GuideFlowAI
Designing an answer—not another search box.
The information existed. The experience of reaching it was broken.
Internal policies, documents and team knowledge were distributed across systems. People often needed a direct answer, but the available experience required them to search, interpret and ask someone else for confirmation.
The opportunity was not simply to add a chatbot. It was to design a trustworthy path from question to answer, source, responsible contact and next action.
Search returns documents. People often need decisions.
I reframed the interaction around four things a useful answer should provide: a concise response, a visible source, the relevant person and a clear next action.
Answers remain grounded in approved information rather than generated speculation.
Lead with the useful response, then make supporting context available.
Users can inspect the document behind the answer and understand where it came from.
When certainty is limited, the experience points to the appropriate human contact.
One response. Four layers of confidence.
Begin with the information most useful to the person asking.
Connect the response to an approved policy or internal document.
Identify the team or person responsible when human context is needed.
Help the user move forward rather than ending at information retrieval.
A working prototype made the design testable.
I translated the experience model into a public demonstration of document-grounded conversational retrieval. The prototype shows how the interface, answer hierarchy and source connections can operate together in practice.
Public demonstration. Initial response may take a short time while the service starts.
From internal knowledge to guided commerce.
The retail assistant is presented as a concept exploration—not a deployed product. It tests how the same guided-answer model might help a customer describe a need, understand recommendations and move toward an appropriate action.
Help an employee act with confidence.
Help a customer decide with confidence.
A demonstrated interaction model with clear questions for future testing.
Demonstrated
- Working document-grounded conversational prototype
- Interface model for sources, contacts and next actions
- Branded experience designed for nontechnical users
- Transferable framework for guided-answer concepts
Future validation
- Task-completion time
- Answer accuracy and source confidence
- Reduction in repeated questions
- Escalation frequency and user satisfaction