How to Reconcile an Amazon Settlement Report, Step by Step

To reconcile an Amazon settlement report, start with the total and the deposit date, then sort every row by type: sales, refunds, fees, advertising, reimbursements, reserves and adjustments. Add the credits, subtract the debits, and check that the result equals the amount that reached your bank. A short payout is usually explained by one of four things: fees or refunds that were larger than the sales view suggested, advertising taken from the payout, a reserve that rose or was not released, or sales that Amazon has not yet made available. Find the row behind each gap, record it, and send anything unmatched to a bookkeeper. The routine below works one settlement at a time on any marketplace.
In short
- A short payout is usually traced to a fee, refund, reserve, advertising charge or deferred sale line, so sort the rows before you post anything.
- Check the amount column against the summary total first; if they differ, pull the report again.
- Amazon's older settlement report types are being removed, so confirm your tool uses the Flat File V2 report.
- A connector and an AI worker can build and check draft entries, but a bookkeeper should approve each one.
- For a single marketplace with low volume, a spreadsheet checklist may be enough.
Most sellers open the settlement report only when the bank deposit looks wrong. This guide sets out the fortnightly routine that answers that question, which lines usually explain a short payout, and where automation helps. The worked figures use US dollars because the mechanics are the same on every Amazon marketplace, whatever the currency.
What a short payout usually means
A settlement report is the statement behind one Amazon payout. Amazon issues it each disbursement cycle, which is usually every 14 days, and it lists each change to your seller balance during that period. The bank deposit should match the net of those lines. When it does not, the gap almost always sits in a fee line, a refund line, a reserve line, or in sales that have not yet been released to the balance.
Why a short payout costs money
Fees are the largest line for most sellers, so small rate differences add up. Most referral fee categories in the United States sit at 15%, the published range runs from 5% to 45% depending on category, and most categories carry a minimum of $0.30 per unit, according to a 2026 rate guide from Feedvisor. Fulfilment fees also moved in 2026. EcomBalance reports an average increase of $0.08 per unit sold, effective 15 January 2026.
A seller analytics firm, Nova Analytics, measured Amazon fees on 8 October 2026 and found that fees, net of reimbursements and credits, took a median of 31.3% of sales, with the middle half of accounts between 24.2% and 38.6%. That is one firm's sample, so use it as a rough range to test against your own report rather than as a benchmark.
Timing adds a second cost. Amazon's own Seller Central forum reply says that under the delivery-date policy, funds from orders are held until seven days after delivery. Seller Essentials, in 2026, reports that the policy took effect for most US and Canadian accounts on 12 March 2026. A deferred order can appear in the report with a blank release date, so a bookkeeper who reads only the payout will misstate the month.
The bigger cost is hours. Each settlement that is rebuilt from raw rows, and each timing difference that turns out to be harmless, takes review time. That time grows with every marketplace, every refund batch and every new fee type.
The manual way and the automated way
The table below compares a routine done by hand with one where a connector pulls the data and an AI worker prepares the draft entries. In both versions, a person approves what is posted.
| Task | Manual routine | Connector and AI worker routine |
|---|---|---|
| Getting the report | Log in each cycle, open the payments reports and download the file | Report found and pulled on a schedule, saved once per settlement ID |
| Sorting rows | Filter a spreadsheet by type and description | Rows grouped using a mapping table that the bookkeeper has approved |
| Refunds | Search for each order ID, sometimes in an earlier period | Refund rows matched to earlier orders; unmatched rows go to a review list |
| Reserves and deferred amounts | Track in a notebook or a separate report | Balance carried forward as its own line; movements flagged |
| Bank match | Compare the deposit against the report by eye | Deposit compared with the settlement total; differences listed |
| Posting | Type the journal entry | Draft entry created; bookkeeper reviews and approves |
| Who decides | Bookkeeper, after working through every row | Bookkeeper, working from a short exception list |
The fortnightly checklist
Run this list once per settlement. It works by hand today and it is the same list an automated system follows, so it is a good place to test whether your books tie out before you automate anything.
- Pull the settlement report for the period and note the settlement ID and start and end dates. The settlement ID is the unique key for that batch.
- Read the summary row at the top. It carries the total amount paid to your bank and the deposit date, and these two figures are your control total.
- Add up the amount column. It should equal the total in the summary row. If it does not, stop and pull the report again before doing anything else.
- Group rows by transaction type, such as orders, refunds, other transactions, service fees and adjustments.
- Within orders, split the rows by amount type and amount description, so that principal, shipping, commission, fulfilment fees and promotions sit in separate totals.
- Review refunds. Match each refund to an order, in this period or an earlier one, and note any refund with no matching order.
- Review reserves. Check the reserve released from the last period against the reserve held in this one.
- Review deferred and account-level balances. Anything not yet paid out goes on the carry-forward line, not into revenue or expense.
- Match the bank deposit to the control total by amount and date. Note any difference and its cause.
- Post the journal entry, list each exception with its row reference, and send the list to the bookkeeper for approval.
Amazon's column layout explains why the amount description matters. The report holds one row per transaction, with one column per fee type, and every fee, refund and adjustment appears as its own row. The fastest way to find a short payout is to total each amount description and compare it with the same line in the previous period.
Here is an illustration with made-up figures, not Amazon data. Suppose a US settlement shows sales of $10,000, referral fees of $1,500, fulfilment fees of $1,200, refunds of $400, advertising of $600 and a new reserve of $500, with $300 released from the last period. The net is $6,100. If the bank shows $5,900, the $200 gap is more likely a timing item or a sign error on a reserve or reimbursement line than a new fee. Check the reimbursement and adjustment rows first, then the reserve rows, and only then the fee rows. Each candidate is a specific row in the report, so the bookkeeper can confirm it in minutes.
How to automate the routine, step by step
Step 1: Get the report on a schedule
Sellers cannot request or schedule settlement reports. Amazon generates them automatically, and a tool finds them with the getReports operation of the Selling Partner API. The current file is the Flat File V2 settlement report, with the report code GET_V2_SETTLEMENT_REPORT_DATA_FLAT_FILE_V2, and Amazon says it replaces the XML report and the original flat file report. Amazon's changelog lists 11 November 2026 as the date those two older reports are removed. That date has now passed, so confirm which report code your connector, script or saved download uses.
Step 2: Connect the report to your ledger
Two layers are needed. The first is the general ledger, usually QuickBooks Online or Xero. The second is a settlement connector, such as A2X, Link My Books or Taxomate, which sits between Amazon and the ledger. A2X describes itself as reconciling settlement reports with QuickBooks, Xero and Sage. For each of these tools, see the vendor's current pricing before you decide. A small account can build the same logic in a spreadsheet with a mapping tab, but the mapping tab then becomes the thing that needs maintenance.
Step 3: Match the bank feed
Once a settlement deposits, the bank feed entry should be matched to the journal entry the connector created for that settlement. Amazon's own report gives the deposit date, so the match can be checked by date and amount. Any deposit with no matching entry goes to the exception list.
Step 4: Let an AI worker prepare and flag
An AI worker can read the V2 file, apply the mapping the bookkeeper approved, build a draft entry for each settlement ID and check that the draft totals equal the control total. It should flag four things: an amount description that is not in the mapping, a refund with no matching order, a reserve movement above a threshold the bookkeeper sets, and any bank difference. It posts the flags to a review queue, which can be a shared sheet, an email to the bookkeeper, or a message in a shared workspace.
A language model is useful for unclear descriptions, for reading a new email from a logistics provider, or for summarising why a row was flagged. It should propose a classification, not post one. Where a supplier invoice arrives as a scanned PDF, optical character recognition can pull the fields into the same review queue. A WhatsApp Business API message can alert the owner when a gap passes an amount the owner has set, which suits owners who already run daily operations from their phone.
Step 5: Approve and keep the trail
The bookkeeper approves each draft in the ledger. The record should keep the settlement ID, the draft, the reviewer, the approval date and the list of exceptions with their resolution. That trail is what an accountant asks for at month end.
What breaks, and how to prevent it
The report format is the first risk. Amazon's changelog says the older settlement report types are being removed, and a script still calling them will fail or return nothing. Prevention is a test run: Amazon's own changeover guidance points to validating the new report against a settlement you have already reconciled by hand before you switch over.
Duplicate pulls are the second risk. If the same settlement is loaded twice, the ledger double-counts it. Use the settlement ID as the unique key, so that a second load is rejected rather than posted.
Reserves and deferred amounts are the third risk, and the most confusing. A deferred transaction can have a blank actual release date, and some accounts see their reserve as one account-level figure rather than as individual transactions. Prevention is a separate balance line that is carried forward each fortnight, plus a monthly check that the carried balance matches what the seller account shows.
Mapping drift is the fourth risk. A new fee type or description appears, and it lands in the wrong account because it is not in the mapping. Prevention is an exception rule that stops the draft from posting any unmapped row, combined with a quarterly review of the mapping table.
Currency is the fifth risk. Each settlement is in the currency of its marketplace, for example USD for amazon.com and GBP for amazon.co.uk. A multi-marketplace seller should keep one ledger per currency, or convert at a documented rate, before totals are compared.
When not to automate it
Automation is not worth it for every seller. A single marketplace with a few dozen rows per settlement, reconciled by a bookkeeper in under an hour, may be better served by a checklist in a spreadsheet. A connector also needs a ledger to post into, so a business with no double-entry books should fix that first.
Automation also does not fix an unclear chart of accounts. If nobody can say which account a fee belongs to, an automated draft only makes the wrong answer appear faster. Settle the mapping with the bookkeeper before the connector runs.
Finally, an AI worker cannot decide whether a refund was valid, argue a reserve with Amazon, or settle how a fee is treated for tax. It prepares, checks and lists. The bookkeeper and the owner still make those decisions.
What it typically takes
The cost of this work is mostly time, not software. Setup usually goes through four stages: access, mapping, a parallel run and the switch. Access means a Seller Central user or developer permission that lets a tool read the reports. Mapping is a working session in which the bookkeeper assigns each fee type and description to an account, once. The parallel run compares the manual result with the automated draft for two or three settlements before anyone relies on the draft. After that, the fortnightly work is mostly reviewing the exception list.
How long each stage takes depends on how many marketplaces you sell on, how many fee types appear in your reports, and how clean your current ledger is. The free automation audit measures those inputs for your own account, which is what sets the time.
How AiStaffo would automate this
AiStaffo would connect the settlement data, either through the report feed or a connector your bookkeeper already uses, to your ledger and bank feed. Each fortnight the AI worker would pull the settlement, check its total, group the rows by type using a mapping the bookkeeper has approved, and build a draft entry for each settlement ID. It would match the bank deposit, carry forward reserve and deferred balances, and place anything it cannot match on a review list with the row references. The owner still approves the draft entries, decides how to answer Amazon on reserves and disputes, and confirms any new fee type before it is mapped. Book a free automation audit
Questions people ask
How often does Amazon send a settlement report?
Why is my Amazon payout smaller than my sales?
What is the DD+7 rule on Amazon?
Can an AI tool reconcile an Amazon settlement by itself?
Which accounting software works with Amazon settlements?
How do I know a settlement report is complete?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.






























