Automation closes the parts of the month that are deterministic: payout matching, transaction coding, fee classification, bank reconciliation. It stops at four places where a person has to decide what is true. Those four are costing policy, accrual judgment, the exception queue, and the final review that puts a name against the numbers. In a multichannel ecommerce business they consume most of the close time even when everything else runs clean.
Knowing where the boundary sits changes what you buy and what you staff.
APQC's Open Standards Benchmarking data on general accounting, reported by CFO.com in its Metric of the Month series and drawn from a survey of roughly 2,300 organizations, puts the median monthly close at 6.4 calendar days from trial balance to completed consolidated statements. Top-quartile organizations finish in 4.8 days or fewer. The bottom quartile takes 10 or more.
Those are large organizations with dedicated finance staff. A seller doing $2M to $30M with a bookkeeper and a fractional controller is often slower, and the reason is rarely transaction volume. It is that the inventory and settlement work has to happen before the close can start at all.
Software applies a costing method. It does not choose one, and it does not decide what belongs in the unit.
FIFO or weighted average is a policy decision with real consequences for reported margin and for the balance sheet. So is what counts as landed cost. Ocean freight, obviously. Duty, obviously. Inbound drayage, usually. Customs broker fees, usually. Inspection at origin, arguable. Demurrage on a delayed container, arguable and consequential.
Each of those is a judgment about whether a cost brought the goods to their present condition and location. Once decided, the software applies it consistently forever, which is exactly what you want. But somebody has to decide, document it, and hold the line when a container arrives with an unusual charge.
The month ends. Three things are true and none are in the ledger.
A supplier shipped goods on the 29th under terms where title passed at origin. The container is at sea, unbilled and unreceived, and it belongs on your balance sheet.
An advertising invoice for the last week of the month will arrive on the 8th of the following month. Spend happened. Expense belongs in the month it happened.
A marketplace held back a reserve that will release in two weeks. It is a receivable, not a reduction of revenue.
Automation can be configured to accrue for known recurring items. It cannot decide the boundary cases, and the boundary cases are where a close goes wrong.
Every automated system produces items it declined to handle. A payout that does not tie. A fee type it has not seen. A refund that cannot be matched to an original order. A negative inventory quantity.
The queue is the most important artifact the software produces and the one most often ignored. A queue that is always empty means the confidence threshold is set too loose and the system is guessing. A queue that grows month over month means the books have quietly stopped being reliable and nobody has noticed yet.
Someone has to look at the finished statements and ask whether they are plausible. Gross margin moved four points. Why. Inventory rose while sales rose faster. Does that make sense. The fee ratio on one channel changed. Did the marketplace change a schedule or did something break.
This takes forty minutes and catches more errors than any control in the stack, because it is the only step where a person compares the output against their own knowledge of the business.
Four channels, 9,400 orders, one warehouse plus marketplace fulfillment. Everything synced. Every payout reconciled to the penny. The close looks finished on day three.
The statements say:
Four items were never touched by automation.
A container that landed on the 27th. Goods received, supplier invoice not yet entered. 8,400 units at $6.20 supplier cost, plus $19,320 of freight and duty. Inventory understated by $71,400, and once landed cost is allocated the true unit cost is $8.50, not $6.20.
Returns not split. 341 units refunded. 218 came back sellable, 123 did not. At $8.50 landed cost that is $1,853 that belongs back in inventory and $1,045.50 that belongs in a write-off account. The automation booked the refund against sales and stopped.
A transfer in flight. 1,600 units left the warehouse on the 28th for a fulfillment center, received on the 3rd. At $8.50 that is $13,600 sitting in neither location on the last day of the month. Absent an in-transit treatment they vanish from the count.
A TikTok Shop reserve. $8,420 held back, posted as a reduction of revenue rather than a receivable.
Corrected, gross margin moves to roughly 41.2 percent and inventory on the balance sheet rises by $86,853. Every one of those adjustments required a human to know that a thing happened. None of them were classification errors. The automation did its job correctly and the statements were still wrong, which is the point.
For sellers on newer channels this bites hardest, because reserve and fee behavior is least familiar. Getting TikTok Shop accounting posted correctly is a good example of a task where the mechanics are automatable and the treatment decision is not.
Automate to the edge of ambiguity, then stop deliberately. Set confidence thresholds so that anything the system is unsure about lands in a queue rather than getting a plausible guess. Silent posting is not accuracy, it is invisibility.
Name the reviewer. One person, with authority to fix the cause rather than clear the flag. This is the single highest-return decision in the whole process.
Sequence inventory before the ledger. Receipts, transfers, returns disposition, and count adjustments all have to be recorded before COGS is meaningful. Most slow closes are slow because this happens last.
Keep a standing accrual list. The same six items come up every month. Write them down. The list turns judgment into a checklist for the routine cases and reserves attention for the unusual ones.
Measure the queue, not the speed. A three-day close with 400 unresolved exceptions is worse than a seven-day close with none. Track the count and the age of open items alongside the calendar days.
The human work above is only tractable if the mechanical work is genuinely finished. That means each marketplace payout broken into transactions and fee types rather than posted as a lump sum, FIFO costing applied per transaction rather than estimated at period end, and stock tracked by warehouse with transfers held as in transit until received.
That is the job ConnectBooks does across Amazon, Shopify, Walmart, eBay, and TikTok Shop into QuickBooks or Xero, and it is also why the company publishes an accuracy guarantee rather than an accuracy claim.
ConnectBooks has announced Crunch, an AI CFO built on that reconciled data, with a waitlist open ahead of its release. An analysis layer belongs after the close, reading numbers a person has already signed off on.
Do not aim for a close with no humans in it. Aim for a close where humans only touch the four things that require judgment, and where the queue that surfaces those four things is small, aged in days rather than months, and owned by someone with a name.
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
