As the sole UX designer responsible for Honeywell's conversational AI interfaces, I designed, tested, and iterated on IVA/IVR call flows and web chatbots across three distinct enterprise divisions — each with its own user base, domain vocabulary, and support complexity.
OVERVIEW
Conversational AI Across Enterprise at Scale
Conversational AI
IVA / IVR
Chatbot UX
Dialogue Flow
🔒 Conversation flows, dialogue scripts, and internal UX documentation are proprietary to Honeywell. No visuals are available for public display.
Honeywell operates across multiple high-stakes enterprise divisions — from aerospace components and avionics to fire safety systems and building automation. Each division runs its own customer-facing support infrastructure, including web chatbots and phone-based IVA/IVR systems that handle high volumes of technical queries, order management, and escalation routing daily.
As the sole UX designer on conversational AI, I was responsible for the end-to-end UX of these interfaces — from initial user research and analytics review through flow design, UX copy, usability testing, and iterative refinement. Critically, while the underlying technical architecture was shared, each division required bespoke conversation design tailored to its users, domain language, and support scenarios.
The Core Challenge
Designing for two fundamentally different modalities — voice (IVA/IVR) and text (chatbot) — each with distinct constraints, failure modes, and user expectations
Maintaining a shared structural framework across three divisions while ensuring each felt tailored to its specific user base and domain
Identifying and closing conversation loopholes — edge cases where users fell into dead ends, unrecognised intents, or unhelpful loops
Designing for users in high-stakes, time-pressured contexts — a fire safety technician or aerospace engineer calling support has very different needs from a typical consumer
My Responsibilities
End-to-end UX ownership of all IVA/IVR and chatbot interfaces across three Honeywell divisions
User research and analytics review to identify pain points, drop-off patterns, and intent mismatches
Conversation flow design, dialogue tree restructuring, and UX copywriting
Usability testing — both moderated studies and analysis of live usage data
Competitor analysis and benchmarking across enterprise conversational AI
Cross-functional collaboration with AI/NLP teams, product owners, and divisional stakeholders
Work Stream 01
IVA / IVR - Voice Experience Design
Interactive Voice Assistant
Interactive Voice Response
Phone Support Flows
The IVA/IVR system is the voice-based front door to Honeywell's support infrastructure — handling call routing, self-service resolution, and live agent escalation across thousands of daily inbound calls. Voice interfaces are unforgiving by nature: users cannot scan ahead, cannot re-read a missed prompt, and will abandon a call the moment a flow feels broken or circular. Designing for this modality required a level of rigour distinct from any screen-based work.
Design Challenge
Voice flows have no visual affordances — every interaction relies entirely on the quality of the dialogue, prompt timing, and error recovery
Identifying loopholes: paths where users became trapped in repetitive loops, triggered unrecognised intents, or were routed incorrectly
Designing graceful failure states — what happens when the IVA cannot understand the user, and how to recover without frustrating them into hang-up
Balancing self-service containment with appropriate escalation — ensuring users who genuinely needed a live agent could reach one without friction
My Contribution
🎧 On-Hold Audio Testing
One distinctive aspect of this work was owning the on-hold audio experience — an often-overlooked but psychologically significant touchpoint in voice support. I researched the cognitive and emotional effects of different hold music styles, conducted structured testing to evaluate options, and implemented the selected audio experience across the IVR system. This kind of end-to-end ownership — from conversation flow to the sound a caller hears while waiting — reflects the full-spectrum nature of voice UX design.
Testing Approach
Systematic loophole testing , scripted walkthroughs of every possible call path to surface edge cases, unhandled intents, and circular routing before go-live
Analytics-driven review — identifying high drop-off points, repeated reprompts, and escalation patterns in live call data
Usability testing with representative users to validate prompt clarity and flow logic
Iterative refinement post-deployment based on real call behaviour and updated analytics
Work Stream 02
Chatbot UX — Web Conversational Interface
Web Chatbot
FAQ Architecture
Live Chat
Multi-division
Each Honeywell division website hosts its own chatbot — a web-based conversational interface serving customers, partners, and technical users with self-service support. While the underlying chatbot architecture was shared, the UX of each required meaningful tailoring: the questions aerospace engineers ask are categorically different from those of a fire safety technician or a building systems operator. My role was to ensure each chatbot felt native to its division, not like a generic support widget.
Key Problems Solved
The database-powered FAQ system used a nested filter structure that was causing significant user drop-off — users were getting lost in multi-level filters and abandoning before finding answers
UX copy across chatbot responses was inconsistent and often technically dense — not written for the scanning, conversational reading pattern of chat interfaces
No clear pathway to live agent support — users who couldn't self-serve had no graceful escalation route
Flow logic across divisions was generic — not accounting for the distinct intent patterns and vocabulary of aerospace, fire safety, or building technology users
My Contribution
💬 Live Chat Feature Design
A key deliverable was designing and implementing the live chat escalation feature — a critical pathway for users whose needs the bot couldn't meet. This involved designing the full handoff flow (from bot conversation to agent queue), the queue state UX (wait time communication, position indicators), and all associated UX copy. Getting this right was essential: a poorly designed escalation path creates frustration at exactly the moment when a user most needs help.
Design Focus Areas
Filter hierarchy redesign — flattening over-nested FAQ category structures to reduce abandonment
Conversation flow tailoring — adapting shared bot architecture to three distinct division user bases and vocabularies
UX copy — writing and refining all chatbot dialogue, prompt text, error messages, and escalation copy
Live chat feature — end-to-end flow design for bot-to-agent handoff, queue UX, and agent-side interaction patterns
Analytics-driven iteration — using engagement data to identify drop-off points and validate redesigned flows post-deployment
Division Context
Designing Across Three Distinct User Bases
Aerospace
Fire
Building Technologies
The central design challenge unique to this role was that the same structural chatbot and IVR framework had to serve three fundamentally different enterprise user bases — each with its own domain language, urgency levels, and support expectations. A fire safety technician calling a support line and an aerospace engineer using a chatbot to check part availability are not just different users — they are in categorically different contexts, with different stakes and different tolerances for friction.
Aerospace
Users: Engineers, MRO technicians, procurement teams
Primary needs: Part availability, order status, technical documentation, pricing
Context: High technical specificity — part numbers, certifications, compliance requirements
Chatbot: Honeywell Assist — a named, public-facing product serving the aerospace support portal
Fire & Life Safety
Users: Fire safety technicians, installers, building managers
Primary needs: Product support, installation guidance, compliance queries, fault resolution
Context: Safety-critical domain — users may be troubleshooting under time pressure in the field
Tone: Precise, authoritative, and efficient — no ambiguity in a safety context
Building Technologies
Users: Facilities managers, systems integrators, building operators
Primary needs: Product information, technical support, system configuration guidance
Context: Broad product range — HVAC, access control, security, energy management
Tone: Practical and solution-oriented — users are operational, not exploratory
What Tailoring Actually Meant in Practice
Intent mapping — identifying the top queries specific to each division through analytics and research, then structuring flows around those intents
Vocabulary — UX copy, quick replies, and prompt language calibrated to each division's domain terminology
Escalation thresholds — when and how aggressively to offer live agent handoff differed meaningfully across divisions
FAQ architecture — the category structure and filter logic of the FAQ system was independently designed for each division's content landscape
