft Bookings, with FR/NL routing
  • Funnel-tracking events for greeting, guardrail triggers, booking initiated/clicked, and call CTA

Screenshot of the original Aviza Copilot Adaptive Card prompting the visitor to book a Microsoft Bookings appointmentOriginal flow: the Adaptive Card prompts the visitor to book via an external Microsoft Bookings link.

Booking Flow Evolution

The initial Adaptive Card booking flow — a card prompting the visitor to book via an external Microsoft Bookings link, then confirm back in chat — was later evolved in a follow-on phase into a fully conversational, in-chat booking experience. The assistant now checks real-time availability, presents open time slots directly inside the conversation, and lets the visitor select and confirm a slot without leaving the chat window. Booking validation and lead details are captured and stored directly to SharePoint, removing the external booking-and-confirm round trip entirely.

Screenshot of the evolved Aviza Copilot conversational booking flow showing real-time available time slots selected directly in chatEvolved flow: real-time available slots are presented and confirmed directly in chat.

Outcome

Replaced a compliance-risk chatbot with a proactive, bilingual assistant that leads with value, keeps every conversation inside safe, compliant boundaries, and — through the follow-on phase — handles booking entirely in-chat with real-time availability and SharePoint-backed lead capture.

Technology Stack

  • Microsoft Copilot Studio — bilingual agent orchestration and deployment
  • GPT-4o — Generative Answers grounded on approved content
  • Adaptive Cards — bilingual Microsoft Bookings routing
  • Conversational Booking — real-time slot selection, validation, and SharePoint lead storage
  • Funnel-tracking events — visibility into greeting, guardrails, booking, and call CTA

Discuss a Similar Project →