Accountant Offering AI-Enabled Automation: What to Ask Before Hiring

Ask what is automated and what a person checks, which tools touch your data and where it goes, who signs off the numbers, what changes in the fee and the written scope, and how you can leave with your records. Get each answer in writing, and walk away from anyone who promises specific savings.

"AI-enabled automation" covers everything from a bank-feed rule that has existed for years to a language model drafting your management commentary. The label on its own tells you nothing: what matters is which tasks run without a person, which tools see your records and who signs off. If you are still choosing a firm in general, start with how to choose an accountant who uses AI well.

Follow me on Instagram@sagnikteaches

What the phrase can mean, and what to ask about each meaning

Ask the firm to say which of these it means, because the risk and the questions differ:

Connect on LinkedInSagnik Bhattacharya

If the answer is "all of the above", ask which one the proposal depends on most. Then make that the subject of the rest of the meeting.

Subscribe on YouTube@sagnikteaches

Questions about what is automated and what a person checks

This is the heart of the pitch, and it is where vague answers show up. Put each task to the firm and compare the answers with the pattern below.

Ask for a number, not a feeling: "Last month, how many transactions were matched automatically, how many went to review, and how many did your reviewer change?" A firm that tracks this can answer. One that cannot is not measuring its own error rate. If you want a primer on what automation can handle before you meet, can AI do my bookkeeping? sets out the split.

A sample exceptions report to ask for

Ask the firm to show you, or promise you, a monthly summary like this one (illustrative figures for a small optician's practice):

MONTH-END EXCEPTIONS REPORT (illustrative)
Transactions this month ............................ 612
Matched automatically (high confidence) ............ 541
Sent to reviewer ................................... 71
Reviewer changed the suggestion .................... 9
  - recoded: 6 (stock vs equipment: 2, repairs vs capital: 3,
              personal drawing: 1)
  - split across categories: 3
Unmatched at month end ............................. 4 (listed below)
Reviewed by: [name], signed off by: [name], on [date]

Nine changes out of 71 reviewed is about 12.7 percent of reviewed items and 1.5 percent of all transactions. The figure is not a pass mark. What matters is that it exists, that the changes are explained, and that somebody's name is on the sign-off.

Questions about the tools and where your data goes

Your books contain bank details, supplier terms, staff pay and sometimes customer data. Ask these seven questions and expect specific answers.

1. "Which tools touch my data? Please list every one." Include the accounting platform, the capture tool, any assistant used for drafting and any add-on the firm uses on its own account. A written tool register beats a verbal list.
2. "Are they business accounts?" Business plans of the major assistants (ChatGPT Business or Enterprise, Claude Team or Enterprise, Microsoft 365 Copilot, Gemini in Workspace) do not train on business content by default, while consumer plans let users switch training off in their settings. Your records should never sit in a free personal account.
3. "Does anyone at the firm use personal AI accounts for client work?" The firm should be able to say no, and say how it enforces that.
4. "Can you switch AI off for my file?" Some vendors do not allow it. Intuit says QuickBooks Online's AI features cannot currently be switched off individually, so a firm working in QuickBooks cannot promise a version of your file without them. The honest answer names what can and cannot be turned off.
5. "Where is my data stored and how long is it kept?" Ask for the location and the retention period, and whether files are deleted from the capture tool and any assistant when you leave. If you have customers in the EU, the GDPR turns this from a courtesy into a legal question.
6. "Who has access, and how is it protected?" Look for named staff, individual log-ins (not shared passwords) and multi-factor sign-in.
7. "What do you do if a tool has a breach or changes its terms?" A firm should have a plan, including telling you.

The tutorial on whether it is safe for an accountant to use ChatGPT with client data goes deeper on the data side, including what a firm should redact.

The tool register to ask for

Ask the firm to complete a simple table. This one is illustrative and filled in for a small bookkeeping engagement:

TOOL REGISTER (illustrative)
Tool               Purpose              Your data it sees        Training on content?   Who has access
Accounting suite   Ledger, bank feeds   Full ledger, bank feed   Check vendor terms     2 bookkeepers, 1 partner
Capture app        Read bills/receipts  Supplier invoices        Check vendor terms     2 bookkeepers
Business-plan AI   Draft commentary     Summary figures only     Off by default         Partner only
Email platform     Client messages      Correspondence           Check vendor terms     All staff

"Check vendor terms" is a legitimate answer only when followed by a link or a date. If the firm cannot say, ask them to find out before you sign.

Questions about review, sign-off and who is responsible

Automation moves the work, not the responsibility. Ask who is accountable, and test the answers with a realistic mistake.

Say the software codes a $4,000 equipment purchase as repairs. In a month-end review, a person reading the list of items over $1,000 should catch it. If nobody reads that list, the error reaches the year-end accounts and changes your tax figures. So ask, "What does your review look at, and what is the size threshold for items a person always checks?" A good answer sounds like: "Everything above $500 is read by a person, and any item the software was not confident about." A poor one sounds like: "Our AI is very accurate."

Then put these to the firm:

• Who is the named accountable person for my file, and who covers them?
• What happens when you find an error that automation caused? How and when will I be told?
• Does the engagement letter say who is responsible for AI-assisted work, and does it cap liability? (Ask a qualified adviser to read the limits.)
• Does your insurer know you use AI tools in client work?
• Will a person sign my filings, and can I see who?

The wider list of slips that damage client trust is in seven AI mistakes that put an accounting firm's client trust at risk; read it before the meeting so you recognise the pattern.

Fees: comparing two proposals line by line

Fees are where "AI-enabled" can hide extra charges or missing services. Take an illustrative veterinary practice that handles 220 supplier bills and 60 expense receipts a month, so 280 documents. It receives two proposals.

The sums: Proposal B with no queries is $480 plus $25.21 plus $90, which is $595.21. Add two hours at $85 and it becomes $765.21, or $9,182.52 over twelve months. Proposal A is $650 times twelve, or $7,800. B is cheaper only if the practice needs under about 39 minutes of queries a month, because $650 minus $595.21 is $54.79, and $54.79 divided by $85 is 0.645 hours.

Three further points hide in that table. First, Dext's Business plan covers 250 documents a month and the practice sends 280. Dext's page says documents over the limit still upload but are not read until the next billing date or an upgrade, so ask who pays for the larger plan. Second, ask whether the capture tool is needed at all: Xero says its built-in Smart Document Capture is included on all its plans, and if your books are in Xero it is fair to ask why a separate paid tool adds value. Third, ask what moves the fee: more documents, more bank accounts, a second entity or a payroll run.

Volume triggers need numbers too. Suppose Proposal A's small print adds $1.50 a document above 300 documents a month. If the practice grows to 560 documents, the extra 260 documents cost $390, and the "flat" fee becomes $1,040 a month. That is why the question "what moves the fee?" belongs in writing before you sign.

Neither proposal should promise a saving. If one says "our AI will cut your accounting costs by a third", ask for the working. Nobody can know that without seeing your records. For a method of scoring any AI quote, how to evaluate an AI implementation proposal or quote is a good companion.

The wording you want in the engagement letter

Verbal assurances disappear. Ask for these points to be written into the engagement letter or an attached schedule. AI engagement letters for accountants shows how firms draft scope clearly, which is helpful to know when you read theirs. An illustrative clause you can ask for:

Use of automated tools (illustrative wording)
1. We use the tools listed in Schedule 2 to capture documents,
   suggest transaction coding and draft reports.
2. A named member of our team reviews all items over [amount] and
   all items the software marks as low confidence. A partner
   reviews each month-end pack before it is released to you.
3. We do not enter your records into any tool not listed in
   Schedule 2, and we will tell you before adding or replacing one.
4. We do not use personal or consumer AI accounts for your records.
5. You may ask us to switch off optional AI features on your file;
   we will tell you what that changes in time and fees.
6. When you leave, we return your records in [formats] within
   [days] and delete copies held in our tools within [days].

Numbers in square brackets are yours to negotiate. The point is that each commitment can be checked.

How the answers differ across three kinds of business

The same questions get different priorities depending on what is in your books.

A care agency has payroll at the centre and staff data everywhere. The questions that matter most are which tools see pay and personal details, and whether anything with names and rates goes into a drafting assistant at all. A good firm keeps payroll steps with a named person and sends AI only summary figures.

A music school has hundreds of small payments, often by direct debit, with parents paying for several children. The risk is mismatched receipts: the software confidently applies a payment to the wrong family. Ask how matches are reviewed, and how the firm will show you unmatched or low-confidence items each month.

A veterinary practice has supplier invoices with many lines, stock purchases and equipment. Here the coding threshold matters, because a capital item miscoded as a repair changes the tax result. Ask which items a person always reads, and the size above which they do.

A pilot month that tests more than the demo

If the firm offers a trial month, use it to check the work, not just the dashboard. Give them one real month of bank feeds and bills. When the reconciliation comes back, pick fifteen transactions to inspect: the five largest, five chosen at random, and five that look unusual. Check each against the source document. If two of the fifteen are wrong, ask how they were caught, or why they were not, and what will change. One slip in a sample that size is not alarming. A pattern of misses in one category is. Ask also how long month-end took from the day you supplied everything to the day you received the pack, because speed is the benefit being sold.

Records and exit: leaving without losing your books

Ask before you sign, because it is awkward afterwards. Who owns the subscription to the accounting platform? The safest set-up has your business as the owner and the firm as an invited adviser, so you keep the file if you part ways. Ask what you can export, in which formats, and whether the document archive in the capture tool is included. Ask how long they keep copies after you leave. A good test: request read-only access during onboarding and confirm you can see the ledger and the archive yourself.

Claims that should end the conversation

• A promised percentage saving, or "100 percent accurate".
• A refusal to name the tools, or "it's proprietary".
• No named person who reviews the numbers.
• A request for your bank log-in details instead of a proper bank-feed connection.
• A tool subscription held in the firm's name that you cannot take with you.
• Pressure to sign a long term before a trial month.
• Unwillingness to put AI use into the engagement letter.

Not every item is equally fatal. A promised saving, a refusal to name tools and the absence of any named reviewer are reasons to stop. Pressure over a trial month and a tool held in the firm's name are worth negotiating before you decide, since a firm with good habits will usually fix them without fuss. If you are unsure which category a worry falls into, put it in your email and judge the answer.

A scoring sheet for two or three firms

This is my own rubric, not an industry standard. Score each item 0 (no answer or a warning sign), 1 (partial) or 2 (specific, written answer). Twelve items, so a top score is 24.

Firm 1's total is 2+2+1+2+2+1+2+2+2+2+2+1, which is 21. Firm 2's is 1+0+0+1+1+0+1+1+0+1+1+1, which is 8. Scores like these will not decide the choice for you, but they show where the gaps are and give you something specific to put back to the firm.

A short email you can send before the next meeting

Subject: Questions about the automation in your proposal

Hi [first name],

Thanks for the proposal. Before we decide, could you answer these
in writing?

1. Which tasks in my file are automated, and which are done or
   checked by a person? Who is the named reviewer for each?
2. Please list every tool that will touch my records, the plan type
   for each, and whether content is used for training.
3. Can optional AI features be switched off on my file, and what
   would that change?
4. Where is my data stored, how long is it kept, and how is it
   deleted when I leave?
5. What is the fee for my current volumes, and what would change
   it? Please itemise tools, reports and query time.
6. Will the engagement letter describe the tools and the review
   steps?
7. Could we start with a one-month trial?

Best wishes,
[your name]

The tone is deliberately plain. A firm that answers it clearly is telling you how it will answer you later; one that bristles has told you something else. If you would like a second pair of eyes on a proposal before you reply, that is the kind of job my AI implementation consultation covers.

Checking it is working in the first three months

After you sign, keep checking, because the first month is the best behaved. In month one, read the exceptions report and ask about every item you do not recognise. In month two, compare the override figures with month one: a falling number usually means the rules are improving, a rising one means the software or your records changed. In month three, ask for a short written review: which tasks stayed automated, which went back to a person and why, and what the total cost was against the proposal. If the firm cannot produce those three pieces, you have learned what you needed to know about how it manages its own automation.

Questions owners ask when an accountant pitches AI

Should I pay less because the accountant uses AI?

Not automatically. Compare the whole annual cost for the same written scope, including tools, reports and query time. Then ask what moves the fee if your volume changes. A lower price that quietly drops the review step is a worse deal than a higher price with a named person checking every month.

Can I ask my accountant not to use AI on my file?

Yes, and a good firm will say what that costs in time. Some platform features cannot be switched off individually, so ask what the software does with your data and which steps the firm can do by hand. Keep any AI drafting out of letters and commentary if you prefer.

Who is responsible if automation books something wrongly?

The firm is responsible for the work its engagement letter covers, and you remain responsible for the information you supply and the filings you approve. Ask for the wording on responsibility and limits in writing, ask whether the firm's insurer knows it uses AI, and have a qualified adviser read anything unclear.

What proof can I fairly ask for instead of savings claims?

Ask for process evidence: the list of automated tasks, a sample monthly exceptions report showing what the software handled and what a person changed, a trial month, and the turnaround for a month-end close. Treat any promised percentage saving as marketing, because it depends on your records and volumes.

How do I check a firm is using business-grade AI accounts?

Ask for the tool register in writing, with the plan type for each tool, whether content is used for training by default and who at the firm has access. Business plans of the major assistants do not train on business content by default. Consumer accounts should not hold your records.

Further reads

• How Much Does AI Cost a Small Accounting Firm? — What AI costs a firm, so you can judge its fees.
• How Accountants Use AI to Explain Tax to Clients in Plain English — How a careful firm explains the numbers to clients.
• Best AI Accounting Software for Small Businesses in 2026 — The software your accountant may be pitching, compared.
• AI Receipt Capture and Bank Categorisation: A Bookkeeper's Setup — How capture and categorisation work, step by step.
• AI Bookkeeping for Accounting Firms: Pricing It as a Service Line — How firms price AI bookkeeping, from the other side.
• How Accountants Use AI to Spot Errors in Client Books — How a careful firm uses AI to catch errors.
• Notion for Accountants: Firm Wiki, Client Trackers and Notion AI — A firm wiki, a client tracker database and a costed verdict on Notion AI, with the templates written out field by field.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Dext Business pricing page, Xero pricing plan comparison, Intuit and vendor privacy notes recorded in the pipeline's fact notes, read in October and November 2026. Fees, volumes and report figures in the examples are illustrative.