What Amazon's Smart Match actually does — and the big recovery opportunity it hides
Most shortage management focuses on what you can see: the claims Amazon raises, the dispute window, the recovery rate. There is another category that does not appear in any of those reports — amounts that fell out of Amazon's receive process without surfacing as shortage claims at all. Not disputed because they are never visible as disputes. Not tracked because they do not appear in shortage data. Call it the ghost opportunity: deductions hidden in the receive and reconciliation process that do not show unless you deep-dive the data. Amazon's Smart Match shortage reconciliation is where most of this disappears — and understanding where its logic breaks down is how you find what you are missing.
How Amazon receives your stock
When a shipment arrives at an Amazon fulfilment centre, auto-match is the first system to process it. The accurate ASN arrived ahead of the delivery. The driver made the appointment. Cartons intact, labels pristine, barcodes match. Amazon's execution muscle takes the goods through and closes the invoice.
This is the happy path. It covers approximately 90% of all inbound stock.
Smart Match handles the rest.
What is Amazon Smart Match?
Smart Match is Amazon's iterative shortage reconciliation algorithm. It runs when auto-match cannot close a position — when received units cannot be matched to the correct invoice. It applies tens of sequential matching passes across Amazon's internal data — invoice dates, received dates, quantity variances, invoice age, time thresholds — and resolves the majority of outstanding shortage positions by 35 days after invoice due date.
Those 35 days are why Amazon advises vendors to wait before filing disputes. Not courtesy — Smart Match is still running.
Does filing disputes early feed Amazon's Smart Match?
No. Dispute submissions are not an input to Smart Match. The algorithm reads Amazon's internal data only. Your filings, your documentation, your escalations — none of it touches Smart Match. Filing early does not accelerate anything.
File after the 35-day cycle. What remains at that point is what Smart Match could not close — and that is the only population your dispute team can actually work on.
Where receive errors come from
Receive errors split into two types.
Systemic process errors start before the dock. Missed delivery appointments. Duplicate ASNs. Last-minute ISA changes arriving after Amazon's systems have committed to a booking. Amazon's own scheduling windows closing — sometimes automatically, sometimes as a contract lever. Delays in stow after receipt. These show up in booking data and carrier logs. They follow patterns. They are findable.
Cognitive errors at receive are harder. They happen in seconds, under throughput pressure.
I worked directly with Amazon's FinOps team on enhancements to the auto-match and Smart Match logic. The errors that generate the most volume follow recognizable patterns.
A vendor ships 5-packs of a leading brand's notebooks in transparent film wrap. The FC associate scans the outer unit and counts the pack as one instead of five. Natural inference under speed — but systematic. Pull receive data by associate and station and the undercount appears as a consistent ratio, not noise.
A mixed carton arrives with four large blue T-shirts, three small blue, three red. The associate groups visually by the first product seen, scans one, counts the rest as the same variant. Seven large blue received against four shipped. Three small blue shorted. FC associates are measured on items-per-hour SLA, and the incentive structure and product complexity combine to produce this predictably. PACMAN builds a confusability index to flag which products are at highest risk before errors accumulate.
How Smart Match reconciles — and where it's too conservative in paying out
Every receive error auto-match cannot close creates an RBNI position: Received but Not Invoiced. RBNI positions arise when goods are received against any available PO — sometimes a phantom PO — rather than the correct one. Goods are in the warehouse; the invoice sits unmatched. Smart Match works back through these positions, applying matching criteria in sequential passes, to close them.
The majority of positions close. But Smart Match is overly conservative where it cannot confirm vendor attribution. This comes from the days when Amazon actively bought from distributors as lowest cost — a distributor might hold stock from multiple vendors, and a receive error could result in Amazon paying the wrong supplier. Better to not pay than to pay the wrong one. That conservatism still sits in the algorithm.
It also degrades where the matching signal is weak: an invoice outside the expected age window, a quantity variance in the ambiguous range, two RBNI positions both plausibly consistent with the incoming stock but indistinguishable by the available criteria. A timing slip cascades into a sequencing failure. A system misalignment produces an exception the algorithm cannot resolve. These positions stay open.
Amazon does not hold them indefinitely. Unresolved RBNI is eventually put through Amazon's P&L — an internal adjustment.
Your hidden opportunity
A small GB vendor with a modest known shortage bucket found a hidden opportunity twice its size — amounts they had no visibility on and were not disputing. A large EU vendor with substantial known exposure found an equal amount again sitting on top of it.
PACMAN was built by founders who engineered these systems from inside Amazon — those insights are baked into how it works.