Why the ledger keeps coming back to a person
Will AI replace bookkeepers? Not as whole jobs, but the task mix is moving quickly. Much of the day is structured and rule-bound: coding bank transactions, posting debits and credits, matching invoices to payments, and checking figures against receipts and statements. Accounting software has chipped away at that work for years, and better pattern matching has made the coding suggestions sharper.
What stays with people is the awkward middle. Reconciling an account that refuses to balance takes a search through dates, duplicates and timing differences. Classifying an unusual payment means knowing how the owner runs the business. Then there is the part no model signs for: answering an auditor’s query, explaining a variance to a manager, and standing behind the numbers when a bank or a tax authority asks.
Scale sets the stakes. The Bureau of Labor Statistics counted about 1,373,680 bookkeeping, accounting, and auditing clerks in the United States, with median pay of $50,670 a year, and projects employment falling 5.6% between 2025 and 2035 (BLS, 2025). That is erosion and fewer openings, not a job disappearing. Our coverage figure, which estimates the share of task time AI can handle today, comes out at 47 for this occupation; how coverage is measured explains what goes into it.
What software runs, what it drafts, and what people keep
The work AI can run with little help is the repeatable posting layer: pulling line items off scanned invoices and receipts, categorizing routine bank transactions, and computing and recording totals into the ledger. The share of task time in that group is 18%. These tasks have clean inputs, a right answer, and an audit trail, which is exactly where current tools are strongest.
The assisted group is larger than most headlines suggest. Bank and account reconciliation, and preparing draft financial statements and summary reports, both move faster with a model doing the first pass while a clerk checks the exceptions and signs off. That slice sits at 69% of task time. Speed goes up; the review step does not vanish.
Then there is the work still sitting with people, at 13% of task time. It includes handling questions from owners, vendors and auditors about discrepancies, and deciding how a messy or one-off item should be treated in the books. Compare this split with a neighboring role using the side-by-side job comparison.
What the evidence actually shows
No study has yet tested AI against working bookkeeping clerks on the real job. Our quality-parity grade for this occupation is D, and that grade means the comparison has not been measured, so we publish no parity number for it. Claims that software does 95% of bookkeeping perfectly are vendor marketing, not a tested result.
A fair test is easy to describe. Give an AI system and a qualified clerk the same real set of books for a month: bank feeds, card statements, supplier invoices, accruals and prepayments, plus a handful of planted errors and odd items. Score both on reconciliation accuracy, how many exceptions were caught rather than silently miscoded, and how well each explained the result to a reviewer. Until something like that is published and repeatable, the honest answer is that the posting layer is clearly automatable and the exception layer is untested. Our approach to grading is set out in the quality-parity method, and the full scoring rules sit on the methodology page.
When the picture could change
Most likely between 2035 and 2046 (8 in 10 of our scenarios). The replacement-year method explains how that window is built and what it does and does not claim.
Two things could pull the date forward. The first is tooling: bank feeds, e-invoicing and standardized data formats remove the messy inputs that currently force manual handling. The second is cost. The annual software cost range shown above runs far below the cost of the clerk hours it displaces, which is the kind of gap that pushes small firms to buy rather than hire.
Two things hold it back. Accountability is one. Someone has to approve the close, answer the auditor and carry the consequences of a misstatement, and that signature has stayed human. Data hygiene is the other. Plenty of small-business paperwork still arrives as paper, photos, or a shoebox of receipts, and this job keeps a small physical component covered only by fixed automation, as the robotics panel above shows. For a wider view of roles under similar pressure, see the jobs most at risk list.
How bookkeepers stay needed
Lean into the tasks the machine leaves behind. First, own exception handling: the entries that do not reconcile, the duplicate payments, the supplier credit nobody logged. Second, become the person who explains the numbers, to owners, lenders and auditors, in plain language. Third, take responsibility for controls and review: who approves what, how errors get caught, and whether the automated coding is actually right month to month.
Two skills carry the most weight. One is configuring and auditing the automation itself, including coding rules, bank feed mapping and spot checks on machine-coded batches. The other is advisory basics: cash flow, margins, payroll and tax timing, the questions a small business owner asks after the books are closed.
What to do: pick one client or entity and document every exception you resolved this quarter, because that log is the evidence of work software did not do.
If you want to move up or sideways, the closest paths are payroll and timekeeping clerks, billing and posting clerks, and accountants and auditors, which usually needs a degree and often a license. The financial clerks family page sets this job beside its neighbors, and the accounting firms sector page shows how the same work scores inside practices. To see where this role sits among every job we score, open the full rankings.