Catch-Up Bookkeeping With AI: Rescue Months of Untouched Books

A bookkeeper can use AI to clear untouched bank feeds and receipts by splitting the backlog into months, identifying duplicate or missing evidence, proposing categories for repeat transactions, and drafting a short client query list. Match existing bills before adding expenses, review every exception, and reconcile each account to its statement before moving to the next month.

The fastest-looking route, accepting hundreds of suggested categories in one pass, can create a slower clean-up later. Bank feeds may contain transfers, refunds and payments against bills already entered. AI can group and explain patterns; it cannot certify why a payment occurred or turn an unsupported document into valid evidence. Keep decisions traceable to the original line.

Follow me on Instagram@sagnikteaches

Scope the rescue before touching the ledger

First establish the last date on which each bank, card and clearing account reconciled correctly. Do not assume the previous year-end is clean just because a report exists. Inspect the latest completed reconciliation, bank statements, opening balances and any unexplained suspense account. Record the backlog period and account list in a one-page scope sheet. If a bank connection stopped downloading for several months, obtain missing statements or a complete export before you estimate how much work remains.

Connect on LinkedInSagnik Bhattacharya

Count transaction volume by month and source: bank lines, card lines, sales invoices, supplier bills, receipts and manual journals. Then sample at least one busy month and one quiet month. Note the percentage with attachments, the number of obvious transfers, the number of unreconciled deposits and whether the chart of accounts has been used consistently. These observations determine whether AI-assisted grouping will help or merely produce a large list of guesses. The broader choice of software is covered in selecting an AI bookkeeping platform for a practice.

Subscribe on YouTube@sagnikteaches

Define the engagement boundary in writing. Is the job to bring bank accounts current, reconstruct sales, investigate old balances, prepare records for a tax adviser, or all of these? State the number of accounts and months, the evidence the client must supply, how many query rounds are included, and what happens if the opening balance is unreliable. A catch-up quote that says only “six months of bookkeeping” hides the hardest variables. Keep the source files intact and make a backup or practice-approved copy before bulk edits.

For an illustrative kitchen fitter, the bank feed has been untouched for five months. It shows 780 bank and card lines, 160 photographed receipts, 45 open bills and two payment processor deposits each week. The first sample finds 11 bank lines that appear to pay bills already entered and six transfers between owned accounts. The bookkeeper therefore plans a matching pass before any bulk expense rules. The case numbers illustrate a scoping method, not a promised processing time.

Build a month-and-source control sheet

Make one control row per account per month. Include statement opening balance, statement closing balance, downloaded line count, ledger line count, receipt count, missing documents, unresolved client queries, reconciliation difference and reviewer initials. This sheet prevents a team from working months out of order and concealing a gap with a later adjustment. Assign an owner and a status such as waiting for statement, in categorisation, in reconciliation or ready for review.

Use the statement month as the work unit, even if the accounting period differs. Start with the earliest unclosed month because its closing balance becomes the next month's opening balance. Download bank and card statements directly from the financial institution or the client's secure files. Compare the statement's first and last transaction with the feed. A feed that begins on the third day is incomplete even if the software says “connected”. Import missing lines carefully; preserve transaction IDs or another duplicate check before adding a file.

An illustrative control entry reads: “Card, month 1: opening $4,200 owed; closing $5,135 owed; 92 statement lines, 90 feed lines, 71 receipts, two missing lines, eight client queries; reconciliation pending.” The two-line count difference is a stop signal. It could be a card refund and a fee absent from the feed. An AI summary that declares the month complete because most receipts have been attached is wrong. The reconciliation test, not the attachment percentage, closes the row.

A second example is a bank account with 160 feed rows but 159 statement rows. Sort by date, amount and reference and check for a duplicate import near the connection reset. Do not delete merely because two amounts match: recurring supplier charges often share a value. Compare source references and statement occurrence. If the duplicate is confirmed, document which software row was excluded and who approved it. This small control protects every later monthly balance.

Separate match, transfer and new expense

In an accounting system, matching a bank line to an existing bill or invoice links the payment to a record already in the books. Categorising it as a new expense creates a new record. Intuit's QuickBooks Online help article on matching (updated 14 August 2026) puts it plainly: match when you already entered a record for the transaction; categorise when there is no existing record, because that creates a brand-new one. The same article says that where the record has already been reconciled, you exclude the downloaded line so it does not create a duplicate, which is the situation a reconnected feed can produce when it re-downloads months the client part-processed. Match or categorise is the first question for every bank line, before choosing an expense account. A backlog can have many entered bills whose payments were never matched. If you add all those payments as new expenses, costs and liabilities become wrong at once.

Run the highest-confidence categories first. Identify transfers between the client's own accounts, invoice receipts, bill payments, payroll settlements and merchant processor payouts. Keep an exception field for ambiguous pairs. For example, an illustrative $3,400 payment to a supplier on 18 April may match bill 1048 for $3,400. If the bank description truncates the supplier name, compare date, amount, bill status and client evidence. Do not create a second materials expense just because the payee text looks unfamiliar.

A processor deposit needs a different reconstruction. Suppose the sales report shows $2,000 gross takings, $70 fees, and a $1,930 deposit. Posting only $1,930 as sales understates both revenue and fees. Use the processor statement to link the deposit to gross sales, fees, refunds and any reserve movement. AI can lay out the arithmetic and identify missing payout IDs, but the bookkeeper must check the settlement report. For a focused comparison, reconciling payouts and fees with AI goes deeper into this pattern.

Refunds and chargebacks are another trap. An illustrative $240 credit from a supplier should be linked to the original purchase or credited bill, not automatically placed in sales. A $240 customer refund has the opposite direction and may involve an invoice adjustment. Have the AI group candidate pairs by amount and date but include a “why this might not match” field. A good review list shows evidence, not a confident label.

Use rules for repetition, with a sample first

For repeat, low-risk transactions, draft a rule from consistent bank text, direction, amount pattern and account. Test it against a sample that includes both likely matches and near misses. QuickBooks documents bank rules for categorisation, but a rule is a standing instruction: a poor one can multiply an error across months. Keep match-first checks and require approval where the source evidence is weak. Record the rule owner, effective date and examples that justify it.

The mechanics matter for a catch-up job. Intuit's QuickBooks Online bank-rules article (updated 11 September 2026) says one rule can hold up to five conditions built on Description, Bank text or Amount; a Test Rule option checks the conditions against transactions already waiting for review; and only one rule applies per transaction, with the highest-priority rule applied first. It also says to use auto-post with caution, because matching transactions are then added to the books without your review. On a backlog, that suggests a routine: build each rule with auto-post off, test it against the month you are working on, read every hit, and only then let it suggest categories for later months. Where two rules could catch the same merchant, put the narrower one higher in the list, because the first rule that matches wins.

One illustrative rule covers a monthly software subscription: bank description contains a distinctive merchant string, amount is usually $49, money is leaving the operating account, and there is no open bill for that merchant. The bookkeeper inspects three months and the supplier invoices before applying it. A fourth month shows $98 because two seats were added. The rule can still suggest the category, but the amount change must be reviewed against the new invoice. Avoid a rigid amount rule that silently excludes the changed month.

Here is a prompt for a sanitised transaction sample. It requests candidate groups rather than posting instructions:

Group these anonymised bank lines by likely transaction type.
For each group give: row IDs, proposed treatment, evidence needed,
and a counterexample that would make the rule unsafe.
Prioritise matching an existing bill, invoice or transfer before
creating a new income or expense. Mark anything uncertain QUERY.
Do not infer a tax treatment and do not claim a receipt exists.

An illustrative output might propose: “Rows 14, 47, 81: recurring software merchant; likely software expense; verify invoice and billing change on row 81.” That is useful. If it says “rows 14, 47, 81 are fully deductible”, strike that claim. The transaction description alone cannot support it, and treatment can depend on facts outside the feed. An approved rule should code a bookkeeping account only after the bookkeeper checks the invoices and the client's chart.

Bulk reclassification also needs boundaries. Suppose 24 fuel transactions were coded to general travel, but the client operates both a delivery vehicle and a personal vehicle. A merchant-only rule would wrongly treat both alike. Split the candidate set using cardholder, vehicle record or client confirmation, and leave unknown rows in a query queue. This is where receipt capture and bank categorisation controls can help, but evidence still determines the final entry.

Turn uncertainty into one client query list

Clients answer specific questions faster than a vague request for “all missing receipts”. Create a query row with account, date, amount, bank description, suspected purpose, document needed and a reply deadline agreed for the engagement. Group related questions by month and theme. Do not send raw AI reasoning or confidential data through an unapproved channel. Use the firm's secure portal or established document exchange, and send a short notification separately if needed.

An illustrative query reads: “On 12 May, $1,180 left the operating account with reference ‘materials’. Was this payment for supplier bill 557, a new purchase, or a transfer? Please upload the invoice or confirm the payee and project.” That gives the client three plausible answers without suggesting that the bookkeeper has already decided. If a receipt is missing, record the response and its limitations. Do not manufacture a document or backfill an imagined description.

For five months of backlog, send one staged list rather than five separate emails in a day. Start with questions that block month 1's reconciliation or affect large balances. Then release the next month's list as earlier answers are applied. An illustrative first round has 18 questions; seven relate to transfers, five to unmatched bills, four to undocumented card spending and two to sales deposits. The bookkeeper can prioritise the two deposits because they affect revenue completeness, while the client can assign receipt collection to a staff member.

AI can draft a polite message from the approved query table. A sample opening is: “Hello [first name], we have grouped the remaining questions for the first month. Please reply in the portal beside each line and attach the relevant statement or receipt. Items 3 and 7 are holding up the bank tie-out, so please start there.” Review it for tone and accuracy. The technique is developed further in chasing missing client records before deadlines.

Reconcile month by month, then review the ledger

Once transactions are matched or coded, compare each account's ledger balance to the statement closing balance at the same date. Investigate outstanding payments, deposits in transit, duplicated lines, missing fees and date errors. A zero difference is necessary but not sufficient: two opposite errors can cancel. Review the list of uncleared items and the largest or oldest entries. Save the reconciliation report and the exception list with the month control sheet.

For the illustrative kitchen fitter, month 1's bank statement closes at $28,420, while the ledger initially shows $29,160: a $740 difference. The bookkeeper finds a $490 customer receipt imported twice and a $250 card settlement fee missing from the ledger. Removing the duplicate receipt takes the ledger to $28,670, and posting the fee takes it to $28,420. The next check is whether the remaining $490 receipt is applied to the right customer invoice and whether the fee agrees to the processor report. A numerical tie-out alone cannot prove the coding is right.

Review the profit and loss and balance sheet after the account ties. Look for negative liabilities, unexpected income spikes, old clearing balances, credit card payments recorded as expenses and unusual month-to-month swings. An illustrative $6,000 “sales” spike may actually be an owner transfer. If AI flags it, open the underlying bank lines and client explanation before changing the ledger. Compare customer and supplier ageing reports with the client where those subledgers are in scope. This catches errors that bank reconciliation alone misses.

If the previous closed month's opening balance is wrong, stop and escalate rather than forcing a journal into the current month. If a historic reconciliation has to be reopened, document the reason, the original and corrected balances, and who authorises it. A qualified accountant should advise on material prior-period changes and tax consequences. The bookkeeper's job is to preserve an audit trail and state clearly which months are confirmed versus provisional.

Price the clean-up as a defined job

Estimate from sampled effort, not only from the number of months. Separate one-off setup and opening-balance investigation from repeat transaction processing, client query rounds, reconciliation and manager review. An illustrative internal estimate might allow four hours to map accounts and import gaps, 12 hours for five months of routine transaction work, six hours for exceptions and queries, five hours for reconciliations, and three hours for final review: 30 hours in total. This is a planning example, not a market rate or a guarantee. Add a stated contingency for missing evidence and define how extra work is approved.

A fixed fee can work if the client supplies complete statements and the sampled exception rate is stable. If the first month reveals 80 undocumented payments rather than the sampled 10, pause, show the evidence and agree a revised scope before continuing. Alternatively, use a capped discovery stage followed by a separate completion quote. Explain what the client will receive: reconciled months, a query register, unresolved-items list and management hand-off. Do not sell “AI cleanup” as if the tool can certify the books.

After completion, give the client a simple prevention routine: upload receipts weekly, review feed suggestions before they accumulate, reconcile each account monthly and answer queries through one channel. Keep the categories and rules that survived review, with an owner for changes. If the monthly routine is still too heavy, examine rolling out AI in a bookkeeping practice as a controlled process improvement rather than repeating another rescue in six months.

Hand over a clean record and a clear exception list

The completion pack should make it possible for another bookkeeper to understand what was done without repeating the entire investigation. Include the period and accounts covered, statement copies or secure references, reconciliation reports, material journals, the final query register, rule changes and a short list of unresolved evidence. State the cut-off date for incoming transactions. A month closed on a Tuesday may have later bank entries that belong to the next period; the hand-over should say so plainly.

An illustrative exception summary might read: “Month 4, card account: $180 supplier charge has no invoice; client says it was a replacement fitting for project 42; coding remains provisional pending the document. Month 5, bank account: $620 deposit has no remittance reference; held in clearing until customer confirms.” This is stronger than a general note saying “some receipts missing”. The amount, month, source and next action are visible. The reviewer can decide whether the unresolved items are material to the engagement and whether an adviser should be involved.

Ask a second person to sample the final ledger where the risk warrants it. They can choose one large deposit, one transfer, one bill payment, one refund and one AI-suggested recurring category, then trace each from statement through ledger to supporting evidence. If even one sample fails, investigate the affected group before signing off all similar entries. Record which rules were tested and how many exceptions they produced. This quality check gives the firm evidence that the process worked, instead of relying on the AI tool's confidence language.

Questions before clearing an old ledger

Should I bulk-accept suggested categories?

Only after testing a narrow rule on a sample and checking the source documents. A similar bank description can cover different purposes, and a match to an existing bill is different from adding a new expense. Keep high-value, unusual, mixed-use and unclear transactions in a human review queue.

What if the client cannot find all receipts?

Log each missing item with date, amount, merchant and the question it blocks. Ask the client for the document or a clear explanation through the agreed secure channel. Do not let AI invent a receipt or assert a tax treatment. Your qualified accountant or adviser should decide how unresolved evidence affects the accounts.

When is a catch-up month complete?

A month is complete when its opening balance is trusted, all bank and card lines are matched or coded, exceptions are resolved or documented, and the closing balance ties to the statement. Review unusual balances and retain the reconciliation report before opening the next month.

Further reads

• AI Bookkeeping Software vs a Human Bookkeeper: Accuracy and Cost — Explain which judgements remain with a bookkeeper.
• Bookkeeping AI Reviews: How to Test a Tool on Your Own Client Files — Test a categorisation tool against familiar records.
• AI Bookkeeping for Accounting Firms: Pricing It as a Service Line — Compare a clean-up job with ongoing service pricing.
• AI Bookkeeping Agents: What They Do Alone and What You Must Review — Set safe approval boundaries for accounting agents.
• AI in Bookkeeping and Tax: The Year-End Handoff to Your Preparer — Prepare the eventual hand-off to a tax preparer.
• 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.
• Bank Statement PDF to Excel With AI: Check Every Line Before Import — A bookkeeper's routine for converting statement PDFs to Excel, then proving the result with balance, count and sign checks before anything is imported.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Intuit QuickBooks Online help articles on setting up bank rules (updated 11 September 2026) and on matching bank and credit card transactions (updated 14 August 2026), both read 4 November 2026.