Sort receipt and invoice photos into expense categories
Staff snap receipts on their phones, and a decision model reads each photo, picks one expense category from your chart of accounts, and sends only the unsure ones to the bookkeeper for a decision.
Uses image input, which TypeSafe AI labels a preview feature for Jev. Test it on your own sample images before relying on it.
The use case
Receipts arrive as phone photos in a shared folder or inbox, and someone has to look at each one and pick a category before the books can close. The decision here is narrow: which one of your fixed expense categories does this receipt belong to, such as meals, fuel, office supplies, software, travel, or other. Jev looks at the photo and returns one category with a confidence score. Receipts at or above a threshold you set, for example 0.85, are filed into that category in a review sheet. Anything below it, plus anything labelled other, lands in a short queue for the bookkeeper. The bookkeeper still decides how each item is posted, whether it is reimbursable, and whether a receipt looks wrong. Jev only sorts; it never approves a reimbursement.
Example workflow
- Staff upload receipt photos to a shared Google Drive folder or forward them to a receipts inbox in Gmail.
- A Zapier or Make scenario picks up each new PNG or JPEG and sends it to Jev with the image input preview header turned on.
- Jev answers one Choice question, which expense category fits this receipt, with the allowed answers meals, fuel, office supplies, software, travel, and other.
- Answers at or above the confidence threshold are written to a Google Sheet with the photo link, the category, and the confidence score.
- Low-confidence answers and anything marked other are posted to a Slack channel for the bookkeeper to categorize by hand.
- The bookkeeper reviews the sheet weekly, corrects any wrong categories, and posts entries to the accounting system themselves.
Tools
- Jev (TypeSafe AI)
- Google Drive
- Gmail
- Zapier or Make
- Google Sheets
- Slack
- QuickBooks Online
Risks and guardrails
- Image input is a preview feature in Jev 1.14 and TypeSafe says it is not production-ready yet — guardrail: test it on a few hundred of your own photos against a person's labels before relying on it, and keep a person checking every answer until the results hold up.
- Faded, crumpled, or cropped receipts can be misread — guardrail: a Noul question asks whether the receipt is legible, and anything below the threshold goes to the bookkeeper.
- A wrong category can flow into tax reports — guardrail: Jev writes only to a review sheet, and the bookkeeper posts every entry.
- Photos may show card numbers or personal details — guardrail: restrict the folder to finance staff and delete originals on your normal retention schedule.
A similar pattern in a public build: https://madewithjev.com/builds/invoice-accounting. See also Decision models.
Related recipe
Receipt Capture and Categorization for Bookkeeper Review: An agent extracts vendor, date, amount, and tax from receipts in email, Drive, or photos, suggests a category from your own chart of accounts, and queues drafts in QuickBooks or Xero for a bookkeeper to approve.
Want it set up for you? Request a quote.
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Public build with a similar pattern: madewithjev.com/builds/invoice-accounting
- Company invoices, classified for the books: original post or repo by iagolast: x.com/iagolast/status/2100694100160745746
- Google Drive (official page): developers.google.com/workspace/drive (checked 2026-11-19)
- Gmail (official page): developers.google.com/workspace/gmail (checked 2026-11-19)
- Zapier MCP (official page): zapier.com/mcp (checked 2026-11-09)
- Make MCP server (official page): developers.make.com/mcp-server (checked 2026-11-09)
- Slack MCP server (official page): docs.slack.dev/ai/slack-mcp-server (checked 2026-11-09)
- QuickBooks Online MCP server (official page): github.com/intuit/quickbooks-online-mcp-server (checked 2026-11-09)