Prosights Excel Add-in
Sole designer at an AI B2B startup, rethinking how finance teams work in Excel
Prosights is a B2B AI startup helping financial analysts cut hours of repetitive work in Excel. As the company pivoted from a web interface to a Microsoft Excel Add‑in, I redesigned core workflows and introduced a branded Excel AI Chat experience.
My Role
Product designer
Skills
Product Design
Motion Design
Design System
Prototyping
Timeline
Jun-Jul 2025 (1 month)
Team
1 Product Manager, 2 Engineers, 1 Designer (me)
Context
Why move from web to Excel Add-in?
ProSights started as a web-based platform where analysts could upload files like PDFs or PNGs, and convert them into Excel-ready tables and charts. These outputs were downloaded and then opened in Excel for further work.
Feedback from users revealed 2 big opportunities:
1. Remove friction — Let analysts run extractions and create charts without leaving Excel.
2. Unlock advanced capabilities — Enable an AI “agent” inside Excel that can read cell values, understand formulas, and act directly on the spreadsheet’s live context.
The Excel Add-in pivot was the answer: keep analysts in their core workspace while giving them AI-powered tools that feel native to Excel.
Overall Challenge
My role was to design cohesive, brand-aligned, and Excel-native experiences. I identified core usability issues, redesigned flows to match analysts’ mental models, and advocated for clarity and ease of use across the product.
Constraints
• Excel Add-in’s narrow minimum pane width (320px)
• Tight iteration timelines
• Technical limits (response times, etc.)
My work naturally split into two phases:
Design Goals & Success Metrics
Goals
• Adapt core web flows to Excel’s Add-in pane without losing usability.
• Ensure consistent responsive behavior across varying pane widths.
• Strengthen brand recognition in the new AI Chat experience.
• Maintain fast iteration cycles without slowing development.
Success Metrics
• Usability score >80 from analyst feedback
• Strong retention rates across core & AI features
• No broken layouts at minimum pane width (320px)
• AI Chat interactions feel distinctively “ProSights,” not generic
Phase 1: Transitioning from Web to Add-in
Redesign navigation and the flagship flow for Excel’s constraints.
Navigation Redesign
When I joined, the navigation prototype existed but felt confusing and not well-suited for the Excel Add-in environment. It lacked a clear information architecture, didn’t prioritize key actions, and took up more space than necessary in Excel’s narrow vertical pane. I explored several layouts to keep the left navigation as narrow as possible while still making it easy to discover and use.
The solution we implemented used:
• A default narrow bar with only icon buttons.
• Opening the Chat item triggered the chat page and a secondary panel for Chat History, displayed right beside the narrow bar.
• Navigation order and grouping were adjusted to reflect usage priority.
During exploration, I also proposed organizing Chat History by Excel file, making it easier to find past conversations. This wasn’t developed in the release but was documented for future opportunities.
Add-in Navigation
Before
After
Iterations: Through several iterations, I moved from a floating collapsed panel to a narrow icon-based sidebar, ultimately introducing a secondary panel that appears only when AI Chat is opened: streamlining navigation and saving horizontal space.
Extract Table Flow Redesign
Extract Table is Prosights’ flagship feature, allowing analysts to pull structured data from PDFs/PNGs directly into Excel.
Problem
The founder initially asked me to “make it prettier” and fix responsiveness issues, as the UI clashed when the pane was resized. But after trying it myself, I realized the issues ran deeper:
• I often clicked the wrong buttons and wasn’t sure how to proceed.
• A quick usability test with a few analysts confirmed the same confusion and inefficiency.
Goal
• Improve usability and reduce misclicks.
• Ensure responsiveness across varying pane widths.
• Create a more professional, polished experience aligned with Prosights’ brand.
01. Select Document (static)
Before
After: Introduced a new design pattern to prevent from users' learned habit of drag-and-drop due to technical limitation.
02. Select Pages from Selected File
Before
After: Improved selection clarity with distinct “Select All” and “Clear” states, introduced a clear ‘Selected File’ label above it to establish context.
03. Extracting Pages/Processing…
Before
After: Made extraction progress clearer and more intuitive: placing it directly under the uploaded file, simplifying page cards into a vertical status list, and turning the bottom bar into a floating, minimized element that stays visible with clear progress feedback.
04. View/Extraction History
Before
After: Made downloads unambiguous in Extraction History with clear Excel icons and removal of misleading file type names.
Impact
• Made navigation more compact and discoverable, optimizing usability within Excel’s constrained task pane.
• Reduced task completion time and misclicks in the Extract Table flow, making interactions faster and clearer.
• Ensured layouts held up at minimum pane width, improving reliability across varied screen setups.
• Received strong positive feedback from analysts and engineers after release.
Phase 2: Introducing the Excel AI Chat Experience
Built chat layout, custom spinner, and multi-stage loading states.
Laying the foundation for the AI Chat Assistant
With the Add-in foundations established, the next step was launching Prosights’ AI Chat Assistant; a context-aware tool that acts directly on live spreadsheet data. The first use case focused on changing table format settings through natural language, helping analysts style tables to firm standards in seconds instead of minutes/hours.
Problem
1. Perceived slowness — AI responses took several seconds, risking user drop-off.
2. Brand disconnect — The generic interface made the product feel like a commodity, not a differentiated Prosights tool.
Solution
• Built Prosights’ first Add-in design system, defining a consistent visual language and interaction patterns to unify the UI and scale for future features. This also included establishing the foundational chat layout.
• Designed a custom AI loading animation, a morphing mathematical symbol tied to analysts’ core work to keep users engaged during wait states. I sketched concepts, iterated with Claude and Cursor for prompt-engineering, and delivered a working code snippet, which significantly reduced development time.
Designing Multi-Stage Loading States for Build Formula Flow
After establishing the AI Chat foundation, we wanted to solve another core pain point for analysts — the manual effort of writing and linking formulas.
Context
• Analysts spend hours manually creating hundreds of formulas after importing raw financial data.
• We envisioned an AI-assisted workflow where analysts could enter a natural-language prompt, and the system would: 1) detect where formulas should be applied, 2) insert them automatically, and 3) confirm completion, all within seconds.
This would speed up workflows, reduce human error, and position Prosights not just as a data-import tool, but as a core productivity driver inside Excel.
Design guidelines
Before designing, I aligned with the team on four key principles:
• Clarity: Keep interactions simple enough for non-technical users.
• Transparency: Surface progress clearly so users never feel stuck.
• Control: Provide undo options to let users experiment safely.
• Scalability: Ensure design patterns could extend to future use cases, like Q&A or forecasting.
01. Detect → 02. Execute → 03. Confirm & Audit
Enhancing Feedback Between Loading Stages
We noticed short pauses between loading steps that made the interface feel idle. To maintain continuity and user trust, I added a secondary feedback channel above the input field that explain AI's actions in real time.
-> Ensured users always saw evidence of progress, even during short pauses, reinforcing transparency and engagement throughout the workflow.
Iterations: I explored several versions of the confirm stage to make information easier to review, audit, and undo. To prevent visual clutter when the pane resizes, I grouped changes by action type instead of using tags, removed redundant elements, and relied on icon buttons. I also proposed an individual audit feature—initially exploring hover-based icon visibility to reduce visual noise—but we ultimately kept the default design for clearer user feedback.
Impact
What I learned
Working as the only designer on a team like Prosights, where AI was aggressively integrated into the workflow, was tough at first. The pace felt overwhelming, i felt lonely being an advocate for usability and design.
But by proactively identifying usability issues, communicating with them, and showing how design could fix them, I grew a lot as a designer. I also leaned into AI myself, sketching, iterating, and even bringing working code snippets alongside my designs. This experience not only helped me grow as a designer with ownership, but also gave me confidence that I can thrive in fast-moving, AI-driven product environments.
If I could do it differently, I’d spend more time shadowing analysts directly. Most of my understanding came from conversations with the co-founder and engineers. It worked, but seeing analysts in action could have revealed even smaller workflow details.
I’m looking for new opportunities where I can learn, grow, and work alongside thoughtful people. If you’re interested, or just curious to hear more about this project, I’d love to chat! :)
