Christy Carroll | AI strategist + design consultant
why pick a lane?
The most interesting problems live between disciplines. I work across AI strategy, product design, conversation design, user research, and code.
Every AI team needs someone experienced who can talk to engineers, listen to users, design for uncertainty, and speak up when the technology isn’t fully baked. I’m that gal.
featured work
PolicyAI
When 80% accuracy feels like 30%
Trust diagnostics for an LLM-powered policy assistant
80% technical accuracy, 30% perceived accuracy, inconsistent adoption
AI-adapted heuristic audit revealed the gap: missing content architecture, not model failure
Delivered trust-focused playbook before scaling 3x
PolicyAI
When 80% accuracy feels like 30%
Trust diagnostics for an LLM-powered policy assistant
80% technical accuracy, 30% perceived accuracy, inconsistent adoption
AI-adapted heuristic audit revealed the gap: missing content architecture, not model failure
Delivered trust-focused playbook before scaling 3x
Intelligent guidance
Designing for what AI can't do (yet)
Multimodal redirection for a top-5 US bank
Brittle IVR design over-relied on AI understanding customer intent
Built adaptive flow that works with AI limitations instead of against them
Shipped to production, adopted across lines of business
Intelligent guidance
Designing for what AI can't do (yet)
Multimodal redirection for a top-5 US bank
Brittle IVR design over-relied on AI understanding customer intent
Built adaptive flow that works with AI limitations instead of against them
Shipped to production, adopted across lines of business
what I’ve worked on
Document intelligence: DocAI
UX audit, research, and strategy for an AI system interpreting policy documents for financial underwriters. Research revealed the trust calibration gap: 80% model accuracy, but only 30-50% user confidence, with professionals bypassing AI summaries altogether in favor of manual verification (or using a different AI tool). I designed evidence-surfacing patterns so users could verify results and build appropriate trust.
Education: Personalized tutoring assistant
Designed conversational guidance that helped college students ask better questions, stay engaged with material, and build confidence in their own reasoning instead of relying on answers alone.
Enterprise assistants: Visa, IBM, & ADP
Conversation patterns, tone frameworks, and error-handling for internal AI that supports complex workflows while respecting employee expertise and autonomy.
Healthcare: Patient appointment scheduling
Shaped conversational flows for sensitive, high-stakes interactions where clarity, accessibility, and graceful recovery from errors were as important as speed.
Human-AI collaboration: Real-time agent assist
Research with live agents testing prototypes for concurrent customer handling — up to three at once. Observed how they interpret AI suggestions in the moment, where cognitive load accumulates, and what support actually helps instead of interrupts.
Channel strategy: IVR self-service playbooks
Playbooks and decision frameworks for when voice interaction can resolve an issue, when to escalate to humans, and when traditional UI is the better choice.
Evaluation frameworks: LLM quality beyond accuracy
Developed evaluation approaches measuring tone, frustration handling, condescension, and trust impact—the dimensions that predict adoption, not just task completion.
what I’ve worked on
Document intelligence: DocAI
UX audit, research, and strategy for an AI system interpreting policy documents for financial underwriters. Research revealed the trust calibration gap: 80% model accuracy, but only 30-50% user confidence, with professionals bypassing AI summaries altogether in favor of manual verification (or using a different AI tool). I designed evidence-surfacing patterns so users could verify results and build appropriate trust.
Education: Personalized tutoring assistant
Designed conversational guidance that helped college students ask better questions, stay engaged with material, and build confidence in their own reasoning instead of relying on answers alone.
Enterprise assistants: Visa, IBM, & ADP
Conversation patterns, tone frameworks, and error-handling for internal AI that supports complex workflows while respecting employee expertise and autonomy.
Healthcare: Patient appointment scheduling
Shaped conversational flows for sensitive, high-stakes interactions where clarity, accessibility, and graceful recovery from errors were as important as speed.
Human-AI collaboration: Real-time agent assist
Research with live agents testing prototypes for concurrent customer handling — up to three at once. Observed how they interpret AI suggestions in the moment, where cognitive load accumulates, and what support actually helps instead of interrupts.
Channel strategy: IVR self-service playbooks
Playbooks and decision frameworks for when voice interaction can resolve an issue, when to escalate to humans, and when traditional UI is the better choice.
Evaluation frameworks: LLM quality beyond accuracy
Developed evaluation approaches measuring tone, frustration handling, condescension, and trust impact—the dimensions that predict adoption, not just task completion.
how I work
Document intelligence: DocAI
UX audit, research, and strategy for an AI system interpreting policy documents for financial underwriters. Research revealed the trust calibration gap: 80% model accuracy, but only 30-50% user confidence, with professionals bypassing AI summaries altogether in favor of manual verification (or using a different AI tool). I designed evidence-surfacing patterns so users could verify results and build appropriate trust.
Education: Personalized tutoring assistant
Designed conversational guidance that helped college students ask better questions, stay engaged with material, and build confidence in their own reasoning instead of relying on answers alone.
Enterprise assistants: Visa, IBM, & ADP
Conversation patterns, tone frameworks, and error-handling for internal AI that supports complex workflows while respecting employee expertise and autonomy.
Healthcare: Patient appointment scheduling
Shaped conversational flows for sensitive, high-stakes interactions where clarity, accessibility, and graceful recovery from errors were as important as speed.
Human-AI collaboration: Real-time agent assist
Research with live agents testing prototypes for concurrent customer handling — up to three at once. Observed how they interpret AI suggestions in the moment, where cognitive load accumulates, and what support actually helps instead of interrupts.
Channel strategy: IVR self-service playbooks
Playbooks and decision frameworks for when voice interaction can resolve an issue, when to escalate to humans, and when traditional UI is the better choice.
Evaluation frameworks: LLM quality beyond accuracy
Developed evaluation approaches measuring tone, frustration handling, condescension, and trust impact—the dimensions that predict adoption, not just task completion.
how I work
Document intelligence: DocAI
UX audit, research, and strategy for an AI system interpreting policy documents for financial underwriters. Research revealed the trust calibration gap: 80% model accuracy, but only 30-50% user confidence, with professionals bypassing AI summaries altogether in favor of manual verification (or using a different AI tool). I designed evidence-surfacing patterns so users could verify results and build appropriate trust.
Education: Personalized tutoring assistant
Designed conversational guidance that helped college students ask better questions, stay engaged with material, and build confidence in their own reasoning instead of relying on answers alone.
Enterprise assistants: Visa, IBM, & ADP
Conversation patterns, tone frameworks, and error-handling for internal AI that supports complex workflows while respecting employee expertise and autonomy.
Healthcare: Patient appointment scheduling
Shaped conversational flows for sensitive, high-stakes interactions where clarity, accessibility, and graceful recovery from errors were as important as speed.
Human-AI collaboration: Real-time agent assist
Research with live agents testing prototypes for concurrent customer handling — up to three at once. Observed how they interpret AI suggestions in the moment, where cognitive load accumulates, and what support actually helps instead of interrupts.
Channel strategy: IVR self-service playbooks
Playbooks and decision frameworks for when voice interaction can resolve an issue, when to escalate to humans, and when traditional UI is the better choice.
Evaluation frameworks: LLM quality beyond accuracy
Developed evaluation approaches measuring tone, frustration handling, condescension, and trust impact—the dimensions that predict adoption, not just task completion.
experience
Senior AI Design Consultant — Slalom / Capital One
Conversational and AI UX for enterprise document intelligence, including research, heuristics, and trust calibration frameworks addressing the gap between model accuracy and user adoption.
Conversational Architect — Amelia.ai (now SoundHound AI)
Text and voice agents in healthcare and service contexts, focusing on multi-turn flows, error recovery, and safe escalation paths.
Lead Product Designer — IBM Watson & Cloud Garage (2015–2020)
Led design for early Watson AI products including education and customer intelligence applications—full product design, not just conversation flows. Later, I ran AI-focused design engagements through IBM Garage, facilitating enterprise design thinking workshops and helping Fortune 100 clients define conversational patterns, evaluation frameworks, and AI integration strategies. This is where I developed the foundational thinking about human-AI collaboration that became my trust calibration work.