From 10 Years of Manual Receipt Entry to a 3-Minute AI Review
The Challenge
Every week, 8-12 paper receipts from Polish grocery stores, pharmacies, and other shops needed to be manually read and typed into a Google Sheet. Each receipt took 2-3 minutes to process: reading abbreviated item names, deciding on categories, typing amounts into the correct cells. That added up to roughly 25 minutes every Sunday - and skipping a week meant the backlog doubled. Existing receipt-scanning apps all required switching to their own format, categories, and export structure. The spreadsheet layout - 16 columns across four categories, monthly tabs with Polish names, sum formulas in fixed rows - had been in use for a decade and contained irreplaceable historical data. Any solution had to write directly into this existing format without changes.
Facing Similar Challenges?
Let us help you overcome these obstacles like we did for Personal Project
Get Your Free AssessmentOur Solution
We built a phone-first web app that processes receipts through a two-step AI pipeline. Step one: Google Cloud Vision reads the text off the receipt photo, handling messy thermal print, Polish abbreviations, and tilted camera angles. Step two: Claude Haiku parses the extracted text into structured JSON using tool-use mode, returning items with names, amounts, categories, and group labels. The system writes directly to the existing Google Sheet via the Sheets API, placing each item in the correct column and row of the correct monthly tab. A batch mode allows uploading 15-20 receipts at once, processing them in the background with a file-system queue. Merchant learning tracks category corrections per store and applies them automatically on future receipts. The system also handles Polish-specific receipt quirks like discount lines printed as separate rows and OCR misreading years in dates.
Technologies & Tools
Implementation Timeline
3 Weekends from start to full deployment
The Results
Weekly receipt processing dropped from 25 minutes of manual typing to about 3 minutes of reviewing AI-categorized items.
Over a full year, roughly 20 hours of repetitive data entry work were eliminated entirely.
Claude Haiku correctly categorizes items about 90% of the time out of the box, rising to 95% with merchant learning applied.
Google Cloud Vision free tier (1,000 calls/month) plus Claude Haiku at $0.001 per receipt. No subscription fees.
Want Similar Results for Your Business?
See how we can help you achieve comparable improvements
Start Your TransformationWhat Our Client Says
"I didn't need a new expense tracking system. I needed someone to type receipts into the system I already had. Now I dump a week's worth of receipts, review for three minutes, and the spreadsheet I've relied on for a decade fills itself."
Ready to Achieve Similar Results?
Join Personal Project and hundreds of other businesses who have transformed their operations with AI automation.
Fast Implementation
Get results in weeks, not months
Proven Results
Track record of successful implementations
Expert Support
Dedicated team throughout your journey
Trusted by businesses across industries:
More Success Stories
Discover how other businesses have transformed with AI automation