Is Bookkeeping a Dying Field? Depends Which Half You Sell

You went in for coffee and career advice. You came out with a sentence you can’t stop replaying: “Honestly? I wouldn’t get into bookkeeping now. It’s finished. QuickBooks does it, the bank feed does it, and in three years AI does the rest.” He said it kindly, which somehow made it worse.

So: is bookkeeping a dying field? Half of it, yes. Data entry bookkeeping is genuinely dying, and it has been dying since bank feeds went mainstream around 2010, not since ChatGPT. Bookkeeping as a service has not been dying, because what clients actually buy has shifted from recording transactions to owning the accuracy of the numbers and being the person who explains them. Those are two different products that share a job title, and a CPA who conflates them will give you advice that is half right and completely useless.

What follows is the arithmetic, the parts automation has demonstrably failed to take, the business-model change that decides what you earn, and a blunt section at the end about who should actually leave.

The dying field claim, against the actual BLS numbers

BLS tracks this occupation as “bookkeeping, accounting, and auditing clerks,” SOC code 43-3031. As of the 2025 data:

Employment: 1,532,400 jobs
Projected change 2025 to 2035: -6%, a loss of 85,600 positions, taking the occupation to 1,446,800
Median pay: $50,670 a year, or $24.36 an hour
Lowest 10%: under $36,000. Highest 10%: over $74,550
Projected average annual openings: 144,100 a year

Read the last two lines together. The occupation is shrinking by roughly 8,560 jobs a year on average and simultaneously expects about 144,100 openings a year. Over the decade that is around 1.44 million openings against a net loss of 85,600. Seventeen openings for every position the projection deletes.

BLS is explicit about where those come from: all of them “result from the need to replace workers who transfer to other occupations or exit the labor force, such as to retire.” That is the whole mechanism. A declining occupation with a large, ageing workforce still has to refill nearly a tenth of its seats every year. A shrinking occupation and a hiring market are not contradictory; they are the normal condition of any large, mature, older-skewing trade.

Now the comparison that should stop your CPA mid-sentence. Accountants and auditors, the job he has, sits at 1,595,200 positions with a projected +5% and about 115,300 annual openings: fewer than the “dying” occupation, from a larger and growing base. His field is healthier per worker. It is not busier at the door.

Bookkeeping ranks fifth on the BLS list of largest projected job declines, behind cashiers (-200,600), general office clerks (-156,200), customer service representatives (-141,800) and secretaries and administrative assistants (-114,100). Look at that list honestly. Every one is a job defined by keying, filing or reading from a script. That tells you which part of bookkeeping the projection is killing.

Your CPA is right about half of it

Refusing to concede this point is how bookkeepers end up in trouble. The following work is gone or going, and no amount of positioning brings it back:

Manual transaction entry. Bank and card feeds pull the line items in automatically. That single change, which predates the current AI cycle by well over a decade, removed the bulk of what an entry-level bookkeeper used to do all day.
Receipt and bill keying. OCR through Dext, Hubdoc or the capture built into Bill.com and Ramp reads vendor, date, total and tax off a photograph well enough that retyping it wastes your client’s money.
Rules-based categorisation. Every recurring, identically-named vendor payment gets a bank rule once and codes itself forever. Machine suggestion has pushed this into merely-similar transactions too.
Reconciliation matching. Matching a cleared bank line to an existing entry is pattern work. Software is better at it than you, is not bored, and does it in a second.
Basic report production. Nobody pays for a printed P&L. It is a button.

If your working week is mostly those five things, your CPA is describing your job accurately and you should read this as a warning, not reassurance. The rate for that work has fallen for fifteen years and will keep falling, because it is competing against software, offshore providers and the client’s own bank feed at once.

The half automation never touched

Automation processes clean, structured, repetitive input brilliantly. What it has not solved, and is not close to solving in any way you can hand to an owner unsupervised, is the messy part. In real small business books, the messy part is most of it. A short list of things I have never seen software resolve correctly on its own:

The owner who pays personal expenses from the business account. Software codes the Home Depot charge to Repairs and Maintenance. It was a new deck at his house. No feed, rule or model knows that. You know because you asked, and because the same card was at a garden centre that afternoon.
Loan payments. A single $2,183 debit is part principal, part interest. Automation books the whole thing somewhere. Getting it wrong distorts profit every month and quietly corrupts the balance sheet until someone applies for financing.
Merchant and marketplace payouts. The $8,412 that landed from Shopify is not revenue. It is gross sales minus refunds, processing fees and chargebacks, plus a payout timing difference straddling month end. Amazon settlements are worse. Getting revenue right here is judgment plus a reconciliation nobody wrote a rule for.
Owner compensation. S-corp shareholder wages versus distributions, guaranteed payments in a partnership, an owner draw miscoded as payroll expense. Classification decisions with tax consequences, which the client will never raise because the client does not know they exist.
Payroll edge cases. A new employee in a state where the business has no registration. Tipped wages and the tip credit. Fringe benefits that belong on a W-2. An employee misclassified as a 1099 contractor for three quarters.
Sales tax. Since South Dakota v. Wayfair in June 2018, over 40 states impose economic nexus, commonly at $100,000 in sales or 200 transactions, with real variation: Connecticut requires both tests met, New York uses $300,000 and 100 transactions, several states dropped the transaction count. Nothing in the client’s software watches this for them.
Clean-up. Eighteen months of auto-categorised transactions, an undeposited funds account holding $61,000 of fiction, a negative inventory balance, a prior bookkeeper who reconciled by forcing an adjustment. One of the most reliably profitable engagements in the trade, and it exists because of automation, not in spite of it.
Explaining it. The owner does not read financial statements. He reads his bank balance. “You were profitable in April and still can’t make payroll because $94,000 is sitting in receivables over 60 days” is the most valuable sentence most bookkeepers say all month.

Notice the pattern. Automation handles the transaction. It does not handle the exception, and it cannot tell which transactions are exceptions.

Why the person who owns the accuracy gets paid

One thing survives every wave of tooling: somebody has to be accountable for the numbers being right.

When a business applies for a loan, gets a sales tax assessment, sells, gets audited or has a partner dispute, the question is never “which software produced this.” It is “who stands behind it.” That cannot be automated, because it is not a task. It is a liability position occupied by a person.

It is also why the pay gap inside one job title is so wide. BLS puts employees between under $36,000 and over $74,550, and the top of the independent market is a multiple of that. The difference is almost never speed of data entry. It is whether the client believes you own the outcome. A bookkeeper who messages on the 8th saying “your gross margin dropped four points in July, it’s the new supplier’s freight terms, here are the two invoices” is not doing bookkeeping in the sense your CPA meant. She is very hard to replace at any price.

Practical consequence: carry errors and omissions cover, put your engagement scope in writing, and stop saying “I just do the books.” You are not just doing the books. You are the accuracy.

From hourly to a monthly fee, and what “advisory” honestly means

The strongest evidence the service has not died is what happened to how it is sold. The 2024 AICPA and CPA.com Client Advisory Services Benchmark Survey, covering 206 CAS practices and calendar year 2023 data, found median practice growth of 17%, median net client fees per professional of $156,250 (up 29% on 2022), and only 10% still using hourly billing as their primary pricing method. Practices with a defined industry niche reported 38% higher median CAS revenue. It is self-reported, from firms that already chose this model, so read it as the successful end of the market rather than the average. The direction is still unambiguous.

Hourly billing is the trap automation sets. Every efficiency you gain cuts your own invoice. Get 30% faster at a client and you have given yourself a 30% pay cut. Fixed monthly pricing inverts that: efficiency becomes margin, and the client stops flinching when they call you.

As for “advisory,” most of what is sold under that word is inflated. Here is what it looks like in a real engagement with a $1.4m excavation contractor:

Books closed by the 10th every month, without being chased
A rolling 13-week cash forecast, updated monthly, so he knows whether he can buy the second truck in March or November
A 30-minute call with three numbers on the screen and one decision to make
Job-level margin on the last ten completed jobs, which is how he found out his commercial work loses money
A flag when something is wrong before his CPA finds it in March

That is it. No dashboards nobody opens, no “CFO services” branding. Monthly, fixed-fee, worth several times a transaction-processing fee, and no product on the market does it.

Pick a niche the software does not understand

Generic small-business bookkeeping is the segment under real pressure, because generic is what a template handles. Specialisation is the strongest available defence, for a simple reason: general-purpose tools encode general-purpose accounting. Industry-specific judgment is exactly what they lack. Niches with genuinely unusual accounting, where the knowledge takes a year to build and then compounds:

Construction and specialty trades. Work in progress, over- and under-billings, percentage-of-completion revenue, retainage receivable, job costing with proper burden allocation, certified payroll, AIA billing. A contractor’s balance sheet is meaningless without a correct WIP schedule, and almost nobody outside the niche can build one.
Restaurants and bars. Daily sales journal from the POS, tips payable versus tip credit, comps and voids, prime cost tracking, inventory and waste. The operating rhythm is weekly, not monthly.
Ecommerce. Multi-channel revenue recognition, payout reconciliation across Shopify, Amazon, Stripe and PayPal, landed cost and inventory valuation, returns reserves, nexus across dozens of jurisdictions. The niche where clients most often arrive with eighteen months of nonsense.
Nonprofits. Fund accounting, donor restrictions and release from restriction, grant tracking, functional expense allocation, the Form 990. Structurally different from commercial accounting, and boards need it explained out loud.
Agencies and professional services. Project profitability, utilisation, pass-through and reimbursable costs, retainer revenue recognition, contractor mix.

Two clients in one niche make the third dramatically faster. Five clients in five industries make you slow forever. That is the whole argument.

What it pays, and who should actually leave

As an employee, the ladder is what BLS describes and it is not thrilling: a median of $50,670, top decile above $74,550, construction the best-paying major industry at $54,890 against retail’s $45,830. Roles labelled staff accountant, accounting manager or controller sit above the clerk classification, and that is the realistic internal path.

Independently the arithmetic changes, because you are no longer selling hours. Model it yourself rather than trusting anyone’s advertised averages: monthly fee per client, times the clients you can genuinely service at your quality bar, minus software, insurance, tax and subcontract help. Most solo practices run out of capacity somewhere between fifteen and thirty monthly clients, depending on niche and how much clean-up is in the mix. The lever is fee per client, not client count, and the only thing that raises it is being demonstrably better at one kind of business than a generalist can be.

Now the uncomfortable part. You should seriously consider leaving if: your work is overwhelmingly transaction entry for clients who chose you on price; you do not want to talk to clients and would prefer to be left alone with the data; you have avoided learning payroll, sales tax and any industry-specific accounting; or you are seven years in and still billing hourly for tasks the software already does. That position is not stable and it will not become stable.

You should stay, and expand, if: you like the diagnostic part, you are willing to own the numbers in writing, you can hold a conversation with an owner about his business rather than his ledger, and you are prepared to pick a niche and go deep rather than take whoever calls. That version of the job is not in decline. It is in short supply, which is a very different thing.

Skills worth building in the next twelve months, in order: payroll to the point of handling multi-state and edge cases; sales tax nexus and filing; one industry’s specialist accounting, properly; cash flow forecasting; clean-up as a packaged, priced engagement; and explaining a variance in two sentences to someone who has never read a balance sheet. Nothing on that list is threatened by a bank feed.

Your CPA was describing a job that is genuinely disappearing. He just wasn’t describing the one worth having.

Questions people ask next

Is a bookkeeping certification worth it? Credentials help with credibility when you have no client list yet, and with little else. Buyers almost never ask about them. What closes work is a specific answer to “have you handled a business like mine,” which is an argument for a niche, not a certificate.

Should I become a CPA instead? Only if you want the work accountants do. BLS shows accountants earn a higher median ($83,680 versus $50,670), but the occupation posts fewer annual openings than bookkeeping clerks. A self-employed bookkeeper with a well-priced niche book can out-earn a mid-level employed CPA. Different career, not a promotion.

Can I still start a bookkeeping business from scratch in 2026? Yes, on one condition: start in a niche, price monthly from the first client, and refuse hourly transaction work even when you are hungry. Generic and hourly means competing head-on with the exact thing being automated.

If you run a bookkeeping or accounting practice and the routine plumbing is eating the hours you would rather spend on client work, that is the part AB7 Solutions works on: automating document intake, categorisation and reconciliation workflows, wiring your ledger to the CRM and practice tools around it, and adding trained remote finance support for clean-up and month-end volume so you can take on niche clients without hiring ahead of the revenue. To talk through what is worth automating and what should stay with a human, call +1 321 341 7733, email ab@ab7solutions.com or director@ab7solutions.com, or read more at www.ab7solutions.com.

Sources: US Bureau of Labor Statistics, Occupational Outlook Handbook: Bookkeeping, Accounting, and Auditing Clerks and Accountants and Auditors (May 2025 wage data; 2025 to 2035 projections); BLS Employment Projections, Occupations with the largest job declines, 2025 to 2035; AICPA PCPS and CPA.com, 2024 Client Advisory Services Benchmark Survey (206 CAS practices, fielded May to July 2024, covering calendar year 2023, self-reported); Sales Tax Institute, South Dakota v. Wayfair economic nexus FAQ.