Turn Any Receipt Into Clean, Checked Expense Data

• ThinkCol built an AI expense tool combining OCR, line-item data extraction, smart categorisation, and behavioural anomaly detection.
• The tool reads receipts in nine or more languages, including handwritten Khmer, Bengali, and Sinhala, and reached 95%+ accuracy on vendor, date, total, and currency in testing.
• It runs entirely on-premises, so financial data never leaves the client's environment.

A global apparel manufacturer's roughly 7,000 staff submit expenses from across Asia, with receipts in nine or more languages, including handwritten Khmer, Bengali, and Sinhala. Finance entered each claim by hand — about five minutes per claim — and duplicates and policy breaches routinely slipped through.

ThinkCol built an AI expense tool combining OCR and data extraction (merchant, date, tax, and every line item), smart categorisation mapped to the client's chart of accounts that learns from finance's corrections, and behavioural anomaly detection that catches duplicate receipts across employees, per-diem breaches, and unusual spend. Staff photograph a receipt on iOS, Android, or web and verify it side by side; finance now triages by exception only.

Tested on real multilingual receipts, the tool reached 95%+ accuracy on vendor, date, total, and currency, and 90%+ on subtotals and tax, at around 10 seconds per standard receipt. It runs on-premises, so financial data never leaves the client's environment.

Why customise? Off-the-shelf expense apps are cloud-only, check fixed rules rather than behaviour, and aren't built around the client's finance stack.