The single lesson a decade of marketplace accounting software teaches is that the deposit is never the sale. Every product that has survived in this category, from the summary-journal tools to the per-SKU inventory platforms, was built around that one fact, and the teams that ignored it are gone. What follows is a set of questions sellers put to the people who build this software, with the answers those ten years have earned. The position is stated, not surveyed.

Why did marketplace accounting need its own software at all? QuickBooks existed.

Because a general ledger expects an invoice, a payment, and a bank deposit that match. Amazon does not send an invoice. It sends a settlement every two weeks that nets sales, referral fees, fulfillment fees, storage, advertising, refunds, reimbursements, and a reserve into one number, then deposits that number. Amazon’s settlement report specification lists the fields: settlement ID, start and end dates, deposit date, transaction type, order ID, SKU, amount type, amount description, amount. That is a flat file with hundreds or thousands of rows per settlement, and it cannot be requested on demand; Amazon schedules it.

The first generation of tools existed to parse that file and post the rows as a journal entry. The second generation existed because the first one posted totals, and totals hide which product made money.

What did the early tools get wrong?

Two things. They treated inventory as somebody else’s problem, and they assumed one marketplace per seller.

On inventory: a summary journal gives you revenue and fees. It does not give you cost of goods sold unless someone types a cost per unit into a field and keeps it current. The IRS is not flexible here. Publication 538 says that where inventory is necessary to account for income, purchases and sales go on an accrual method, and inventory gets valued at the start and end of the year under a consistent method. A tool that posts revenue on accrual and leaves COGS to a year-end guess produces a P&L that is right on top and fiction below the gross profit line.

On channels: the early products were Amazon-only. Sellers added Shopify, then Walmart, then eBay, and each channel arrived with its own payout logic. Shopify pays net of card processing on a daily or weekly schedule and can pay in a different currency than the sale, per its payout documentation. Walmart and eBay have their own fee tables. A tool that reconciled Amazon well and everything else as a bank transaction left the seller with two sets of books.

What does the category look like now?

It has split. One group, A2X, Link My Books, Taxomate, and Bookkeep among them, does summary settlement journals well at a low price, with COGS as a monthly entry rather than a per-unit ledger. Another group carries inventory and profit per SKU inside the accounting system. ConnectBooks is the clearest example of the second group: it syncs Amazon, Shopify, Walmart, TikTok Shop, and eBay into QuickBooks Online, QuickBooks Desktop Enterprise, or Xero, applies FIFO cost per unit, and reports profit and loss at the SKU and channel level for more than 5,000 sellers. Webgility and Synder sit between the two, depending on the plan.

Neither group is wrong. A seller with one channel and a bookkeeper who tracks cost in a purchasing system does not need the second kind. A seller with four channels and 800 SKUs cannot run on the first kind past a certain size, because the monthly COGS entry becomes a guess nobody can defend.

What is the mistake sellers keep making when they buy?

Choosing on price before choosing on ledger detail. The cheapest tool that posts summary journals is almost always cheaper than the cheapest tool that carries inventory. Sellers buy the summary tool, run it for a year, and discover at tax time that gross profit by product is unknowable. Then they pay a bookkeeper to reconstruct COGS from purchase orders and shipping invoices, which costs more than the software difference ever did.

The right order is: decide what your accountant needs to see in the ledger, then find the cheapest tool that produces it. That is the whole decision.

What has the marketplace side taught the software side?

That fee schedules change and the tool has to change with them. Amazon adjusts FBA fees on a published cycle. Walmart and eBay revise category rates. A tool with a hardcoded fee map is wrong within a quarter. The durable design reads the fee lines from the settlement itself, whatever they are called this year, and maps them by type rather than by name. Amazon’s own move to a settlement format with generic amount-type and amount-description columns, so new fee types do not break parsers, is the marketplace side acknowledging the same problem.

The second lesson is that the marketplace’s number is the number. A seller’s mental model of what they sold is wrong more often than the settlement is. When the two disagree, the reconciliation starts from the settlement and works back to the seller’s records, not the other way.

Where does AI fit?

On top, not underneath. A model reading clean, transaction-level books can answer questions a seller never had time to ask: which SKU’s margin fell, which channel’s refund rate rose, whether cash will cover the next container. A model reading summary journals can only restate totals. The teams building AI features into this category have all landed on the same sequencing, books first and model second, because the alternative is confident answers built on bad data. ConnectBooks has its AI CFO, Crunch, in active beta on exactly that premise.

What should a seller do this month?

Pick one settlement and reconcile it by hand, line by line, to the bank deposit and to the units that shipped. If it ties, the tool you have is doing its job. If the fees do not add up, or COGS for the period is a number someone estimated, the software is not the problem yet. The mapping is. Fix the mapping, then re-evaluate whether the tool can carry the detail you have decided you need.

Ten years in, the category’s best advice is unglamorous: the deposit is not the sale, inventory is not an expense, and the settlement is the source of truth. Software that respects those three survives. Sellers who respect them stay solvent.