Vista Recon · how reconciliation works
What happens to the line you can't match.
Settlement reconciliation looks like magic until you see the mechanism. There is no model guessing per file and no remainder swept to a clearing account. Each channel reports in a fixed vocabulary of codes; that vocabulary is mapped to your chart of accounts once; and after that every settlement is assigned deterministically and foots to the deposit on its own.
Deterministic mapping · exceptions to a person · no auto-match percentage
Vista Recon · Amazon settlement
Reconciled81,342 line items · 38 Amazon codes → 10 GL accounts
| GL account | Amount |
|---|---|
| 4000Marketplace revenue | 482,644.27 |
| 2250Sales tax collected | 25,012.14 |
| 2251Facilitator tax remitted | −25,010.37 |
| 5050Marketplace commission | −81,831.10 |
| 5060FBA fulfillment & storage | −86,481.08 |
| 5070Advertising & deal fees | −14,016.30 |
| 4500Promotions & coupons | −4,086.41 |
| 1450FBA reimbursements | 5,556.52 |
| 5080Returns & return shipping | −1,071.75 |
| 6900Adjustments | 168.58 |
| Net deposit → bankties to the penny | 300,884.50 |
AI maps your codes to your accounts once. You confirm it. Every settlement after is assigned deterministically, before it imports: the same answer every time.
Real Amazon settlement, anonymized · every channel ships its own format, mapped once
The short answer
Reconciliation works in four moves, and only the exceptions need a person on every file.
- Every sales channel reports its payout in a fixed, finite vocabulary of codes: the same fee, tax, and adjustment labels recur on every settlement.
- Altura maps that vocabulary to your chart of accounts once, during implementation: AI proposes each code's GL account, a person on your finance team confirms it.
- After that, assignment is deterministic (the same code lands in the same account on every settlement, before anything imports) so the file foots to the deposit on its own.
- Only a genuinely new code surfaces, and it goes to a person as a classified exception, never swept into a clearing account to reconcile later.
The status quo
Most pipelines have one honest weakness: the line nobody mapped.
A marketplace deposit is not one number. It is thousands of orders, fees, refunds, reserves, and taxes the channel already netted before it paid you. The reconciliation question is not “did the cash arrive,” it is where does every component of that deposit belong in the ledger.
The common workaround is a clearing account. Whatever doesn't map cleanly gets swept into an “Other Deposits” or “clearing” bucket so the deposit nets out, and someone promises to explain it later. The books tie, technically, but the difference nobody explained is still sitting in that account at year-end, and it grows every month the queue outruns the people clearing it.
Hand-built flows have the same failure in a different place: a person keying deposits at ten at night so the bank rec ties before close. It works until the volume, or the person, runs out.
Why it is mappable
Every channel ships its own fixed vocabulary, and that is the whole opening.
A settlement file feels infinite because it is long. It isn't. However many order lines it carries, each one is labeled from a finite, repeating set of codes: the same fee, reserve, refund, and tax labels show up month after month. The labels differ from one channel to the next, which is exactly why a single generic connector can't reconcile all of them. Map each channel's vocabulary once, and the length stops mattering.
Amazon Seller (3P)
Settlement reports built from a fixed set of amount-types and amount-descriptions: referral and FBA fees, reserves, reimbursements, promotions, and facilitator tax, line by line.
Shopify Payments
Payout batches of balance transactions (order charges, refunds, processor fees, and adjustments) netted into each deposit.
Walmart Marketplace
Marketplace payouts with their own commission, refund, and adjustment labels, structured nothing like Amazon's.
Stripe
Balance transactions, per-transaction fees, and payouts, where the deposit batches charges and refunds that occurred on different days.
Amazon Vendor (1P)
Remittance advice instead of a settlement: short-pays, co-op chargebacks, and compliance deductions applied against open invoices.
PayPal / Braintree
Transaction-level fees and settlement batches that report on a different cadence than the orders they pay for.
A worked example
One real Amazon settlement, decomposed to the penny.
This is an anonymized Amazon settlement Altura reconciled. It runs 81,342 lines, but every line is labeled from only 38 codesacross 13 amount-types: Amazon's fixed vocabulary. Bucketed by code, the whole file collapses into 10 GL accounts that sum exactly to the $300,884.50 that hit the bank.
Read it as a decomposition, not a loss. The settlement carries $482,644.27 of marketplace revenue; the channel withheld commission, FBA fulfillment, advertising, promotions, and the tax it remits on your behalf, and returned the net. Vista Recon puts each of those components on its own GL line instead of leaving a single mystery deposit, so the deposit is explained, not just recorded.
01
81,342
settlement lines
One Amazon settlement, line by line.
02
38
Amazon codes
Across 13 amount-types: a fixed, finite vocabulary.
03
10
GL accounts
Where those codes are confirmed to land in NetSuite.
04
$300,884.50
net deposit
What hit the bank, and it foots to the penny.
The reconciled entry itself is the card at the top of this page.
The mechanism
Mapped once by a person, then deterministic on every file after.
The trust core is the order of operations: the judgment happens once, up front, with a person in the loop, and everything repeatable happens by lookup, not by guesswork.
Map the vocabulary, once
At implementation, Altura reads a representative settlement from each channel. AI proposes a GL account for every code in that channel's vocabulary, and a person on your finance team reviews and confirms the map before it is used.
Assign deterministically, every file
When the next settlement lands, each line is assigned by lookup against the confirmed map: the same code to the same account, before anything imports. It is a lookup, not a model re-scoring the file each month.
Foot to the deposit
Bucketed by code, tens of thousands of lines collapse to a handful of GL lines that sum to the net deposit. If it foots, it posts as one auditable NetSuite entry: the same entry on every run.
Route exceptions to a person
A line whose code is not in the confirmed vocabulary, a genuinely new fee or adjustment, is held back and classified by cause for review. A person decides, the map learns it once, and nothing is swept into a clearing account.
Why deterministic beats “fully automated”
A deterministic map covers every code you have confirmed, so there is no match rate to quote, and nothing to sweep.
“Fully automated” reconciliation usually means a model decided where lines go and the leftovers went to a clearing account. Deterministic mapping is the opposite claim: every code in the confirmed vocabulary already has a GL home, so the file foots on its own and the only thing that ever reaches a person is a code that is genuinely new: a real change in how the channel pays you, which a person should see. That is why Altura does not publish an auto-match percentage: structural completeness plus exceptions-to-a-human is both true and more defensible than a number.
Questions
How reconciliation works, answered.
Keep reading
Where this fits.
Vista Recon
The product this mechanism powers: settlement and remittance reconciliation into NetSuite.
Reconciliation cost calculator
Estimate the monthly hours and loaded cost your settlement queue takes by hand today.
NetSuite payout reconciliation
The method for matching marketplace, processor, and wholesale payouts to NetSuite activity.
Settlement Pattern Library
Anonymized operator patterns for payout, settlement, and remittance reconciliation.
Want to see your own settlement decomposed to the penny?
Send us one settlement file. We map it to your chart of accounts, reconcile it against NetSuite, and walk you through the posted entry, no integration to buy first.