AI Chatbot
Live evidence showed that a narrower, menu-driven chatbot was more useful than a more ambitious open-ended experience. The work turned complex tuition-reimbursement policy into reliable conversational guidance while protecting accuracy and trust.
Role: UX Writer and Content Designer, Guild
The Context
Guild helps employees use education benefits to pay for school. Tuition-reimbursement policies varied and generated a high volume of questions for Member Services. The team wanted a chatbot that could answer common questions quickly, but the experience had to work within Amazon Lex’s technical limits and avoid turning nuanced policy into misleading guidance.
The working content inventory mapped source answers, conversational copy, fallback language, and review notes.
What I Did
I worked with Engineering to understand Amazon Lex, audited existing call-center content, and reshaped policy answers into conversational patterns. I revised the language for clarity and brand voice, then partnered with Member Services to check each answer against real policy.
We launched the experience to a limited audience so the team could learn from real behavior before expanding it.
The initial launch gave the team evidence about how members interpreted open-ended prompts.
Challenges I Faced
Analytics showed that open-ended questions created confusion. Members could phrase the same need in unpredictable ways, while reimbursement answers had to remain precise.
We narrowed the experience to a menu-driven model. That reduced ambiguity, made supported topics visible, and gave members a dependable path to a live specialist when the chatbot could not answer.
The revised interaction surfaced supported reimbursement topics instead of asking members to guess what the chatbot understood.
Policy content was reviewed against the source material before launch.
Fallback paths made the limits of the chatbot explicit and connected members to human support.
The End Result
The final chatbot handled a limited but valuable set of common reimbursement questions with clearer, more reliable guidance. The constrained model balanced conversational ease with policy accuracy and established a reusable approach that leadership prioritized for additional use cases.
The final experience paired menu-driven choices with a clear escalation path.