2022 to 2023
Building trust in AI assistance
I designed a multilingual assistant inside the IBM Cloud console that answers in seconds and shows exactly where every answer came from.
- Role
- Lead Product Designer
- Team
- 1 researcher and 1 development team
- Duration
- 10 months
- Scope
- Discovery to shipped v1
- 8sAverage answer time
- Down from 15 minutes with a person
- 1,200Questions answered daily
- From no AI answer capability
- $1.2MSaved every year
- Support burden removed
- +18.8%CSAT lift
- From 3.2 to 3.8 out of 5
Make trust visible
- Problem
- Enterprise customers needed faster answers, but they could not trust a system they could not verify.
- My decision
- I made trust visible through source citations, clear limits, and an assistant that stayed inside the console workflow.
- Tradeoff
- The first release answered documentation questions but did not take actions or use account data. That kept risk low while we proved the core experience.
- Shipped result
- We shipped a multilingual assistant that answered more than 1,200 questions a day in 8 seconds.
The assistant in use
The assistant, end to end
The complete assistant experience inside the IBM Cloud console.
Visualizing an answer
Streaming interface components make a complex answer easier to scan.
Code lookup in context
The assistant finds code and keeps the supporting source visible.
Chat to CLI handoff
A natural-language request becomes a command that can run in Cloud Shell.
Available where the work happens
People could open the assistant without leaving the cloud console or losing their place.
Show the work behind every answer
Start with what people need to trust
Research showed that accuracy, transparency, and reliability came first.
Keep the sources attached
Every answer linked back to the IBM Cloud documentation used to create it.