Built around one stubborn problem
Finance teams spend an unreasonable amount of time sorting expenses into the right buckets. We set out in 2019 to change that with AI that does the categorization work for them — accurately, at volume, without constant supervision.
Where the work actually happens
Expense categorization sounds simple until you're dealing with 8,000 transactions a month across 14 cost centers. The manual process is slow, inconsistent, and produces data that nobody fully trusts. Accountants end up re-checking work instead of analyzing it.
Our system reads transaction descriptions, vendor names, and amounts — then assigns categories using models trained on real financial data. It handles edge cases that rules-based logic misses, and it improves as your transaction history grows. Typical setups reach over 91% first-pass accuracy within the first 6 weeks of use.
We work with finance teams across Singapore and the wider region — from mid-sized companies processing a few thousand transactions monthly to operations running well above 60,000. The product runs as an API so it slots into existing accounting workflows without replacing anything that already works.
-
Models trained on financial data, not generic text
Generic NLP models handle language well but struggle with vendor codes, abbreviations, and the shorthand that appears in real banking exports. Our models are trained specifically on financial transaction data.
-
Confidence scoring on every decision
Each categorized transaction comes with a confidence score. Low-confidence items get flagged for human review instead of silently passing through — so the data your team trusts is actually trustworthy.
-
Improving without manual retraining
Corrections made during review feed back into the model over time. After about 3 months of regular use, most clients see their manual review queue shrink by 40% or more compared to the initial baseline.
