Discovery
Problem Statement
How can we design a simple, powerful interface that leverages generative AI to enable finance professionals to interact with complex data using natural language without needing to understand the underlying queries or system architecture?
WHY IS THIS A PROBLEM?
The emergence of ChatGPT in the market sparked interest among businesses seeking innovative solutions for automating the traditionally manual and cumbersome processes of data analysis and insight derivation. Traditional enterprise performance management tools require users to know where to look and how to extract insights from their ERP systems. They involve dashboards, filters, and formulas that are second nature to analysts but intimidating or time-consuming for less technical users. A chat interface could lower this barrier dramatically.
RESEARCH
We began by studying leading conversational AI systems of the time(e.g., ChatGPT, Google Bard) and noted key interaction principles in a competitive analysis. Interaction principles such as progressive disclosure, inline clarification, and recoverability informed our early interface direction. To design a chat-based interface that truly served financial analysts, we recognized the importance of combining industry standards, user expectations, and technical feasibility.
👥
Informal Interviews
Interviewing analysts to understand current workflows, frustrations, and expectations
📓
EPM Workflow Analysis
Analysis of report generation in Oracle EPM through EPM documentation review
🔎
Documentation
Analyzing consumer and enterprise-facing AI tools to identify interaction patterns
💻
Code / Queries
Analyzing REST integrations and EPM capabilities for integration design and logic
We framed our research results around three pillars: how users currently interacted with dashboards and data tools, how they trusted and adopted AI outputs, and what could make enterprise conversational interfaces successful. This helped us identify the needs of our key users.
Financial Analysts needed Speed
Analysts needed to ask quick questions and get actionable responses without sorting through r raw data. During initial research, the ability to type natural language prompts and receive structured output was highlighted as a benefit during month-end and reporting cycles.
Executives Needed Clarity
Leaders wanted top-level summaries with just enough detail to make fast, informed decisions. It was crucial that the AI output included clear data references and confidence indicators, so executives trusted and acted on the insights without a second round of validation.
Planning Teams Needed Consistency
Planning teams needed a way to incorporate AI insights into existing planning flows. From file uploads to custom queries, the system had to support repeatable templates, allow multiple data sources, and provide consistent language across responses.
DETAILED REQUIREMENTS
Based on our exploratory research and initial requirements, the solution needed to use a chat-based interface where users could input questions in natural language. It also required the ability to save frequently asked queries as templates, enabling users to rerun them without retyping.
Additionally, users had to be able to select relevant data sources and apply filters so responses would be grounded in clear, traceable context. The tool also needed to support file uploads—such as reports or documentation—which users could query directly. This extended its utility to both structured database connections and ad-hoc or legacy content.
At its core, this was an MVP design exercise focused on speed, simplicity, and confidence. Rather than building a full-featured product, our goal was to deliver a usable, testable version of a chat-based insight tool that could evolve with feedback and real-world use.
Key Insights
1
Key Insight 01
Without knowing what database, filters, or context was applied, users would likely mistrust or misinterpret AI-generated outputs. Maintaining transparency about selected datasets and parameters helps build credibility and clarity.
2
Key Insight 02
Many enterprise queries are repetitive. Being able to reuse them as templates, prevents rework and supports consistency across teams.
3
Key Insight 03
Especially in a complex financial environment, having sample prompts or hover-based help makes the tool easier to approach and reduces onboarding time.
Outcomes
~75% reduction
in Time-to-Insight
Based on a benchmark taken during researching report extraction and insight synthesis process in traditional BI systems, the new system allowed users to extract relevant insights from data in minutes
2-step reduction
in support dependency
Early feedback suggested that non-technical users were able to complete insight queries independently without assistance from analysts or IT support, thanks to the intuitive interface and context-aware logic.
This project marked my first hands-on experience designing for an AI-driven product. I learned how critical it is to manage user expectations in generative systems and ensure traceability in AI outputs, especially within enterprise environments. Working on the integration between GPT and structured financial systems gave me a new appreciation for the complexity of data mapping and integration. It also deepened my confidence in designing intuitive interfaces that translate complex tools into usable, everyday workflows.
Overview
PROJECT OVERVIEW
The emergence of ChatGPT and large language models in 2022 presented exciting opportunities for the firm and the practice as a whole to rethink how enterprise users interact with data. This initiative was one such exploration, aiming to create an intelligent FP&A assistant powered by GPT-4, capable of querying Oracle EPM data and generating actionable insights through conversation.
I was brought in to lead the interface design, ensuring that the powerful backend solution translated into an experience that was not only functional but also familiar, intuitive, and efficient. My work focused on developing the full user interface, prototypes, visual language, and usability guidelines for the system.
Specifics of this project are confidential and cannot be disclosed. However, I have provided a basic overview of work I have done as part of this project. All assets displayed here-in are generic representations.
CLIENT OVERVIEW
This was an internal innovation initiative driven by a cross-functional Deloitte team. It sought to explore GPT integration into financial workflows and define patterns for future AI-enabled enterprise tools. The primary stakeholders included senior leadership from both India and USA, internal FP&A professionals, data engineers, and developers who needed to evaluate the concept for further adoption.
MY ROLE
I was responsible for shaping the design of the interface, developing high-fidelity wireframes in Figma, and defining the interaction flows that enabled seamless querying of financial data.
Beyond UX, I also collaborated closely with developers to define the architectural framework for integrating Oracle EPM data with GPT-4 through REST APIs, ensuring contextually rich and secure interactions. I played a key role in usability testing, iterating based on feedback, and ensuring the interface met both business and technical needs.
TOOLS & PLATFORMS
🛠️
Core Platforms
Oracle EPM
Open AI API
REST API
🧠
User Research
Figma
Excel
Teams
Design
ARRIVING AT THE MVP
The design process began with a focus on simplicity, familiarity, and functional clarity. Drawing from popular conversational tools like ChatGPT and Bard, we identified design patterns that supported efficient dialogue, reduced friction, and helped users feel in control of the interaction. We also focused on integration with databases and creating a consistent workflow for the user querying the data. The solution used GPT to identify contextual portions of the database, and generate answers to the user's question using contextual data from the database or the files uploaded to the system.
File Upload and Integrated Querying
To support unstructured input, we added a drag-and-drop file upload zone above the chat. Uploaded documents were parsed by the system and indexed for reference. When a user asked a question, the interface showed whether the AI response was drawn from a file or database, increasing trust.
Conversation Flow & Layout
I structured the interface with a persistent chat window, response bubbles, and a dynamic input bar. To maintain clarity during multi-step queries, I incorporated visual dividers and collapsible side panels where users could view source or context information.
Template Management
Given the repetitive nature of enterprise questions, we introduced a lightweight template system. Users could save frequently used prompts and select from a dropdown of saved queries. The UI enabled reusability and quick access to high-priority prompts.
PROTOTYPE
Built and prototyped in Figma, the design showcased a cohesive and task-focused conversational system, starting from login and authentication through to insight generation. From quick-switch database controls on the home screen to intuitive prompt suggestions, the layout was optimized to encourage engagement from both technical and non-technical users. These representative visuals offer a quick overview of how the system worked across its major touchpoints, from login to insight generation.
Each frame above highlights a distinct part of the experience: from seamless SSO login, to a home screen with preloaded prompts, to query differentiation, document upload management, and template editing. These modular designs supported fast MVP development and easy iteration during future phases. This approach allowed us to deliver a simple but functional MVP that could be tested and iterated on quickly. The end result was a focused, usable tool that:
-
Enabled conversation-driven access to financial data
-
Minimized user friction through familiar layouts and defaults
-
Provided a foundation for scalable enhancements as needs evolved
Designing an Intuitive Interface for a custom GPT-driven Enterprise Data Analysis Solution to generate Insights
Figma / Oracle EPM / REST API
Project Snapshot
-
Designed an intuitive chat interface for a GPT-4-powered financial data analysis tool designed to integrate with multiple leading enterprise performance management tools.
-
Enabled real-time querying of Oracle EPM datasets via a conversational UI that allowed users to in effect "talk" to their data.
-
Designed a conceptual integration framework exploring how enterprise finance data could interface with GPT in a secure, context-aware way through REST API.
-
Recognized by the firm with an award for design contribution
Client/Firm
Deloitte
Role
UX Designer
Timeline
2 months
Team Size
~5
Development
As an EPM SME, I was also an integration developer on the initiative responsible for the integration architecture of the tool with the EPM platform. The integration architecture was built to support an end-to-end flow of conversational querying and AI-generated insights. The system was designed to identify relevant data points in response to prompts, enabling contextual querying through AI.
This development approach prioritized a balance between speed and adaptability, making it possible to rapidly test a usable AI interface while laying the groundwork for further capabilities. Validating the solution with stakeholders through workshop sessions helped us better align to their expectations while creating an MVP that was ready to iterate upon in upcoming iterations of the product and identify scope for future features and development.
