Finding financial information shouldn’t require financial expertise.
An AI search experience that helps investment teams find answers in their financial documents and inspect the sources behind them. I co-led design, turning user research into a familiar way to ask, verify, and carry information into their work.
A simple question could send someone all over iLEVEL.
What are the most impactful updates in a portfolio company’s board pack? To answer a question like this, front-office users clicked through different parts of iLEVEL and compiled information in Word or PowerPoint. When they couldn’t find what they needed, they turned to platform admins.
That cost both people time and made access to information depend on expertise with the platform. AI Doc Search aimed to let users ask questions directly of their financial-document library.
First, understand what people needed to ask.
Before exploring the interface, we brought together user interviews, internal subject-matter expertise, and workshops to understand the questions people wanted answered—from administrative staff with less familiarity with iLEVEL to experienced power users.
We organized this “question universe” into four groups: summarizing key takeaways, referencing datapoints, finding documents by content or metadata, and comparing information across documents.
Those needs informed the empty-state prompt examples, alongside the limitations of the technology available to us. The examples introduced useful starting points while helping set expectations for what AI Doc Search could do.
What should asking a document look like?
In a collaborative Crazy 8s session, we explored search-style results, chat, document cards, highlighted passages, and side-by-side review. We were deciding how people would move between a question, an answer, and the material behind it.
Team judgment and user feedback shaped the direction toward a conversation with expandable citations and access to the full document.
Test the whole journey, including what happens after the answer.
I co-led the design with another designer and took the lead on communication and facilitation. I planned and facilitated all usability sessions, led synthesis, presented the insights, and worked with Product to translate them into design decisions.
Our April 2024 prototype study involved five participants—a mix of client and internal users. We explored entry points, follow-up questions, source verification, copying, downloading, and sharing.
Where would you start?
We avoided the words “search” and “chat” when asking participants where they would go, so the prompt wouldn’t point them toward a control.
What would make you confident?
We asked about confidence in the response and what would increase it, then explored citations and the source documents.
What would you do next?
We explored how an answer would enter their work, including expectations for copying, downloading, and sharing.
Inside the test scenario
Participants explored a prototype question about mobility features Uber introduced in 2023, refined the question, and checked the supporting documents. The script used this scenario to evaluate the interface and flow; the board-pack question above illustrates the real-world need.
Start where people already go to ask.
We tested two entry points: a sparkle icon and the existing top search bar. Participants described the search bar as their natural starting point for questions about information in iLEVEL.
the existing search bar
A familiar starting point for a new capability.
This preference guided the entry-point direction. The next challenge was making the scope of AI Doc Search clear.
Make the useful possibilities clear.
There was strong pressure to bring an AI product to market, while the technology available to us still had limits. I wanted the experience to be clear about what it could help with.
The empty state translated the discovery themes into concrete examples: find a datapoint, find a document, identify trends, or summarize. We shaped these examples around both user needs and the technology’s limitations. A short scope note explained that answers came from the document library. That mattered because users saw the search bar as a place to ask about anything in iLEVEL.
Users wouldn’t use an answer without evidence.
We began with the instinct that source traceability would be essential. Testing confirmed it: participants said they wouldn’t use an answer without supporting evidence.
The shipped experience connected answers to numbered citations. Users could expand a relevant passage within the conversation or open the full document alongside it. This gave them ways to inspect the evidence while keeping the answer in view.
Make the answer easy to take with you.
Users also wanted to copy the answer into the work they were already doing. They wanted clean text, without AI branding or watermarks.
Copying answers shipped alongside the source-review interactions. Together, they supported two distinct needs: checking the evidence in iLEVEL and reusing the answer in an existing workflow.
A shipped experience. A clear limit.
AI Doc Search shipped, including inline source previews and answer copying. Users were excited about its potential to make their work easier. After launch, research showed that underlying AI output quality limited the value it could reliably deliver.
I helped turn an open-ended AI initiative into a defined experience, using research to align the team around how users would enter, verify, and reuse answers.
The design could give people a way to check an answer. The answer still had to be worth using.