Claude Cowork turns 47 trip receipts into Excel
Hassid reports a 47-receipt expense workbook with three unclear values flagged for manual entry, completing the process in his estimated ten minutes.
Original prompt
Shared by the source author. See source notes for context.
Create expense report. Categorize by type. Include date, vendor, amount. Calculate totals.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Ruben Hassid
- Original account
- Read the source ↗
- Published
- 4 Feb 2026
- Source link checked
- 15 Oct 2026
- Reported result
- Hassid reports a 47-receipt expense workbook with three unclear values flagged for manual entry, completing the process in his estimated ten minutes.
- Human involvement
- Hassid fills the three flagged cells and retains review of the reimbursement data. Private source receipts are not independently audited.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Place 47 work-trip receipt images into the Cowork workspace.
- Extract date, vendor and amount, categorize expenses and calculate totals in Excel.
- Flag three blurry receipts and have the author complete the missing cells.
Ruben Hassid tested Claude Cowork on 47 receipts from a work trip. He placed the receipt images in a folder and requested an expense report with dates, vendors, amounts, categories and totals. The task required reading multiple files and creating a usable spreadsheet rather than returning a summary in chat.
Cowork produced an Excel file with the receipts logged and the requested fields extracted. It did not silently fill every uncertain value. The author reports that three receipts were too blurry to read clearly and were flagged for verification. He supplied the three missing cells manually.
Hassid estimates the complete process took ten minutes, compared with his expected manual effort. That is his personal observation, not a controlled benchmark, and he does not publish a recording because of the sensitive material. The resulting workbook and correction step establish the reported outcome, while the private inputs prevent an independent row-by-row audit here. The useful boundary is the explicit handoff for illegible documents: an incomplete field becomes a review task rather than an invented number that could flow into a reimbursement total.
Prompt
Original prompt
Source: Blog · Ruben Hassid
Frequently asked questions
How many receipts were processed?
The author reports 47 work-trip receipts.
What happened to blurry receipts?
Three were flagged for verification, and Hassid filled in the corresponding cells manually.
Is the ten-minute figure a benchmark?
No. It is the author’s estimate for this private receipt task.
More details
About this account
Cowork extracted receipt fields, assigned categories, calculated totals and flagged three unreadable records in an expense workbook.
Firsthand account read. Original Substack article visibly dated Feb 04, 2026. Private inputs; timing is a personal estimate rather than a controlled comparison. Reported outcomes were not independently reproduced. Original prompt box contains an exact short excerpt of the published task prompt.
Credit: Ruben Hassid
Source published 4 Feb 2026
Source checked 15 Oct 2026
Tools and topics
Claude Cowork · Receipt images · Excel
Claude Cowork · Receipt images · Excel
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.