Case Study
Details
Category
AI
Enterprise
Fintech
Type
Product Design
Team
5 Designers, Engineers, CEO
Duration
4 months
Norric helps private market teams organize documents, answer DDQs, and collaborate with AI-assisted workflows.



[01]
overview
What is a DDQ?
A Due Diligence Questionnaire is a standardized set of questions that investors send to organizations before making an investment
Institutional teams spend weeks responding to DDQs, RFPs, and compliance requests. Critical information is often buried across spreadsheets, PDFs, and historical submissions.
Our team designed an AI-powered platform that transforms fragmented documents into clear workflows.
[02]
problem
Manual DDQ workflows put accuracy at risk
Time-consuming preparation
Teams look through past documents, verified answers, and coordinated reviews across sources.
Unclear accountability
Outdated answers, lost knowledge, and unclear approval histories make information hard to trust.
Inaccurate responses could create compliance risks and damage investor trust.
[03]
Audience
Personas
Investment Analyst
Reviews investment information and prepares detailed responses across multiple DDQs.
Needs: Quickly find reliable information and verify answers without searching through hundreds of documents.
Investor Relations Manager
Coordinates DDQ submissions across teams and manages investor communications.
Needs: Track progress, assign responsibilities, and ensure submissions are complete and consistent.
Compliance & Risk Professional
Reviews regulatory information and evaluates the accuracy of submitted responses.
Needs: Trace answers to their sources, identify potential risks, and validate AI-generated information.
[04]
early iterations
Flow Exploration
Exploring workflows through rapid iteration
As product requirements evolved, we mapped and explored workflows across onboarding, project creation, data rooms, integrations, and AI interactions.
Rather than refining each screen immediately, we used early flow explorations to identify dependencies, clarify interactions, and align on the evolving product direction.

Flow explorations in a previous version
[05]
solution
Key Product Flows
Question Matching
We designed question matching to surface relevant answers alongside the active question, with confidence levels, answer previews, and AI-generated reasoning. Rather than automatically filling in responses, we gave users the context to evaluate each match before deciding whether to reuse it.
DDQs often contain questions that have already been answered in previous submissions. Reusing those answers could save analysts time, but similar questions don't always have identical responses.
Collaborative Review
DDQs often involve multiple stakeholders across legal, compliance, and investment teams.
We designed workspace-level collaborator management so users could easily assign and remove team members as responsibilities changed. A shared comment section kept feedback and discussions in one place.
Institutional Memory
DDQs are not one-off tasks. Each submission builds on years of organizational knowledge, decisions, and evolving information.
We designed a searchable history panel where users could revisit past AI conversations, compare historical context, and trace how information evolved over time.
AI as a Thought Partner
Complex DDQ workflows require more than answering questions. Teams also need to understand what deserves their attention and what actions to take next.
We designed our AI assistant to provide contextual guidance, suggested prompts, and actionable next steps to help users navigate complex workflows and make informed decisions.
[06]
reasoning
Design Decisions & Tradeoffs
Across these workflows, we explored how much responsibility AI should take on in a high-stakes financial environment. Rather than maximizing automation, we made deliberate decisions about where AI could add value and where human oversight was essential.
Make AI confidence interpretable
A strong question match doesn't necessarily mean the underlying answer is accurate or up to date.
Instead of treating AI confidence as proof of correctness, we made its reasoning visible so users could evaluate recommendations in context.
What we gave up
Frictionless acceptance of AI recommendations
WHAT WE PRESERVED
Transparency and informed decision-making
Introduce intentional friction
We could have designed AI to automatically populate DDQ answers, but chose to preserve a manual review step.
Users had to explicitly select and copy suggested responses into their questionnaire. While this added an interaction, it created a deliberate checkpoint for verification and human accountability.
What we gave up
Full automation and fewer interactions
WHAT WE PRESERVED
Human verification and accountability
Give AI structure, not complete autonomy
We chose a structured Master DDQ alongside the AI agent, rather than relying entirely on open-ended generation.
Reusable answer templates provided a consistent starting point that AI could adapt to different contexts. This also gave users a predictable workspace to review, edit, and manage information independently of AI.
What we gave up
Unrestricted AI-generated responses
WHAT WE PRESERVED
Consistency, predictability, and user control
[07]
scalability
Design System
Building consistency through a changing product
With over 10 design iterations and four revisions of the design system, maintaining consistency became a challenge.
The goal wasn't to finalize a rigid system, but to create a flexible foundation that could adapt to changing requirements.

Component explorations establishing consistent UI patterns
[08]
measurements
Impact
Design Output
10+ design iterations across 4 months
10+ product flows covering onboarding, project management, data rooms, AI interactions, and integrations
4+ design system iterations to refine components, patterns, and consistency as the product evolved
Success Metrics to Validate
30% reduction in time spent finding relevant answers
25% reduction in DDQ question completion time
90%+ accuracy in identifying unsuitable AI recommendations during testing
These are proposed targets, not measured outcomes. Further user testing is needed to validate the product's impact.
[09]
conclusion
Reflection
Learning to design through ambiguity
Working at an early-stage startup meant navigating constantly changing requirements and multiple design iterations within a short period.
One of my biggest takeaways was learning to separate what AI could automate from what users actually needed. It pushed me to question assumptions, consider tradeoffs, and make more intentional decisions rather than simply designing for more features.
If I could take this further, I would prioritize research with actual investment analysts to validate our assumptions, particularly around AI trust, answer verification, and how people navigate complex DDQ workflows.