Designing AI interactions that build user confidence
Leading interaction design and usability testing for an AI writing tool while pivoting design direction based on user insights.
Organization
As the nature of this work is currently confidential, I can’t show many project visuals for the time being. For now, I’ll provide a description of the work I’ve done. If you’d like to learn more, I am happy to discuss on a call.
Role
UX design: design for Word Add-In and web platforms using Figma and our design system to quickly ideate through designs over time.
Rapid prototyping: created testable prototypes for validation and interaction with usability tests, pivoted to prioritize a single platform after learning from users in testing.
Usability testing: created test scenarios, selected participants, facilitated test sessions, analyzed results, and identified key challenges to address in future design iterations.
Engineering handoff prototypes and documentation for development.
Story mapping: worked with stakeholders in UX, product, and engineering to create, revise, and reach agreement on a story map for the feature.
Key deliverables
Prototypes for testing
Story maps for the product team
Engineering handoff prototypes
Design system contributions and updates
Reports on usability test findings
Challenge
For a new AI-supported writing feature, the UX team needed to design for both Microsoft Word Add-In and web platforms while pushing the scope of the product in challenging new ways. How could we design this interaction-rich new feature in multiple platforms in a way that leaves users confident and satisfied?
A flow of a second prototype exploring branching flows and looping behaviors.
After preparing a design as an interactable Figma prototype and designing a suitable remote moderated usability test, test participants quickly showed us glaring issues with how our product team was thinking about the feature.
We identified key challenges that demanded a significant pivot from our original plans. Our change of plans made it possible to iterate and perfect the feature with further tests with users to continue validating our riskiest assumptions.