Accounting Firms Using AI Close the Books 7.5 Days Faster

Most accounting firms using AI are still at the ChatGPT-prompt stage — drafting emails, summarising documents, maybe writing a client letter. Useful, but not where the real numbers are. The real numbers come when AI moves from a tool you open to an agent that learns how your firm works.

Last month, Dext — the document processing platform used by thousands of Australian accounting and bookkeeping firms — launched AI Assist. It’s an AI agent that watches how your team categorises transactions, applies tax treatments, and structures data, then starts automating those same decisions across future workflows. Every suggestion is surfaced for human review before it’s applied. But the shift is significant: the AI isn’t waiting to be asked. It’s learning your firm’s patterns and proposing the next step.

To understand why this matters, consider the volume. In January 2026 alone, Dext processed 31.4 million receipts and invoices globally. At three to four minutes per document manually, that’s more than two million hours of admin work. With the platform, users spent roughly 206,000 hours — a 90% reduction in processing time, according to Dext.

But document capture is just step one. After receipts are processed, someone still needs to categorise them, assign tax codes, check for anomalies, and reconcile. That’s where most of the human hours actually go. And that’s exactly where AI agents like Dext AI Assist are now stepping in.

A 2025 study by Stanford’s Jung Ho Choi and MIT’s Chloe Xie analysed hundreds of thousands of transactions across 79 small and midsize accounting firms and surveyed 277 accountants. The findings, published on SSRN, are concrete.

Accountants using AI finalise monthly statements 7.5 days faster. They support 55% more clients per week. Reporting granularity — the level of detail in financial records — improves by 12%. And 8.5% of total work time shifts from routine processing to higher-value advisory tasks.

Senior accountants in the study treated AI as a collaborator: stepping in when the system’s confidence dropped and applying judgment where it mattered. The AI handled pattern recognition. The human handled exceptions. That’s the model that works.

For an Australian accounting firm, a 7.5-day improvement in close time isn’t abstract. If your firm runs monthly bookkeeping for 50 clients, finishing the month-end cycle a week earlier frees capacity for advisory work, tax planning, or taking on new clients without hiring.

Run the numbers on the time shift. A 10-person firm with a fully loaded cost of $75 per hour per staff member, where AI moves 8.5% of each person’s time from routine admin to billable advisory work — that’s roughly 3.4 hours per person per week. Across the team: 34 hours a week, or $132,600 a year in recovered capacity. The admin leverage implications are hard to ignore.

The firms already using AI aren’t just working faster. According to the Stanford and MIT research, they’re serving more clients with the same headcount and producing more detailed records while doing it. We’ve written before about how WIP write-offs drain accounting firm profitability — faster close times and better record-keeping attack that problem at the source.

Start with what you already pay for. If your firm uses Dext, Xero, MYOB, or QuickBooks, check what AI features have been added in the last six months. Many are switched on by default and sitting unused. Dext AI Assist is available as a free trial until 23 April before moving to a paid add-on.

Then pick one workflow. Month-end close is the obvious candidate — it’s repetitive, high-volume, and the Stanford data shows it responds well to AI. Have one person trial the AI-assisted workflow for a single client cohort and measure the time difference. The firms pulling ahead aren’t using some secret platform. They’re using the same software you are — they’ve just switched the AI features on.

Key takeaways

Want to know where your firm’s admin hours are going?

A 30-minute call can map the workflows where AI would make the biggest difference — and whether the tools you already pay for can do it.