I redesigned the interaction model for Rocket's AI assistant and the patterns I proposed made it to the product roadmap.
Handed off
Aug 2025
Industry
Fintech
Role
Product Design
Team
Conversational AI Designers, Product Designers
TL;DR
Rocket Mortgage's AI assistant guides homebuyers through the mortgage process, but the guidance inside the assistant was largely generic and disconnected from each client's specific situation.
I redesigned the post-offer onboarding experience inside Rocket Assist and introduced new interaction patterns like task cards, document components and a transparent human handoff flow.
The redesign was tested with Rocket Mortgage clients and clients reported that the context-driven conversation and new experience was more helpful and reduced frustration.
92%
found the tailored responses were helpful
75%
said customized recommendations were trustworthy
90%
found transfer option reduced frustration
Generic chat experience at a crucial stage was causing client frustration
Generic chat experience
Broken human handoff flow
The post-offer stage is one of the most emotional and delicate part of the home-buying journey. The clients have just had their offer accepted after many hurdles and what they require is a clear guide on what comes next. This was missing in the current chat experience which was not personalized and gave the same common and generic advice to every home-buyer, irrespective of what stage they were in.
In delicate situations where the client needed a human intervention, the handoff flow was also broken. Actionable next steps that could be taken from the chat were misising.
When I picked up this problem space, the common thread between all the client feedback was the same-
“I'm buying a house; it's a lot of money - it should really be a guided journey”
“It's an authenticated experience so it has my data, but chat historytells me otherwise.”
So what was the missing layer? There were actually three.
Context-aware & personalized conversation
The most common question in the chat logs was some version of 'what do I do next?' I moved the answer into the conversation itself. An onboarding flow tied to the client's specific loan stage, with the most relevant next steps surfacing as task carousels inside the chat.
Personalized greeting with pictures of their home
Next steps in a more visual and interactive form
Interactive in-chat experience
Once clients had a task in front of them, the next gap was the information needed to act on it, inspection reports, appraisal documents, etc. All of this lived outside the Rocket Assist.
I introduced visual components that brought it inside: inspector cards personalized to the client's home, document cards with highlighted insights and prompt buttons that surfaced the next logical action without requiring the client to ask.
In-chat inspector contact cards
Important documents accessible within chat
Transparent & smooth AI-Human handoff flow
The conversation logs showed a consistent pattern where clients would reach a point where they needed reassurance that only a person could give. Rather than treating this as a failure state, I designed the handoff as a feature.
Clients could transfer directly to their Purchase Specialist, see the estimated wait time and schedule a callback if needed. The conversation history carried over so the specialist could pick up exactly where the client left off.
Clear communication on agent availability
Context retention with human handover
What the people who saw the work up close had to say
“This was perhaps her most complex assignment, and Pri quickly mapped key friction points while collaborating with engineers and researchers. Her work helped influence product roadmap priorities.”
Dana Lee
Director of Conversational AI Design & Digital Product Management
“Pri routinely sought out and addressed challenging issues, independently identified critical opportunities for improvement, and delivered results on par with a full-time associate designer.”
Amanda Matzenbach
Conversational AI Design Manager & Mentor
Some things I learned during my internship
01
Thinking in consequences
A PM breaking down every feature into APIs, infrastructure constraints and engineering dependencies reframed for me what good design means in a mature org. A feature is ten decisions that all need to align first.
02
Collaboration as a design tool
Seeking out engineers, researchers and SMEs before being told to is what got me into the conversations that shaped the work most. Waiting to be pointed in the right direction would have been the wrong instinct here.
03
Business needs vs client needs
The inspector card feature scored high in the usability testing but was deprioritized because the backend architecture wasn't within the team's bandwidth. For me it was a business needs rather than a design failure.