Rigor Under Ambiguity: Testing Executive Hypotheses in Financial Services
In the race to build agentic AI, the riskiest move is building the wrong agent. This research helped Intapp decide which ones to build first.
Overview
In the race to build agentic AI products, Intapp launched Celeste, an agentic platform designed to help Financial Services, Legal, and Real Estate professionals with their daily and most important tasks. This work directly informed the use cases the agents were built to execute, and which use cases were prioritized for the major Celeste release in mid-July 2026.
Through a series of 1:1 conversations with executives, I noticed a high degree of uncertainty about where Celeste should go within the Financial Services space despite existing hypotheses. This research was born from that uncertainty.
Project Scope
Company; Intapp
Timeframe: 10 weeks
My Role: Senior UX Researcher
Teams Involved: Design, Engineering, Product Management, Go-to-Market, Industry Solutions, Client Success
Methods: Generative Interviews, Survey, Contextual Inquiry
Tools: Figma, Microsoft Teams, Qualtrics, Slack
Constraint:
Participant access: Reaching the right users was the principal challenge. I worked with Client Success and Product to recruit participants and used an external panel through Cint for the survey phase of this project.
Key Question:
What are the most burdensome (challenging and time-consuming) and most important tasks for financial services professionals?
Business Goal:
Invest in agents that drive adoption, retention, and usage.
User Goal:
Identify areas of opportunity to help senior-level professionals in Financial Services complete their tasks more effectively and efficiently.
Research Phases and Key Insights
Phase 1: Generative Interviews
Goal: Identify the most burdensome and important tasks for financial services professionals, without anchoring on leadership’s hypotheses, and understand why those tasks mattered and where they broke down.
Participants: 11 professionals from Intapp customer firms
Key Takeaways:
Blind spots surfaced: Two tasks outside leadership’s original hypotheses emerged as potential revenue opportunities: financial modeling and forecasting, and client communication management.
Expectations for AI: We didn’t ask about AI directly. It came up naturally toward the end of conversations as participants talked about their workflows. When it did, they expected AI to:
Fit seamlessly into their existing workflows
Include checkpoints where they could validate the output
Help with parts of a task rather than take over the whole thing
Phase 2: Survey
Goal: Test leadership’s hypotheses and find blind spots by identifying which tasks financial services professionals found most burdensome and most important.
Participants: 132 financial services professionals (105 non-customers, 27 Intapp customers)
Key Takeaways:
Hypotheses confirmed: Complex workflows, such as compiling quarterly fund performance reports and analyzing due diligence reports, were rated the most burdensome.
Blind spots validated: Two tasks that weren’t in leadership’s original hypotheses but came up in Phase 1, financial modeling and forecasting and client communication management, emerged as new revenue opportunities.
Phase 3: Contextual Inquiry
Goal: Observe how professionals actually handle the reporting and client communication workflows identified in the survey.
Participants: 8 client-facing professionals
Key Takeaways:
Quarterly reporting: Drafting documents and handing them off for revisions were the most tedious parts of the workflow.
Client communication: High email volume made it easy for messages to get lost, which put client relationships at risk.
Deliverables:
Research findings decks with recommendations
A trust framework adopted by multiple internal stakeholders
How I Used AI: Initial survey question drafting and initial survey and call transcript analysis.
Outcomes Tied Back to the Challenge
Created a trust framework that became a design system standard: The framework moved beyond a research artifact and became a standard teams used to design AI experiences.
Identified 5+ new use cases for Celeste: The research surfaced AI opportunities the team hadn’t considered, including quarterly fund performance reporting and financial modeling and forecasting.
Shaped product, go-to-market, and roadmap decisions: Findings informed the Celeste product experience, GTM and sales messaging, and how the team prioritized what to build.
What I’d do differently: Test AI concepts sooner. Participants raised clear expectations for AI on their own, like checkpoints to validate output and help with parts of a task rather than the whole thing.