Steward — Treasury Intelligence
Reduced monthly treasury work from 2-3 hours to 15 minutes
The Business Challenge
As Treasurer of Jacksonville Southpoint SDA Church, I lead a treasury assistance team responsible for counting collections, managing deposits, and recording accounting entries. My personal responsibility extends beyond the team — I'm accountable to the Board for comprehensive financial reporting.
Board reporting was a tedious manual process: re-entering all bank transactions, all accounting system transactions, and reconciling from fulfillment. Every month, this consumed 2-3 hours of meticulous manual work. The process was not only time-consuming but error-prone and offered no historical perspective for financial planning.
Discovery & Collaboration
The discovery process focused on understanding what the Board actually needed versus what they were getting. The existing reports were flat — numbers on a page with no trend analysis, no ministry-level breakdowns, and no ability to ask questions about the data. I identified that the real value wasn't just automating the existing report, but building an intelligence layer that could transform how the church leadership understands its finances.
I also recognized that an AI-powered chat function would allow Board members to ask natural language questions about financial data — comparisons between ministries, trend analysis over periods, and planning questions — without requiring them to become spreadsheet experts.
The Strategic Decision
I chose to build a Streamlit application with AI integration rather than a more complex web app because the primary users are Board members who are not tech-savvy. Streamlit provides a clean, simple interface that feels approachable. The AI chat function uses Claude to enable financial planning questions in natural language.
The auto-categorization system was designed to improve over time: starting at 75-80% accuracy and improving to approximately 95% through a learning loop. The remaining 5% are manually adjusted during the upload workflow — a deliberate design choice that keeps a human in the loop for edge cases.
The Build
Tech Stack: Python, Streamlit, Claude AI, Make.com, Smartsheet
Key Features:
- Upload transactions — app processes and matches all transactions automatically
- Scrubs out personal data
- Reports at ministry level and organization level
- Three years of history for trend analysis
- AI chat function for financial planning questions
- Standard evaluation questions for the most recent month
- Chat supports general questions about particular ministry, time period, or comparisons
- 75-80% auto-categorization initially, now improved to ~95%
The Outcome
- Monthly treasury work reduced from 2-3 hours to 15 minutes
- Auto-categorization improved from 75-80% to ~95% over time
- Board initially hesitant (members are not tech-savvy) but after using the tool, feedback was overwhelmingly positive: the system "really promotes good transparency from our Treasury operations"
- Change management success: hesitance to appreciation
- Three years of historical data now available for trend analysis and financial planning