01What "AI bookkeeping" actually means
The term covers a range of tools, from bank-feed rules with a little machine learning to systems that read documents, suggest categories, draft reconciliations and write variance notes. For an accounting firm, the useful question is not "does it use AI?" but "which steps of the close does it take off my team, and how do I check its work?"
02What AI does well
- Suggesting categories. Given your chart of accounts and past coding, AI can propose a category for each transaction, often with high accuracy for recurring vendors.
- Matching and reconciling. It can match bank lines to invoices, bills and payouts, and draft a reconciliation that highlights what does not match.
- Spotting anomalies. Duplicates, unusual amounts and transactions that break a pattern can be flagged for a person.
- Reading documents. Invoices, receipts and statements can be read and their figures extracted. See extracting data from W-2 and 1099 PDFs for the tax side of this.
- Drafting narrative. First drafts of variance notes and close memos, which a person then edits.
03What AI still can't do
- Understand intent. A $2,100 payment to a bank could be a loan repayment, an equipment purchase or an owner draw. Only the client knows.
- Make professional judgements. Accounting treatment, policy choices and materiality are decisions for qualified people.
- Know what changed in the business. A new location, a new line of business or a dispute with a supplier changes how transactions should be read.
- Take responsibility. The firm signs off on the books and the advice. Software cannot.
This is why the best AI bookkeeping tools are designed to draft and suggest, and to stop at the point where judgement begins.
04The design that keeps you in control
When you evaluate a tool, look for these features:
| Feature | Why it matters |
|---|---|
| Human approval before posting | Nothing changes the books until a person agrees |
| Confidence score on each suggestion | Reviewers can focus on uncertain items |
| The reason for each suggestion | You can see why, not just what |
| Differences raised as questions | Nothing is forced to balance |
| Source links | Every figure traces back to its document or bank line |
| Consistent rules | Your chart of accounts and rules apply the same way every month |
| An audit trail | You can see who approved what, and when |
We explain the approval pattern in how to automate transaction categorisation and still have a person approve it.
05How to evaluate AI bookkeeping software
- Test on closed months. Run the tool on a month you have already closed and compare its suggestions with your final books.
- Look at the uncertain items. How does it present low-confidence suggestions? Is the reason helpful?
- Check traceability. Pick ten figures at random and see whether you can trace each to its source in one click.
- Measure review time, not automation rate. A tool that automates 90% but makes the remaining 10% hard to review may cost more time than it saves.
- Check where it works. Does it work with your ledger, such as QuickBooks Online, Xero or Zoho Books, and your chart of accounts?
- Ask about data. Where is client data processed and stored, and what do the terms of any AI provider say about using your data?
06How AI changes your team's work
AI does not remove the need for bookkeepers; it changes what they do. Instead of typing and coding, they review suggestions, resolve exceptions and talk to clients about the items only the client can explain. That is higher-value work, and it scales better: one reviewer can approve far more transactions than they could code by hand.
For firms short of staff, the combination of AI drafting and a managed preparation team can add capacity without hiring. See how to outsource bookkeeping without losing quality.
07Questions to ask vendors about AI and data
- Which AI providers do you use, and do their terms allow them to train on our data?
- Where is client data processed and stored?
- Can we turn AI features off for specific clients?
- How are suggestions logged, and can we export the audit trail?
- What happens to our data if we stop using the product?
08A worked example: one month with AI-drafted books
Take an illustrative dental practice with around 400 bank and card transactions a month.
- Import. Bank and card feeds bring in October's transactions.
- Draft. The software suggests categories for every line. Recurring vendors, such as the payroll provider, the dental supply company and the landlord, are suggested with high confidence and a clear reason.
- Review the uncertain few. About a dozen items fall below the confidence threshold: a new vendor, a refund, and a $2,100 payment to a bank that could be a loan repayment or an equipment purchase.
- Ask the client. The bank payment becomes a one-tap question in the client portal. The client answers "loan payment" the same day.
- Approve. The reviewer approves the high-confidence items in bulk after scanning them, then resolves the uncertain items individually.
- Reconcile. The software drafts reconciliations for each account; the reviewer confirms they tie to the statements.
- Close. Draft statements and variance notes are produced, edited and approved by the firm.
The reviewer's time goes into the dozen items that needed thought, not the hundreds that did not.
09How to roll out AI bookkeeping in your firm
- Start with a few clean clients. Clients with consistent vendors and connected feeds show results fastest.
- Agree thresholds. Decide what confidence level allows bulk approval, and keep it conservative at first.
- Train reviewers on the queue. Reviewing suggestions is a different skill from coding transactions. Show reviewers how to read the reason and the alternatives.
- Track corrections. Every correction is a signal. Recurring corrections become rules.
- Review after three months. Compare review time and error rates with the months before.
10Questions clients may ask
Clients sometimes ask whether "a robot" is doing their books. A clear answer helps:
- Software drafts the routine work so your team can spend more time on review and advice.
- A qualified person reviews and approves everything before it reaches the books.
- Your firm remains responsible for the work and the advice.
- Their data is protected under your firm's security and confidentiality obligations.
11Signs a tool is not right for your firm
- It posts to the books without approval.
- It cannot explain its suggestions.
- It forces differences into suspense or plug accounts.
- It does not work in your ledger or with your chart of accounts.
- The vendor cannot say where data is processed or how AI providers use it.
12How Marksford Books works
Marksford Books categorises transactions and drafts reconciliations for human approval. Each suggestion shows its confidence and the reason. Differences become questions, which can go to the client through ClientBase. Books supports the close with draft statements, variance notes and a close memo for partner review. Books is priced per client; see pricing.
FAQFrequently asked questions
Can AI replace a bookkeeper?
No. AI can draft categories and reconciliations, but people are still needed to review suggestions, resolve exceptions, understand the client's intent and make professional judgements.
Is AI bookkeeping accurate?
Accuracy varies by tool, client and data quality. Recurring vendors are usually categorised well; unusual transactions need a person. Judge tools by how clearly they show uncertainty and how easy they make review.
Is client data safe in AI bookkeeping tools?
It depends on the vendor. Ask where data is processed and stored, which AI providers are used and whether they may train on your data, and how access is controlled and logged.
What should an accounting firm look for in AI bookkeeping software?
Human approval before posting, confidence and reasons for suggestions, differences raised as questions, source links, consistent rules and a full audit trail.
Does AI bookkeeping software work with QuickBooks Online and Xero?
Many tools do. Check that the tool works in your ledger with your chart of accounts rather than requiring a separate system.
→Tools for this topic
→Related guides
This guide is general information for accounting professionals. It is not legal, tax or accounting advice. Marksford is not a CPA firm.