What Small Accounting Firms Can Safely Hand to AI, and What They Shouldn't

We build and run accounting and tax-filing software, including a system that turns bank statements and invoices into ledger entries with AI. The question we hear most from small firms is not "should we use AI?" but "which parts are safe?" This is our answer.

The evidence: the gains come from routine work

Researchers at Stanford and MIT surveyed 277 accountants and analysed hundreds of thousands of transactions at 79 small and midsize firms. Compared with non-users, accountants using AI:

• closed the month-end books 7.5 days sooner;
• recorded 55% more weekly client support;
• moved about 8.5% of their time (roughly 3.5 hours a week) away from routine data entry;
• kept more detailed ledgers: 12% more unique accounts used to categorise transactions.

And the demand is coming from clients too. Intuit's 2026 survey of small business owners found bookkeeping entered their top three uses of AI in January 2026. Its advice for accountants was one question: "Who checks the output?"

Green: hand these over now

• Chasing clients for missing documents. Polite, persistent reminders with a list of exactly what is missing.
• Drafting routine emails (engagement reminders, deadline notices) that a person sends.
• Meeting and call notes, written into the client file.
• Reading statements and receipts into draft entries, with the source line shown next to each one.
• Summarising a new rule or notice, with a link to the official text so you can check it.

Amber: AI suggests, a person approves

• Categorising transactions. AI is good at the 80% that repeat every month. The unusual 20% is where the errors hide.
• Matching for reconciliation. Let it propose matches; never let it force a balance.
• First-pass review of a return for missing items or numbers that changed a lot from last year.
• Answering client questions from your own approved notes, with a person reading before it sends.

Red: keep these human

• Submitting anything to a tax authority or a regulator.
• Tax positions and advice, where a confident wrong answer becomes your liability.
• Sending numbers to a client that nobody has checked.
• Pasting client data into free AI tools whose terms let them keep or train on it. Use a business plan with data protection, or don't use it for client data at all.

How to start this month

1. Pick one green task, usually document chasing, and run it for a month.
2. Measure hours saved and errors caught, not how impressive it looks.
3. Only then try one amber task, with a review step built in from day one.

Automating a messy process makes the mess faster, as we wrote in the automation trap. If you want help choosing the first task or building the review step, tell us about your firm.

Sources

• Journal of Accountancy, Calculating AI's impact on CPAs: New study quantifies time savings, 20 August 2025. Study by Jung Ho Choi (Stanford) and Chloe Xie (MIT): 277 accountants, 79 firms, close time, client support, time reallocated, ledger granularity.
• Intuit QuickBooks, 2026 AI Impact Report, May 2026. Survey of small and midsize business owners; bookkeeping entering the top three uses; "Who checks the output?"