Automating Budgeting Workflows with AI
Capstone Project
— SEP 2025 - MAY 2026 —
PFF, LLC is a consulting firm that develops tools to help nonprofits, local governments, and small agencies manage their budgeting more effectively. For their emergency management clients, we redesigned InCEP, a legacy budget planning platform, into Treasora, an AI-assisted system that streamlines how mission assignments, cost estimates, and transactions move through a multi-role approval chain.
8 months
My Role:
Product designer
Team:
Team of 5
4.8 / 5
6 users rated the experience highly in usability testing
~ 10%
error rate across core workflows, as seen from usability tests
Imagine you're a financial analyst during a hurricane response. You need to create a Mission Assignment so emergency work can begin fast…
InCEP Platform
Manual data entry slows workflows
Complex approval processes
No visibility into approval status
Instead, you're stuck in InCEP
RESEARCH
—
Competitor's Analysis
Some insights include:
There is an opportunity to eliminate the manual burden of the legacy InCEP platform rather than simply adding an AI layer on top.
If the tool requires extensive training, it will face adoption resistance.
The strongest dashboard patterns share top-level KPIs at a glance with drill-down detail below.
Mapping InCEP Workflows
Some insights include:
Emergency financial workflows are structured around three core entities: Mission Assignment (MA), Cost Estimate (CE), and Transaction.
Cost estimates define how funding is allocated and must be approved before any spending can occur.
Approval is embedded throughout the workflow, governing both funding decisions and actual spending.
Understanding User Needs
Financial Analyst
Requestor
Approver
Director
Organizational Workflow
The current platform treats a multi-role, time-sensitive workflow like a series of isolated forms with no intelligence, visibility, or continuity between steps.
IDEATE
—
How Might We
Faster Creation
How might we help financial analysts create mission assignments and estimate costs without manual data entry?
Informed Purchasing
How might we help requestors track spend against allocation before committing to a purchase?
Confident Approvals
How might we help approvers surface discrepancies and make confident decisions without line-by-line review?
Real-Time Oversight
How might we give directors real-time visibility into budget utilization across all mission assignments?
Design Decisions
Core Principle: AI informs, humans decide.
PROTOTYPE
—
Design Strategy: Focus, Iterate, and Scale
To manage the complexity of a multi-role financial system, we focused on key modules first and iterated from desktop wireframes to high-fidelity designs. This structured approach enabled faster feedback, clearer workflows, and scalable system development.
We made early sketches and wireframes to define core layouts and key workflows. We also explored multiple design directions for different interaction patterns and system structures.
Design System
We built a comprehensive design system from scratch in Figma with 85+ reusable components, including buttons, tables, navigation, and form elements, so that every screen across web and mobile inherited a shared visual language.
TEST
—
We tested the prototype with 6 participants deliberately recruited across a spectrum, from seasoned InCEP users to first-time users with no financial planning background. Testing focused on 4 tasks that mirror the full approval chain: create a Mission Assignment, create a Cost Estimate, approve the estimate, and create and approve a Transaction.
Key Insights
Discoverability & Clarity
Users struggled to understand key actions.
The “Create Cost Estimate” link is not obvious.
Terminology (CE, MA, funding lines) is confusing for new users.
AI Visibility
AI features were valuable but often missed.
AI Review is not prominent.
Trust & Control
Users were unsure how much to trust AI outputs.
Confusion around confidence score & data source.
Users wanted to see the “why” behind confidence scores.
The Approval Chain in Action
Financial Analyst
Create Mission Assignment
Instead of manually transcribing every field, the analyst uploads the source document and reviews what the system extracts.
Financial Analyst
Create Cost Estimate
For cost estimation, AI references similar past mission assignments and suggests funding line amounts.
Approver
Review submissions and approve/reject
AI generates a findings summary ranking every issue by severity.
Requestor
Create Transaction
For transactions, AI suggests spend plan lines and object class codes based on the expense description.
Director
Monitor spending and ensure compliance
The budget dashboard gives the director real-time visibility into spend across every mission.
This eight-month project with a real client, a real approval chain, and real constraints taught me more than any classroom project could. Here's what stayed with me:
Building a design system from scratch taught me that the system isn't the components, but the decisions baked into them. Every token, spacing rule, or state variant is a decision you make once, so you never have to make it again.
Designing for four users sharing one workflow is negotiation. Every screen had to serve the person using it without breaking the experience for the person who touches it next. The hardest design problems weren't visual; they were figuring out what the Approver needs to see that the Analyst didn't know to provide.
AI is easiest to design when you stop asking "where can we add AI?" and start asking "where is a human doing work a machine should handle?" That reframe eliminated half our ideas and sharpened the rest. The features we cut were more important than the ones we kept.
