Overview
The goal of the POC was to explore ways to enhance its functionality with LLMs & natural language AI.
With a natural language model backing the project my challenge was to create an intelligent search that could solve existing challenges faced by users of Story™.
The prototype I designed served as the blueprint for our search feature today in both internal and consumer facing products. It has been adopted across 4 different products serving 120,000 users worldwide.
To avoid spilling “secret sauce”, sensitive information has been omitted. All views are my own and may not reflect those of JPMorganChase.
92%
Search success rate
8 sec
Average time to result
41%
Reduction in help center tickets
Source: Q2 2025 performance metrics.
Problem
This was the research question we started with. What should search be capable of, and how do we design it to scale across internal & consumer facing JPMorgan products?
Challenge I
Challenge II
Problem, continued
Quick wins vs user value
I hosted two whiteboarding sessions with the team to narrow down our areas of focus for search. We decided to pick problems that fit the criteria of (1) having great immediate value, and (2) being deployable quickly.
Many of the problems our research team identified came from friction in our core product experiences — a perfect application for a quick search to solve.
Exploration
I was very happy we started with defining happy paths as they helped us see how search might fit into the holistic experience.
Design
Placement of the search
We tackled cross-product scalability by embedding the search within the global nav, an element present in all of our products.
This also made search accessible from anywhere — a requirement, since user friction could appear anywhere.
Quick wins vs user value
I hosted two whiteboarding sessions with the team to narrow down our areas of focus for search. We decided to pick problems that fit the criteria of (1) having great immediate value, and (2) being deployable quickly.
Many of the problems our research team identified came from friction in our core product experiences — a perfect application for a quick search to solve.
Displaying a launch menu upfront
We decided to group possible search queries into buckets, and make them visible on launch for inspiration. Each bucket solves one of the four user problems we decided to solve with our search.
Final Design
Outcomes
Source: Q2 2025 performance metrics.
More work
