AI in Bookkeeping and Tax: The Year-End Handoff to Your Preparer

AI can smooth a year-end handoff by drafting three documents from the ledger: a queries list of items the preparer will ask about, an adjustments log of journals and reclassifications with reasons, and a missing-document chaser for the client. The bookkeeper checks every line against evidence, and the preparer reviews the pack again before relying on it.

Year end tends to go wrong in the gap between two competent people. The bookkeeper knows why a payment was split; the preparer, three weeks later, sees a lump sum and writes an email. The client replies a week after that. A handoff pack with the reasons attached removes most of those loops, and AI makes writing it fast enough that it actually gets done. What it cannot do is decide whether a treatment is right. That stays with the humans on both sides.

Follow me on Instagram@sagnikteaches

What goes in the pack

Build one folder per client containing six things. The first three are the ones AI helps with most.

Connect on LinkedInSagnik Bhattacharya

1. Queries list: items you are unsure about or that look unusual, each with its evidence and your proposed answer.
2. Adjustments log: every year-end journal, reclassification and correction, with the reason and the source.
3. Missing-document chaser: the outstanding items, written for the client, with a deadline.
4. Standard reports: trial balance, transaction detail for the year, aged receivables and payables, bank reconciliations for each account at year end.
5. Supporting schedules: fixed asset additions and disposals, loan balances, stock count if relevant, payroll summaries and any contractor payment list.
6. A cover note: one page stating the period, the software, the cut-off date, who prepared what and anything the preparer must know.

The checklist of what must be ready before the pack is assembled belongs to the books themselves; the year-end accounts checklist goes through that. The pack starts where that checklist ends: the books are reconciled, and the question is how to hand them over cleanly.

Subscribe on YouTube@sagnikteaches

The queries list: from ledger scan to a clean table

Export the year's transactions to a spreadsheet and strip it down to date, payee, amount, account and memo. Remove anything that identifies individuals beyond what the preparer needs (employee names, for instance), and replace client or supplier names with codes if your policy requires it. Then ask the model to flag, never to fix:

Below is a year's transaction list from a small business. Flag lines a tax preparer would likely ask about. Do not change categories.
Look for: round-sum payments to individuals; payments to the owner or related parties; one-off large amounts; items in the wrong period (dates within 5 days either side of year end); possible duplicates (same payee and amount within 7 days); payments over [amount] with no memo; transfers between accounts that do not net to zero; negative expense lines.
Output a table: row reference, issue type, why flagged, suggested question for the client or bookkeeper.

Illustrative output rows: "Row 412, one-off large amount, $8,400 posted to Office Expenses on 14 March, question: is this equipment that should be capitalised, with an invoice?" and "Row 733, near year-end, $1,950 invoice dated 2 January posted in December, question: should this be in the next period?" The bookkeeper works down the list, adds the answer and the document name, and deletes false alarms. A list of 40 flags might reduce to 22 real queries. Models over-flag, which is better than under-flagging, but expect it.

What they also miss: anything that does not look odd in a spreadsheet. A transaction can be a perfect round of normal expenses and still be wrong because the client personally owns the vehicle. The queries list is a supplement to your own knowledge of the client, not a replacement.

A filled-in extract

Keep a "Status" column: open for preparer, waiting for client, resolved. It shows at a glance what each side owes.

The adjustments log

Every journal you post at year end gets one row: journal number, date, accounts debited and credited, amount, reason in one sentence, document reference. Bookkeepers often keep this in their heads or in a scrawled note; the preparer then reconstructs it from the ledger. AI is useful in two ways here.

Drafting reasons from journal narratives. Export the journals and ask for a plain-English reason for each, using only the narrative and accounts. "Accrued December electricity based on November's bill, estimated $310, supplier statement due in January." You then confirm or correct. The model writes tidy reasons; you supply the truth.

Checking the log against the ledger. Give it the journal export and the log and ask it to list journals missing from the log and log entries with no matching journal. Models do this comparison reasonably well on short lists; verify with a count of rows both sides so that nothing is dropped.

Compare list A (journals exported from the ledger) with list B (my adjustments log).
Output: (1) journals in A with no entry in B; (2) entries in B with no journal in A; (3) entries where the amount or date differs.
Do not guess matches. If unsure, put the row under (1) or (2) and say why.

The preparer will care most about three kinds of entries: estimates (accruals, prepayments, provisions), reclassifications between categories, and anything touching the owner's own transactions. Flag those three types in a column so they are quick to find.

The missing-document chaser

This is the safest first AI task in the whole handoff. You have a list: a receipt for a $2,300 purchase, a statement for a closed account, a signed loan agreement, a mileage log. The client has had polite requests all year and still hasn't answered. A good chaser is specific, short and dated.

Write a message to a small-business client asking for the documents below. Tone: friendly, direct, no jargon, under 150 words. Say the deadline is [date] because our preparer starts on [date]. List each item with the date and amount so they can find it. Explain in one sentence why we need it. Do not mention tax rules or advice.
Items:
1) Invoice for $2,300 paid 14 March to [supplier]
2) Final statement for the closed savings account
3) Signed loan agreement for the vehicle loan
[Sign-off: [your name]]

Illustrative output: "Hi [first name], we are pulling together your year-end pack and need three items by [date] so [preparer] can start on time: (1) the invoice for the $2,300 payment to [supplier] on 14 March, (2) the final statement for the closed savings account, (3) the signed vehicle loan agreement. We need the invoice to confirm what was bought, the statement to close the account correctly and the agreement to record the loan balance. If you cannot find any of them, tell us and we will work out what to do." Edit for your voice, add the real names in your own system, and send it yourself. Offer a one-click way to reply, such as a shared upload folder. Draft a second, firmer version for the one-week reminder, and a short third for the phone call. For general wording help, see ChatGPT prompts for bookkeepers.

Who checks what

The handoff has two sets of eyes, and each should know what the other has done.

Include the rule that AI output is never submitted anywhere: not to a tax authority, not to a lender, not to the client as a number. It is internal drafting; the figures in the pack come from the accounting software and are the bookkeeper's responsibility.

The cover note the preparer reads first

A preparer opens this page before anything else, so it decides where their time goes. A model can draft it from the queries list and log, but every line is the bookkeeper's statement. For the trades firm in the worked example below it might read (illustrative):

• Period and software: 12 months to 31 December; cloud accounting package; period locked on 24 January.
• Prepared by: bookkeeper, with the client's office manager supplying documents.
• Bank and card accounts: four, all reconciled to year-end statements; one savings account closed in August, final statement attached.
• Adjustments: nine journals in the log, three of them estimates (two accruals and a stock adjustment), marked in the type column.
• Open for the preparer: six queries, including the part-financed van and the owner transfers, each with a proposed answer.
• Waiting for the client: two documents, chased twice, due by the date in the chaser.
• Known weak spots: mileage records are estimates for the first quarter; the owner transfers have no loan agreement.

The last line matters most. A preparer who is told where the books are weak spends time there; one who is not finds out later, at a worse moment.

Handling client data safely

Ledgers hold client confidential data. Before using any tool, check three things: your engagement letter (does it allow third-party processing?), the vendor's terms (business plans such as ChatGPT Business and Claude Team say they do not train on your content by default, according to the vendors), and your own policy (what may be pasted). A practical compromise is to export only the columns the task needs, replace names with codes, and keep the key in your own system. Accounting software's built-in AI is a different case: it already reads the ledger, so before a client asks, check the vendor's current help pages for what each AI feature does with their data and which features can be switched off, because the options differ by feature and change often. Either way, tell clients in your engagement letter that AI-assisted tools may be used and that a person reviews all output. The longer discussion of rolling out controls across a practice is in rolling out AI in a bookkeeping practice without losing control.

Timeline: three weeks before the preparer starts

1. Three weeks before: reconcile all accounts to year end; run the flag prompt; send the first chaser with a deadline one week before the preparer's start.
2. Two weeks before: work through the queries list; post adjustments; update the log; send reminders for outstanding documents.
3. One week before: compare the log with the journal export; run the standard reports; freeze the period; chase the last items by phone.
4. Handoff day: send the pack with the cover note; agree who the single contact is on each side.
5. Two days after: a 15-minute call to walk through open queries, instead of an email chain.

With 30 clients, this staggers: nobody should be sending 30 chasers on one day. Schedule clients across weeks by complexity.

A worked example: a six-person trades firm

A bookkeeper looks after a six-person electrical contractor (illustrative figures). The year's ledger has 1,850 transaction lines. The flag prompt returns 46 flags; after review 24 are real queries. Eleven resolve from existing documents in the files. Seven need the client, so the chaser lists seven items with dates and amounts. Six are for the preparer: a van purchase part-financed, a $12,000 owner transfer, two accruals, a stock adjustment for cable drums, and a possible refund from a supplier credited to the wrong account. The adjustments log has nine journals, each with a sentence of reasoning drafted by AI and edited. Time: the bookkeeper estimates 4.5 hours for the pack compared with about 8 previously, mostly because writing reasons and chasers took minutes instead of an hour. The preparer's feedback is the real test: she reports that she had to email the client twice instead of her usual six times. That is one firm's illustrative picture, but it shows where the saving sits: in fewer rounds of questions, not in the ledger work.

A second example: inheriting messy books mid-year

Say a bookkeeper takes on a café owner in October, and the previous records are a spreadsheet plus a bank feed that was never reconciled. The handoff to the preparer in January is harder, because the bookkeeper cannot vouch for the first nine months. The honest pack says so. The cover note states: "Records from January to September were prepared by a previous provider and have been reconciled by us to bank statements only; supporting documents are available for about 60% of items." The queries list gets a new column, "confidence", with three levels: evidenced, bank-agreed only, estimated. AI helps by sorting the 900 older lines into those levels from memo fields and document-name matches in the upload folder, which the bookkeeper then samples. A preparer given this view can decide where to spend time, which is worth more than a pretend-clean ledger. The lesson is to let the pack describe the state of the books as it is.

A third example: the contractor payments list

Many preparers need a list of payments to self-employed contractors, with identifying details collected. From a year of supplier payments, ask the model to group payments by payee and flag payees that are likely individuals and not companies, with the total paid and number of payments. It then lists which payees have no tax identification or form on file, based on a second list you provide. Illustrative output: "Payee C-14, 6 payments, $7,820, no form on file." The bookkeeper verifies each against the bank and the supplier file, and the chaser requests the missing forms. Whether a payee must be reported, and the thresholds, depend on the rules the client falls under, so the preparer decides. See the year-end contractor forms checklist for the specifics if that applies to you.

The preparer's intake: what to ask the bookkeeper for

If you sit on the receiving side, you can ask for the pack in a form that suits you. Request the cover note first, then the open queries only, then the log, then the reports. Ask for the reconciliations with the bank statement balance and the ledger balance on the same page, and for the prior-year closing balances to be shown beside the opening balances. A preparer's own AI use is similar: ask it to read the pack and list the ten questions most likely to need the client, checked against the pack's own evidence. Never ask it to produce a tax computation or classification that you rely on without recalculating from source. For the busy-season practice side, see how small tax practices use AI through busy season.

When the bookkeeper and preparer disagree

Sometimes the bookkeeper categorised the van loan as an expense and the preparer wants a liability. The adjustments log is the way out: the preparer posts or requests a reclassifying journal, the bookkeeper records it with the reason, and the ledger changes once, visibly. Resist both sides keeping separate versions. If disagreements are frequent on the same item type (owner transactions, say), agree a short written convention and add it to the pack's cover note for that client. A model can draft the convention from the two parties' emails, but only the two humans can agree it.

Keeping it going all year

The best year-end handoff is one that is mostly done by October. Run the flag prompt quarterly on the latest quarter's transactions, clear the queries while the client still remembers, and chase documents monthly. At year end the pack is a roll-up, not a rescue. A small recurring calendar entry per client (one hour a quarter) is enough, and the log grows steadily instead of being reconstructed in January. Quarterly runs also catch categorisation drift early, such as a recurring supplier that started landing in the wrong expense account after a bank-feed rule changed.

Where it goes wrong

• Treating a flag as a finding. The model says "possible duplicate" and the bookkeeper deletes a payment. Always check against the bank statement first.
• Letting reasons drift. A model writes a confident reason ("accrual for unbilled services") that the bookkeeper has not verified. Reasons must come from evidence, so require a document reference on each row.
• A chaser with the wrong amounts. Copy errors in a long list. Check each line against the ledger before sending.
• Pack versions drifting. The bookkeeper posts a late journal after sending the pack. Freeze the period and re-send the log with a version date.
• Leaving the preparer out of the design. Ask your preparer what they would like to see first; some prefer a spreadsheet, others a PDF. Fit your pack to them.

How to tell it worked

Track three things across your first year-end with the pack: the number of back-and-forth emails per client, the days between handoff and first draft returned, and the number of adjustments the preparer made that were not in your log. If the third number is high, your queries list is missing something and you should look at why. Ask the preparer for a five-minute debrief. For the next stage, where the preparer explains the result to the client, see how accountants use AI to explain tax in plain English. If you would like help designing the pack around your own clients, that is what my AI implementation consultation covers.

Questions bookkeepers and preparers ask about the handoff

Which year-end task is safest to give to AI first?

The missing-document chaser. It turns a list of unmatched or undocumented items into polite, specific requests to the client. Nothing in the books changes, and a person reads every message before sending.

Can AI decide how a transaction should be categorised for tax?

It can suggest, but a qualified person decides. Treat AI category suggestions as queries to confirm with the client or preparer, never as final classifications, particularly for items with tax treatment such as assets, vehicles, meals or owner transactions.

Should the preparer see the AI's working?

Share the queries list and adjustments log, not the chat. The preparer needs the reasons and the evidence, so each line should say what was queried, the source document and the outcome. The chat history adds noise.

Can I upload client ledgers to ChatGPT or Claude?

Only under a plan and agreement your firm has approved, and only what is needed. Many bookkeepers export a trimmed list with names reduced to codes. Check your engagement letter and the vendor's terms, since client confidentiality duties still apply.

Further reads

• Payroll Year-End Checklist: Where AI Helps and Where It Misleads — The payroll part of year end, with AI's limits.
• Year-End Tax Planning Letters With AI: Personalise Without Errors — The preparer's side: planning letters to clients.
• Can AI Do My Bookkeeping? What Still Needs an Accountant — What still needs an accountant when AI does the bookkeeping.
• AI Bookkeeping Software vs a Human Bookkeeper: Accuracy and Cost — Accuracy and cost of AI bookkeeping against a person.
• Is AI Worth It for a One-Person Bookkeeping Business? — Whether AI pays off for a one-person practice.
• AI Receipt Capture and Bank Categorisation: A Bookkeeper's Setup — The categorisation work that sits behind a clean pack.
• Can AI Do My Business Taxes? What It Can Safely Prepare — How far AI can go on the tax side itself.
• AI Software for Tax Professionals: Run a Pilot and Measure Cost per Return — A four-week pilot plan for a small tax practice: one task, pass or fail rules, a consent and WISP check, and a cost-per-return formula.
• AI Bookkeeping in Canada: GST/HST Coding and Remittance Checks — A bookkeeper's review routine for GST/HST coding and payroll remittances: formulas for the maths, AI for the judgement calls, CRA pages for the dates.
• AI Tax Organiser: Build One That Chases Missing Documents — Turn a prior-year return into a client document list, track uploads row by row, and send reminders that name exactly what is still missing.
• Catch-Up Bookkeeping With AI: Rescue Months of Untouched Books — A bookkeeper's month-by-month recovery method, with exception lists, client queries, reconciliation checks and a scoped clean-up quote.
• AI Bookkeeping Software Challenges: Where It Goes Wrong and the Fix — A problem-by-problem repair routine for a small bookkeeping practice, with a verification check after every fix.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business and Claude pricing pages (training defaults), checked 4 November 2026.