AI accounting for an online seller

An online seller has better data than almost any other business. Every order timestamped, every fee itemised somewhere, every payout documented.

And the books are still hard, because the data arrives from four sources on four timetables in four formats, and none of them agrees with the bank without work.

Why revenue is genuinely difficult here

In a shop, revenue is what came through the till. Online, the same sale has at least four defensible figures.

What the customer paid. Including shipping they paid and vouchers they used. What the platform recorded as order value. Often before a platform-funded discount is stripped out. What was settled to you. Net of commission, transaction fee, shipping subsidy, advertising and any hold. What you can recognise. Which depends on delivery and the return window, not on when the money moved.

None of these is wrong. They answer different questions. The problem is that a business tends to pick one, use it everywhere, and then be surprised that margin, tax and cash never quite reconcile. See why your payout is not your revenue.

The four reconciliations that have to complete

Orders to settlement. Did every order you fulfilled appear in a payout, with the fees you expected? Settlement to bank. Did the payout arrive in full, on the date stated? Returns to refunds. Did every returned item produce a credit, and did the original fee come back or not? Stock to sales. Does the movement out of inventory match what you sold, across every channel?

The first two are where money is found. The third is where it leaks. The fourth is where the balance sheet stops being true.

Automation completes all four nightly. The manual version completes maybe two of them, monthly, when there is time.

What multiple channels actually costs you

Not effort in proportion to channels — effort in proportion to channel differences.

Each platform has its own fee structure, payout schedule, return policy, and definition of order value. Three channels means three reconciliation logics, three settlement calendars and three ways for a discount to be funded. See each marketplace has different fees and payouts.

The consequence is that a business at three channels usually has a house view of margin that is right on the largest channel and wrong on the other two — and the other two are where the growth is being planned.

Where the margin actually goes

Worth stating plainly, because the answer is rarely the one assumed.

Advertising, commission and shipping subsidy typically dwarf payment fees, and they are the three most likely to be posted as a single lump rather than attributed to the orders that incurred them. Attribute them per order and the picture changes: a proportion of any catalogue is being sold at a loss, and it is often the products the business believes are its best sellers, because volume attracts promotional spend.

That finding does not require AI. It requires the fee data to be itemised against orders, which is what a deep integration does and a shallow one does not — see why integration depth beats feature count.

The cash-flow shape nobody warns you about

Online selling front-loads costs and back-loads receipts. Stock is bought and paid for, advertising is charged immediately, platforms settle on a delay, and holds and reserves extend it further.

The result is that growth consumes cash faster than a shop does. A month of strong sales can leave less in the bank than a quiet one, which is alarming if the books have not made the timing visible and unremarkable if they have.

Automation helps here mainly by being current. A forecast built on reconciled data three days old is useful; one built on a spreadsheet six weeks old is decoration.

What to automate first

Your largest channel, order to payout, end to end. Not a partial sync — the whole chain, until payouts reconcile to the bank without a spreadsheet.

Resist adding channels two and three until the first completes. The reconciliation logic you build understanding on the first channel is what makes the others quick, and doing three shallowly produces three unfinished reconciliations instead of one finished one.

Common questions

Why is revenue hard to define for an online seller?

Because the same sale has several defensible figures — what the customer paid, what the platform recorded as order value, what was settled after commission, advertising and shipping subsidy, and what can be recognised given delivery and the return window. Each answers a different question, and problems arise when one figure is used for margin, tax and cash flow alike.

What reconciliations does an online seller need?

Orders to settlement, settlement to bank, returns to refunds, and stock movement to sales across every channel. The first two are where underpayments are found, the third is where money leaks through fees that never come back, and the fourth is what keeps the balance sheet true.

Does selling on more channels multiply the accounting work?

It multiplies it in proportion to how different the channels are rather than how many there are. Each platform has its own fee structure, payout timetable, return handling and definition of order value, so three channels means three reconciliation logics rather than three times the volume of one.

What should an online seller automate first?

The complete order-to-payout chain for the single largest channel, carried through until payouts reconcile to the bank without a spreadsheet. Adding further channels before the first one completes tends to produce several partial reconciliations rather than one finished one.

Related: reconciling sales across multiple marketplaces · cash vs accrual accounting for online sellers · managing cash flow across multiple channels