Business owners kept calling for features they already had. So we brought the features to the conversation.
Chase's AI assistant for business banking customers wasn't closing the gap between what owners asked for and where those features lived in the app. In six weeks, I designed 40+ conversations that connect each request directly to the right feature.
- Role
- Conversation designer
- Engagement
- Accenture consultant, embedded at JPMorgan Chase
- Timeline
- 6 weeks
- Tools
- Voiceflow
- Team
- Chase conversation design + UX/UI teams
Support calls were telling us where the app broke down.
Business owners were calling customer service, or getting stuck with the AI assistant, to ask for things the Chase app could already do.
For an owner, every call is time away from running the business. For Chase, it's a cost and a sign that features aren't being found.
How might we turn the assistant into the fastest route from a question to the feature that answers it?
I owned the conversations, from data to Voiceflow.
- Analyzed support calls and chat logs to find what business owners couldn't find
- Defined and prioritized 40+ intents
- Designed flows, sample dialogs, utterances, deep links and fallback paths
- Partnered with Chase's conversation design and UX/UI teams, reporting to a Chase manager
- 01Data researchCalls + chat logs
- 02Synthesis40+ intents, 4 themes
- 03PrioritizeVolume × friction
- 04Conversation designFlows in Voiceflow
- 05HandoffReady to ship without me
40+ intents, four themes.
We sorted every request from the calls and chat logs into intents, then grouped them by what owners were trying to get done.
Moving money
- "How do I send a wire?"
- "Set up an ACH payment"
- "Zelle for my business"
Managing access
- "Add an employee"
- "Change what my bookkeeper can see"
Finding documents
- "Download last month's statement"
- "Where are my tax forms?"
Account maintenance
- "Update my business address"
- "Order checks"
- "Lock my card"
It wasn't a missing-feature problem. It was a findability problem.
Two calls that shaped every conversation.
Should the assistant complete tasks in chat, or send owners to the app?
The call: it was faster to ship, sensitive steps stayed in the app's secure flows, and every link taught owners where the feature lives.
What happens when the assistant doesn't understand?
The call: no dead ends. Every path ends at a feature, an option or a person.
The happy path and the recovery path.
All mockups recreated to respect client confidentiality.
Designed, handed off, launched after my engagement.
40+ intents ready to ship
Flows, utterances, deep links and fallbacks clear enough for the Chase team to build without me. [Add sign-off or manager feedback]
Built to move clear metrics
Fewer support calls, higher containment, deep-link click-through and a lower fallback rate.
Insight beyond the assistant
The intent data showed where the app itself was hard to navigate, which is useful input for the UX/UI team.
"As a consultant, you rarely see launch day. So I design for the handoff: work clear enough to ship without me in the room."