How to Choose an Outsourced Bookkeeping Service That Uses AI

Choose a service that keeps your books in an accounting file you own, such as your own Xero or QuickBooks subscription, names the AI tools that categorise and reconcile your transactions, has a named person reviewing every month-end, and commits to a close date and error-fix process in writing. Then test it on a month of real transactions.

Outsourced bookkeeping services now use AI heavily for categorising, matching and reading receipts, and that's much of what makes low monthly prices possible. The risk is volume with too little review. Businesses with payment-platform payouts, refunds and prepaid packages are exactly where categorisation software slips, because the same bank line can mean several different things. You're looking for a service whose people understand your money well enough to catch those slips.

Follow me on Instagram@sagnikteaches

The example throughout is a yoga studio with around 600 members, selling monthly memberships by direct debit, class packs through its booking platform, some retail (mats and water bottles) and occasional workshops. It processes about 1,400 transactions a month and pays 14 freelance teachers. The owner had been doing the books on Sunday evenings and wanted to stop.

Connect on LinkedInSagnik Bhattacharya

Your file, their people: the rule that protects you

The most important question has nothing to do with AI: where do your books live? Some services keep your records inside their own platform. Others work inside an accounting file, such as a Xero or QuickBooks subscription, that is in your name and paid for by you, with the bookkeeper invited as a user.

Subscribe on YouTube@codingliquids

The second arrangement is the one to insist on, and there's a recent reason why. Bench, an online bookkeeping service, shut down on 27 December 2024. Its clients' books lived in Bench's own software rather than in ledgers the clients owned, and that software became inaccessible the day it closed, leaving clients to recover their records and find new help at year-end. Any provider can close, get acquired or change its prices; if your ledger is yours, you change bookkeeper and keep every transaction, rule and attachment. The wider set of checks for any supplier that might disappear is in what happens if your AI vendor shuts down.

So the first filter is simple. If a service says your books will live in its proprietary system and you'll "get reports", ask whether it will work in a file you own instead. If it won't, move on, however good the price.

Where AI works in a bookkeeping service, and where people check

Most of the AI in a modern bookkeeping service is inside the ledger software itself, plus a few specialist tools. Knowing which is which helps you ask sensible questions. For the broader question of what software can now do on its own, see can AI do my bookkeeping.

Two product facts are worth having in mind. Intuit says QuickBooks Online's AI features can't currently be switched off individually, so a service working in QuickBooks needs a review routine rather than an off switch. And if a service promises that bills will be fetched automatically from your suppliers' online portals, ask exactly how: Hubdoc, a popular capture tool, stopped fetching bills from supplier portals back in 2022, and not every provider has updated its sales material.

Twelve questions to send before any sales call

Outsourced bookkeeping services vary from a single freelance bookkeeper to large online firms with teams in several time zones. The same questions work for all of them, and written answers are easier to compare than a sales call. The yoga studio's email, filled in:

Subject: Bookkeeping for [studio name] - questions before we talk

Hi [first name],

We're a yoga studio: ~1,400 transactions a month, memberships by
direct debit, class packs via our booking platform (paid out net of
fees), some retail, 14 freelance teachers paid monthly. We use Xero.

1.  Will you work in our own Xero file, with us as the subscriber?
2.  Which AI or automation tools will touch our data, beyond Xero?
3.  Are those tools on business plans that don't train on our data?
4.  Who reviews the AI's categorisations and matches, and how often?
5.  Will we have a named bookkeeper? Who covers their holidays?
6.  By what date each month will the previous month be closed?
7.  How do you handle payouts that arrive net of platform fees?
8.  How do you record class packs paid for in advance?
9.  If you make an error, how do we report it and how fast is it fixed?
10. What access do you need to our bank? (We won't share logins.)
11. Is any of our work subcontracted to another firm?
12. If we leave, what do we get, and what notice do you need?

Thanks,
[your name]

Questions 7 and 8 are specific to the studio's money, and they're the most revealing. A service that answers them precisely has handled businesses like yours. For question 8, the treatment of prepaid packages is ultimately a decision for your accountant, so a good bookkeeper's answer will mention agreeing the approach with them.

How to read the replies: firm, specific answers to 1, 4, 6 and 9 are essential. "Our AI is 99% accurate" in answer to question 4 is not an answer; it's a claim about software, when you asked about people. Vague answers to 11 deserve a follow-up, because subcontracting isn't wrong, but you should know who can see your data. And a refusal on question 1 ends the conversation, for the reasons above.

Comparing quotes by transactions, deadlines and exclusions

Bookkeeping services price in several ways: a fixed monthly fee set by transaction volume, a per-transaction rate, hourly billing, or a package with extras. To compare them, put every quote into one grid. The studio's three quotes, with illustrative figures that describe these particular providers rather than the market:

Service A is cheapest on paper at $350 a month including teacher payments, but it fails the first filter: the books would live on its platform. Service C costs about $380 a month if the hour estimate holds and could drift higher in busy months, and it has a single point of failure. Service B costs more, closes the month ten days earlier and puts everything in writing. The studio shortlisted B and C for a trial month.

If you're still deciding between outsourcing and simply using better software yourself, AI bookkeeping software vs a human bookkeeper compares accuracy and cost, and what AI bookkeeping software costs per month prices the software on its own.

The trial month: testing a service on your own transactions

Any service can describe its process. A trial on one month of your real transactions shows whether it works for your money. Ask each shortlisted service to process the same recent month, paid at its normal rate, and check the result against this list:

• The bank balance reconciles to the cent on the last day of the month, in every account.
• A random sample of 30 transactions is in the categories you'd choose. Two debatable ones is normal; five wrong ones is a pattern.
• Every payout is split into gross sales, platform fees and refunds, so income isn't understated and fees don't vanish.
• Questions arrived about genuinely unclear items, rather than guesses. A good trial produces five to ten sensible questions from a studio like this.
• The close date was met, and the month-end summary matches the ledger.

Record the results in the same grid for each service. It takes an hour of your time and turns the choice from a matter of impressions into a comparison of evidence.

Worked example: class passes, payouts and refunds

The trial exposed exactly the problems the studio's money invites. Service C's first pass, done largely by bank rules the software suggested, had four errors in the 30-transaction sample:

1. Payouts recorded net. The booking platform paid out $18,400 for the month after deducting $540 of fees and $310 of refunds. The payout was recorded as $18,400 of sales. Sales were understated by $850, and neither the fees nor the refunds appeared anywhere.
2. A refund treated as an expense. A $120 class-pack refund paid by bank transfer was categorised as "general expenses", because the rule matched on the word "refund" in the reference.
3. A teacher paid twice in the ledger. One teacher's invoice and the bank payment were both recorded as expenses, doubling $640 of cost, because the software matched the payment to nothing and created a new expense.
4. Retail stock as sales. A $900 order of yoga mats for resale was categorised as "equipment", which is closer but still wrong for stock the studio sells on.

None of these is exotic. Each is what categorisation software does when a bank line looks like something it has seen before. What mattered was the response: Service C fixed all four within two days, explained each, and changed the rules behind them, including a monthly routine to split every payout from the platform's own report. Service B's trial had two debatable categorisations and no errors, and it had asked eight questions, including one about whether workshop deposits should be held separately until the workshop ran.

The studio chose Service B. The deciding factors weren't the error count alone, but the earlier close date, the written exit terms and holiday cover, which mattered for a business whose teachers need paying on time. It kept Service C's details as a fallback, which is worth doing with any supplier.

What a good month-end note looks like

Each month the service should send a short note when it closes your books. It's where AI drafting helps the bookkeeper and where you can see whether a person actually looked. This illustrative note from Service B's second month shows the standard to expect:

"September is closed. All three bank accounts reconcile to the statement balances. Booking-platform payouts are split into $19,120 of gross sales, $560 of fees and $280 of refunds. Two items need you: a $95 card payment to an unfamiliar supplier on the 14th (receipt, please), and whether the $1,200 workshop deposit should stay as income received in advance until the workshop in November; your accountant may want a view. Teacher payments match the rota; one teacher invoiced a class that was cancelled, which we've queried with them."

Every figure can be checked against the ledger, the open questions are specific, and the note shows judgement about what matters. A note that only says "all reconciled, see attached reports" tells you nothing about whether anyone reviewed the software's work.

Your side of the arrangement

An outsourced service can only be as accurate as the information it receives, and AI categorisation gets noticeably better when the inputs are tidy. The studio owner agreed to a short monthly routine:

• Photograph or forward receipts within a week, using the ledger's capture tool, not a shared photo album.
• Keep business spending on the business card, so personal items don't need untangling.
• Download the booking platform's monthly payout report and send it on the first working day, so payouts can be split properly.
• Answer the bookkeeper's questions within three working days.
• Tell the bookkeeper before anything new starts: a new income stream, a new supplier on a contract, a teacher changing how they invoice.

Twenty minutes a month on this list does more for the accuracy of your books than any feature in the software.

Security and access: what to grant and what to refuse

• Never share online banking logins. Bank feeds connect the ledger to your bank through the software's own authorised connection. A bookkeeper who asks for your banking password to "download statements" is asking for the wrong thing; the tutorial on connecting AI tools to a business bank account safely explains the safer routes.
• Use named user accounts with sensible roles. Invite the bookkeeper as a user of your ledger with the access they need, and require multifactor sign-in.
• Ask where documents are stored after capture, and who at the service can see them.
• Check the AI tools' terms. Anything outside the ledger that processes your data should be on a business plan that doesn't train on customer content by default.
• Get confidentiality in writing. Your bookkeeper sees payroll, personal spending on business cards and your margins; the engagement terms should cover confidentiality and data handling. Ask your adviser if anything looks thin.

Leaving cleanly: the exit plan to agree now

Agree the exit before you start, while everyone is keen. The studio's terms with Service B include:

• 30 days' notice from either side.
• The ledger remains the studio's subscription throughout; on exit, the bookkeeper's user access is removed, nothing else changes.
• The final month is reconciled and closed before access ends.
• A written list of bank rules and recurring adjustments (such as the monthly payout split) is handed over, so the next bookkeeper doesn't rediscover them by making the same mistakes.
• Source documents stay attached to transactions in the ledger (in Xero, the area formerly called Files was renamed Documents in September 2026), not in the service's own storage.

That list is short because the most important decision, keeping the books in your own file, was made on day one. If your accountant is separate from your bookkeeper, share the list with them too; the companion tutorial on choosing an accountant who uses AI well covers that side of the relationship, and Xero vs QuickBooks AI helps if the choice of ledger is still open.

Further reads

• AI Receipt Capture and Bank Categorisation: A Bookkeeper's Setup — How receipt capture and bank rules should be set up.
• Rolling Out AI in a Bookkeeping Practice Without Losing Control — What a well-run AI bookkeeping practice looks like from inside.
• How Much Does Automated Invoice Processing Cost per Invoice? — What automated bill processing costs per invoice.
• AI Vendor Lock-In: How to Keep Your Data and Prompts Portable — Keep your data portable, whoever does the work.
• Best AI Accounting Software for Small Businesses in 2026 — The ledgers a good service should be working in.
• Can AI Do My Business Taxes? What It Can Safely Prepare — What AI can safely prepare for your business tax return, where its tax answers go wrong, and a year-end pack you can build before the accountant sees it.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the series fact sheet for the Bench shutdown date, Hubdoc's 2022 change, Xero plans and features (JAX reconciliation, Smart Document Capture, Files renamed Documents) and QuickBooks Online plans and AI settings.