Staffing services · Treasury
76.2% of transactions classify themselves. With 97.5% accuracy, measured against 1,043 real decisions.
The accountant no longer looks up the category for every charge: they confirm or correct what the system proposes.
The business before
A legacy desktop program and hand-built spreadsheets. An accountant reviewed every debit and every credit to assign its category. When an audit came around, explaining why something had been classified that way depended on the memory of whoever did it.
What we did
A treasury platform that works on top of the administrative system the company already used. There was no migration and no double entry.
Run by the system
It imports the bank statement, applies the rules, proposes the accounting category, reconciles against the ledger and projects cash flow against budget.
Decided by a person
They confirm or correct each proposal.
How it is maintained
The rules are edited on screen. If an accounting policy changes, nobody has to call a developer. A rule can be tested before it goes live without touching the real ledger, and every classification writes down its reason.
The number
| What | Before | Today | Type |
|---|---|---|---|
| Transactions that classify themselves | manual classification; no quantitative baseline | 76.2% | measured |
| Accuracy when the system proposes | — | 97.5% | measured |
| Classifying one transaction | 3.5 min | 20 sec | estimated |
| Weekly cash-flow projection | 4 h | 1 h | estimated |
| Time given back to treasury | — | 71 h/month | estimated |
Source. The 76.2% and 97.5% figures come from running the system against 1,043 real transactions an accountant had already classified by hand, comparing them one by one. Scope as of 20 August 2026. The times are estimates against the equivalent manual process and mid-range operating volumes.