The question gets asked two ways. A seller asks it to find out whether they can stop paying someone. A bookkeeper asks it to find out whether to retrain. The answers differ, so both are below, with the labor data attached rather than an opinion.
The U.S. Bureau of Labor Statistics, in its Occupational Outlook Handbook entry for bookkeeping, accounting, and auditing clerks last updated August 28, 2025, projects employment in the occupation to decline 6 percent from 2024 to 2034. The base was 1,613,400 jobs in 2024, projected at 1,519,100 in 2034. Median pay was $49,210 per year as of May 2024.
The same entry projects about 170,000 openings per year on average over the decade, driven by workers leaving the occupation or retiring.
As slow displacement of a task, not sudden removal of a role. A decline of roughly 94,000 positions across ten years, against 170,000 annual openings, describes an occupation that shrinks while still hiring heavily. That is not the shape of a job being eliminated. It is the shape of a job whose entry-level portion is being absorbed while the rest persists.
No, and that is the useful context. Bank feeds, rules engines, and automated matching have been eroding manual data entry since well before language models existed. The current wave extends the same trend to slightly harder classification problems. The occupation has been declining across multiple BLS projection cycles for the same underlying reason.
Manual transaction entry. Rule-based categorization. Matching a deposit to an invoice. Chasing a receipt. Reading a PDF bill and putting the numbers in a form. Producing a standard monthly report pack.
These have a common shape. The correct answer is fully determined by information already sitting in a system, and there is one right answer per item. Software handles that well and gets cheaper at it every year.
Deciding costing policy. Judging whether an accrual is required and for how much. Determining whether a returned unit is sellable. Investigating why the inventory account will not tie to the warehouse count. Explaining a margin swing to a lender who is skeptical. Choosing a sales tax position and defending it. Fixing the mess after a channel migration.
None of these are hard because the arithmetic is hard. They are hard because they require deciding what is true when the records disagree, which is the actual content of the job.
Ask which 90 percent, measured how. If the measurement is transaction count, a high number is plausible and uninformative, because most transactions are small and easy. If the measurement is time spent by an experienced bookkeeper on a real ecommerce close, the number falls sharply, because the time goes to the small share of items that are ambiguous.
A seller with 14,000 monthly orders may have 13,800 that need no human attention and 200 that consume the entire close. The split between the automatable half and the stubborn half is covered in more detail in what AI bookkeeping scales and what it does not.
Some can. If you sell on one channel, hold no inventory or a single stable SKU, do not import, and have no third-party logistics relationship, well-configured software plus a CPA at year end is a defensible setup.
If you run several marketplaces, hold physical inventory, import goods, or use a third-party warehouse, cutting the human is a decision you will discover the cost of later. The usual moment of discovery is diligence, when a buyer asks why the inventory on the balance sheet does not match a count.
Move them off entry and onto control. Reviewing the exception queue. Owning the monthly tie-out between the inventory account and physical stock. Confirming that fee ratios match the published marketplace schedules. Approving accruals and write-downs. Reading the chart of accounts once a year with a skeptical eye.
That work is worth more per hour than the coding they used to do, and it is the work that keeps the automated part honest.
Ask them to explain, without looking it up, how a marketplace settlement becomes journal entries and what happens to a reserve balance. If the answer treats the payout as revenue, keep looking. ConnectBooks maintains an ecommerce accountant directory for sellers who want someone who has already seen this problem.
Toward specialization, yes. The generalist who codes transactions across twenty small clients is doing the work most exposed to automation. The bookkeeper who can reconcile a marketplace settlement to the penny, allocate landed cost across a mixed container, and defend an inventory balance to a lender is doing work that has more demand than supply.
Ecommerce is a good place to specialize precisely because the automation is worst there. Physical goods, multiple channels, and imported cost layers create the ambiguity that software cannot resolve alone.
Settlement reconciliation for the major marketplaces. FIFO and weighted average costing applied per transaction. Landed cost allocation. Inventory tie-outs across multiple locations including marketplace fulfillment centers. Sales tax mechanics for marketplace facilitator states. Working inside QuickBooks and Xero rather than around them.
A seller closes March. The software classifies 11,842 transactions without help and reconciles four channel payouts to the penny. Genuinely useful work, done in minutes.
What remains on the bookkeeper's desk:
Four items. Perhaps five hours. They determine whether March's gross margin is right, and none of them are classification problems. The reason they are tractable at all is that the settlement and cost data underneath them is reconciled at transaction level rather than summarized, which is the same marketplace accounting foundation everything else in this article assumes.
AI will not replace bookkeepers. It will keep removing the least skilled hours from the job, which shows up in the BLS projection as a slow decline rather than a collapse, and it will raise the value of the hours that remain.
For a seller, that means the question is not whether to have a person. It is what to have them do. For a bookkeeper, it means the safest position is inside the problems software handles worst, which for the next several years means inventory, multichannel settlements, and cost.
ConnectBooks has announced Crunch, an AI CFO built on reconciled seller data, with a waitlist open ahead of its release. It is designed to answer questions about numbers a human already made correct. That ordering is not a limitation of one product. It is the shape of the whole category.
Running an e-commerce business comes with plenty of challenges, but ConnectBooks is here to make your life easier. With real-time insights, seamless integrations, and detailed tracking of your profitability and inventory, you can stay ahead of the game. Whether you’re selling on Amazon, Shopify, Walmart, TikTok or eBay, ConnectBooks helps you manage your finances with 100% accuracy and confidence, so you can focus on growing your business.
Ready to level up? Start making smarter, data-driven decisions every step of the way. Try ConnectBooks Free Today or Schedule a Demo
