Nuance · 2019–2022
Designing a multimodal authoring system for the conditional conversations enterprise IVR teams could not safely express in chatbot tools.
Senior UX Designer
Interaction Design Lead
Behavior Authoring
Research + Prototyping
01 · Situation
Before large language models, enterprise voice systems depended on deep conditional logic and intent recognition to decide what a conversation should do next—and what should happen when it failed.
Most authoring tools in the space focused on virtual assistants. IVR teams carried a harder operational burden: no-match and no-input paths, retries, escalation, channel differences, nested business rules, and graceful recovery when recognition broke down.
The existing workaround was a handoff artifact. Conversation designers built long conditional spreadsheets to explain intended behavior to customers and engineers. The logic was difficult to read, expensive to change, and separated design intent from implementation.
02 · Platform context
Mix.dialog was the multimodal behavior layer—not a standalone chatbot builder. It connected language understanding to deployable conversation logic across voice and digital channels.
NLU
Mix.nlu
Defined intents and entities the experience could recognize.
01
Mix.dialog
Authored prompts, conditions, actions, and recovery behavior.
Dashboard
Dashboard
Managed and deployed dialog and NLU configurations.
03 · My role
I led designs with conversation designers, product leadership, and engineering directors. My technical background made me uniquely useful in a space where interface decisions changed how the underlying product had to represent and generate logic. Along the way I conducted research with internal conversation designers, prototyped interaction models and failure states, partnered with engineering on feasible data and UI structures, and created patterns that became templates for the wider tool.
Scope
The work crossed node behavior, shared properties, and the connection between NLU recognition and actions in the dialog.
Anchor pattern
Question & Answer node
Reworked the central intent-discovery loop and used its complex properties as a model for the rest of the authoring experience.
Behavior building
Messages + conditions
Designed how authors created conditional prompts, nested logic, and channel-aware messages without reading or writing generated code.
System connection
NLU + actions
Integrated intents and entities into the QA node so recognized information could map to downstream behavior and actions.
04 · Action
The condition stack became a focused design program: understand the failure, test multiple structures, and let feedback shape the product’s architecture as well as its interface.
01 · Diverge
We failed a lot—and learned why
Code-forward concepts were powerful but alienating. Form-heavy designs hid context. Other versions were hard to scan or too rigid for the depth IVR teams needed.
02 · Listen
Designers needed context before power
Research showed that nested conditions were not only a syntax problem. Authors needed clear starting zones, visible hierarchy, readable content, and confidence about what an edit would affect.
03 · Converge
A visual structure that still mapped to valid logic
The final direction treated conditional groups as directly manipulable objects. Authors could read hierarchy, add branches, move content, and understand the output without translating a spreadsheet or writing code.
04 · Refine
Small behaviors carried architectural weight
Adding, deleting, reordering, and changing an if branch into an else-if branch all needed predictable rules. These details were essential to making a deep system feel safe rather than fragile.
How I worked
This was not a screen-design handoff. The condition model evolved in a continuous loop with conversation designers, product, and engineering. Interface discoveries frequently became architecture conversations.
01
Research real authoring work
Study spreadsheets, tools, handoffs, and nested IVR scenarios.
02
Prototype competing mental models
Tables, forms, graphs, code-like syntax, and direct manipulation.
03
Validate comprehension and control
Test context, nesting, add points, movement, and recovery.
04
Shape implementation with engineering
Align interaction rules with valid generated logic and data.
05 · Result
The work improved deeply nested condition authoring while establishing interaction patterns that influenced the wider Mix platform.
Measurable usability improvement
Research showed improved usability for deeply nested conditions. The current public evidence does not include a shareable numerical measure.
Proven in enterprise contexts
The workflows supported customers including Sony Interactive Entertainment and Rakuten, where reliability and complex branching mattered.
A platform-level template
The QA node and condition work established properties and interaction patterns that informed other nodes and shared product behavior.
Customer context. Sony Interactive Entertainment used Mix.dialog and its condition system in work supporting voice functionality for PlayStation 5. This case study focuses on the authoring system—not Sony’s implementation details.
06 · Later chapter · 2022–2023
The Mix.dialog work exposed a broader need for consistency, accessibility, and a shared language between design and engineering. I later led the design and front-end architecture of Verse.
Let’s make complex systems understandable.
Senior UX Designer · UX Engineer · Design systems and AI tools
Get in touch