Back to Selected Work

CASE STUDY · MORGAN STANLEY

AI Teammate

Voice-First Client Assistant & Advisor Configuration Portal

A voice-first client assistant and advisor configuration portal for service, testing, publishing, and escalation — enabling hands-free access to client data in professional settings.

Voice UXAI AssistantConfiguration PortalMulti-Modal

ROLE

Principal UX Designer

USERS

Financial Advisors, advisor support teams, NLP configuration managers

PLATFORM

Mobile, tablet, and desktop portal

FOCUS

Voice interaction design, AI configuration, multi-modal sync, escalation workflows, privacy controls

TOOLS

Figma, audio prototyping, user research, conversation flow design

CONFIDENTIALITY NOTE

Some details, screens, and workflows are recreated or generalized to protect confidential information while preserving the design challenge, process, and decision-making approach.

01
SITUATION

Situation

Business context

Financial Advisors needed a hands-free way to access client information and capture notes while on the move or in client-facing meetings — without visibly checking their devices.

User context

Advisors in client-facing scenarios couldn't pause to type. They needed voice-first access to portfolio data, notes, scheduling, and action items — with the professionalism and security that financial services demands.

System context

Existing advisor tools were entirely screen-dependent, creating friction in mobile and meeting contexts where hands-free, voice-first interaction is essential for maintaining professionalism.

02
DESIGN CHALLENGE

Design Challenge

Design a voice-first experience that is professional, secure, and seamlessly integrated with the advisor's desktop portal — with a configuration system that lets advisors personalize the AI without requiring technical expertise.

Screen dependency

Advisors couldn't access critical client data during meetings without visibly checking their devices, breaking the flow of professional client conversations.

Privacy and security

Voice interactions in financial services must handle sensitive client data with appropriate privacy controls — especially in shared or semi-public meeting environments.

AI configuration complexity

Different advisors have different practice needs, client types, and communication styles. The AI must be configurable and testable without requiring technical expertise.

Voice privacy controlsMulti-modal syncAdvisor personalizationEnterprise securityProfessional meeting context
03
MY ROLE

My Role

What I led

Voice interaction design and the configuration dashboard for the AI's personality, response logic, data access rules, and escalation behavior.

What I designed

Voice conversation flows, visual pulse feedback system, the advisor-facing configuration portal, and the testing and publishing pipeline for AI behavior changes.

Who I partnered with

NLP engineers, voice interaction researchers, compliance teams, advisor focus groups, and the engineering team building the voice platform.

04
MAKING THE WORKFLOW VISIBLE

Making the Workflow Visible

Prototyped conversation flows using high-fidelity audio mockups to test intent recognition, response tone, error recovery, and the boundary between voice automation and human escalation in professional settings.

VOICE INTERACTION FLOW · INTENT → RESPONSE → ACTION → ESCALATION
05
EXPLORING THE EXPERIENCE

Exploring the Experience

Conversation flow mapping

Designed the branching logic for voice intents, confirmations, disambiguation, and error recovery scenarios.

Visual feedback system

Created the pulse and state feedback indicators showing listening, processing, responding, and ready states.

Privacy control models

Explored push-to-talk, wake-word, and environment-aware activation models for different professional settings.

Configuration portal

Designed the advisor-facing dashboard for customizing AI personality, data access rules, and response behavior.

Testing & publishing pipeline

Built the flow for advisors to test their AI configurations in a sandbox, iterate, and publish to production.

Escalation patterns

Defined the rules and UI for when and how the AI should hand off to a human specialist or supervisor.

06
THE SOLUTION

The Solution

A multi-modal system where voice commands on mobile sync instantly with the advisor's desktop workstation — backed by a configuration portal that gives advisors control over AI personality, behavior, and escalation rules.

PRIMARY SOLUTION · DASHBOARD VIEW

Voice interface

Natural language interaction with visual pulse feedback, confidence indicators, and clear state communication.

Configuration dashboard

Advisor-facing portal for customizing AI personality, data access permissions, response patterns, and behavior rules.

Testing environment

Safe sandbox where advisors can test voice interactions and AI behavior before publishing changes to production.

Multi-modal sync

Real-time synchronization between voice commands on mobile and the advisor's desktop workstation and CRM.

Escalation manager

Rules engine for defining when and how the AI should hand off to a human specialist, with supervisor notification.

Analytics & feedback

Usage patterns, success rates, common intents, and advisor feedback for continuous improvement of the voice experience.

07
KEY DESIGN DECISIONS

Key Design Decisions

Push-to-talk + wake-word hybrid

WHYFinancial meeting environments require explicit control over when the AI is listening. Advisors need to feel confident that sensitive conversations aren't being captured.
TRADEOFFSlightly more activation friction vs. strong privacy guarantees in sensitive professional settings.
RESULTAdvisors felt confident using the AI during client meetings without privacy concerns or client discomfort.

Visual pulse feedback

WHYVoice-only feedback isn't enough in professional settings. Users need clear visual indication of AI state — listening, processing, or ready for the next command.
TRADEOFFScreen real estate and attention for visual feedback vs. the uncertainty of a purely audio-only interaction model.
RESULTReduced advisor uncertainty about whether the AI heard and understood their request, improving trust and adoption.

Advisor-controlled configuration

WHYAdvisors know their practice, clients, and communication style best. They should be able to customize how the AI behaves without needing engineering support.
TRADEOFFConfiguration UI complexity vs. AI relevance, personalization, and advisor ownership of the tool.
RESULTHigher adoption because advisors felt genuine ownership over the AI's behavior and could tailor it to their practice.

Staged publishing pipeline

WHYChanges to AI behavior in a financial context should be testable before going live. Advisors need a safe way to experiment without risking client-facing errors.
TRADEOFFSlower iteration cycle vs. reduced risk of embarrassing or inappropriate AI responses in real client conversations.
RESULTAdvisors iterated more confidently in the sandbox and published with fewer post-launch corrections needed.
08
IMPACT

Impact

USER IMPACT

Advisors gained hands-free access to client data during meetings, improving their responsiveness and professionalism in client-facing scenarios.

BUSINESS IMPACT

Improved advisor meeting productivity by enabling real-time information retrieval and note capture without breaking the flow of client conversations.

OPERATIONAL IMPACT

Reduced the friction of note-taking and action item capture during client conversations, ensuring follow-ups were captured in the moment.

PRODUCT IMPACT

Established the voice UX framework, configuration portal pattern, and testing pipeline used as the foundation for future AI assistant products.

09
REFLECTION

Reflection

What I learned

Voice in fintech isn't just about convenience — it's about making high-stakes information accessible at the speed of thought, in contexts where screens create friction or break professional rapport.

What I would improve

I would invest more in contextual awareness — the AI should adapt its behavior, response length, and interaction style based on whether the advisor is in a meeting, commuting, or at their desk.

How this shaped my approach

This project taught me that voice UX design is fundamentally interaction design for invisible interfaces. Every state must be communicated, every error must be recoverable, and trust must be earned through transparency — not assumed.