Conversation Design for AI-driven Benefits Verification

−

Case Detail

Built a visual conversation builder for Infinitus's AI-driven insurance verification system, enabling faster automation of complex workflows through reusable dialogue components.

Background

Infinitus faced a critical challenge while scaling their benefits verification automation.

In order to build and debug these complex conversations at scale, conversation designers needed to make and manage flows.

The existing tool was becoming a bottleneck — conversation designers struggled with an unintuitive interface, lacked visual feedback, and could not easily reuse components across different flows and customers.

Key challenges

add_circle
The conversation design interface lacked intuitive controls and clear workflows, significantly impacting designers' ability to create and refine automated conversations
add_circle
Designers had no way to visualize the impact of their changes on conversation flows until after implementation, making iteration slow and error-prone
add_circle
Could not easily reuse components across flows and customers
add_circle
Complex workflows required significant human operator intervention

Outcomes

add_circle
Led design and engineering to overhaul the conversation builder tool, enabling users to convert new SOPs into automated dialogues >30% faster
add_circle
Enabled conversation designers to flatten conversations, opening the door for more sophisticated automation
add_circle
Improved response time to customer requests by >20%
add_circle
Standardized component implementation, including providing automatic usage metrics for product and engineering
add_circle
Enabled rapid iteration on conversation designs through improved visual feedback and reduced QA time by >20%

Business Impact

Improved Efficiency

add_circle
Reduced time to create new conversation flows
add_circle
Decreased maintenance overhead for existing flows
add_circle
Enabled rapid testing of automation strategies

Better Scalability

add_circle
Standardized components reduced implementation time
add_circle
Reusable patterns accelerated new development
add_circle
Improved visibility into system behavior

Enhanced Quality

add_circle
Real-time preview enabled earlier detection of issues
add_circle
Standardized validation improved data quality
add_circle
Better visualization reduced implementation errors

Infinitus Flow

This example explains the relationship between the conversations and the call page used by operators. Use arrow keys or swipe to navigate.

We start with the simplest case to demonstrate our new approach: every call needs the patient's full name. Each step in an insurance verification call is a set of questions to fulfill data requirements. Based on the data, we direct to a different "action" or new data collection step.

If the operator we're speaking with doesn't give us the full name, we have follow-up questions to get whichever part of the name we need.

Once we have the full name, we can proceed to the next step. In this case, it's getting the patient's date of birth (an action called "getPatientBirthday"), so we can complete the validation steps, ensuring we're managing care for the correct patient.

Abstractly, each action looks like this. It has a name, a number of opening questions that should allow us to fill as many fields in a single conversation turn, and a set of fields that we need to collect. Each field has additional questions specific to that field, in case we need to ask for more information.

Solution

Redesigned the conversation builder from the ground up, focusing on three key innovations

Visual Flow Editor

Improved visualization and editing of conversation flows

add_circle
Developed a visual interface that enables designers to create and modify complex insurance verification conversations efficiently
add_circle
Visually map and modify conversation flows
add_circle
Preview utterances and responses in real-time
add_circle
Enforce validation rules for outputs against known data formats

Standardized Component Library

Increased consistency and reusability of dialogue components

add_circle
Developed comprehensive set of reusable dialogue components
add_circle
Enforced consistent implementation patterns for dialog components
add_circle
Provided automatic usage metrics
add_circle
Simplified complex interaction patterns into reusable blocks
add_circle
Enabled rapid prototyping of new conversation flows

Output Management System

Improved data quality and system integration

add_circle
Built flexible system for managing conversation outputs
add_circle
Enforced data validation rules at design time
add_circle
Provided clear visibility into data dependencies
add_circle
Enabled cross-organization output standardization
add_circle
Simplified integration with downstream systems

Process

User Research

add_circle
Conducted extensive shadowing sessions with operators and conversation designers
add_circle
Interviewed >20 users across different roles to understand pain points
add_circle
Created detailed workflow diagrams to identify optimization opportunities

Iterative Design

add_circle
Developed multiple design iterations with stakeholder feedback
add_circle
Used high-fidelity designs in production to test with users
add_circle
Conducted usability studies to validate design decisions
add_circle
Iterated based on quantitative usage data

Implementation Strategy

add_circle
Created phased rollout plan to minimize disruption
add_circle
Isolated UI and data changes with middleware to support gradual migration
add_circle
Developed comprehensive documentation and training materials
add_circle
Established new design patterns for future automation tools

Roles

Designer, Engineer, Researcher

Tools

Figma, Visual Studio Code, GitHub

Tech

TypeScript, React, Material UI, GraphQL, JavaScript, JSS, WebPack