Bookkeeping work cut by 62% with a model that learns each client's coding rules.

The story

Marcroft & Co looks after family and privately owned Waikato businesses, the kind of clients who value an accountant who actually knows their operation. Yet much of the team's week disappeared into repetitive bookkeeping — coding bank transactions to the right GL account, client by client, each with their own quirks about how a fuel stop, a tradie's hardware run or a mixed-use vehicle expense should be treated. The generic bank-feed rules in Xero only ever caught the obvious recurring lines, leaving staff to hand-code the long tail and re-answer the same GST-treatment questions month after month. It was steady, low-value work that quietly crowded out the advisory conversations clients actually pay for and the firm most wants to have. The coding also drifted: different staff sometimes treated the same kind of transaction differently, and onboarding a new bookkeeper meant teaching them each client's unwritten conventions one correction at a time. As the client base grew, the firm could see the maths — more ledgers meant proportionally more manual coding, more month-end pressure and less room to grow without simply adding headcount. They wanted to take the grind out of the work while keeping the accuracy and the partner-level judgement that their reputation rests on.

What we built

1. 01

   Built a model that learns each client's own coding history and chart of accounts, so it categorises the way that specific client actually books their world — not a generic, one-size-fits-all ruleset bolted on top of Xero

2. 02

   Auto-coded incoming bank transactions with the correct GL account and GST treatment, attaching a calibrated confidence score to every single line so nothing posts on a hunch

3. 03

   Routed only the low-confidence or genuinely unusual transactions to a person for review, while clean, high-confidence matches posted straight through without touching a keyboard

4. 04

   Built a feedback loop so every correction a bookkeeper makes sharpens that client's model, lifting accuracy steadily month on month rather than repeating the same mistakes

5. 05

   Kept the firm firmly in control with a clear audit trail behind each coding decision, so a reviewer can always see why a transaction was treated the way it was

“It actually learns how each client codes their books, GST quirks and all, so most transactions just fall into place with the right account and treatment. We're reviewing the handful of exceptions now instead of keying every line, and every correction we make makes the next month better. That time goes straight back into advising owners, which is the work we wanted to be doing all along.”

Results at a glance

• The bulk of routine transactions now code themselves, leaving staff to review exceptions rather than key every line by hand
• More consistent GST and GL treatment across each client's ledger, regardless of which staff member is on the file
• New bookkeepers get up to speed faster because the model already carries each client's conventions instead of living only in someone's head
• Freed-up hours redirected out of data entry and back into the advisory work that business owners genuinely value