Vendor statement reconciliation compares a supplier's statement of account against your accounts payable subledger for the same period, classifies every difference, and resolves each to a documented outcome. The result is journal adjustments, applied credits, and a residual list carried forward. Matching is one step of seven, and the value sits in what fails to match.
Every accounts payable (AP) team agrees vendor statement reconciliation should happen. Almost none do it across the whole supplier base, and laziness is not the reason.
A supplier statement is that supplier's view of your account, and your subledger is yours. They disagree for mostly ordinary reasons, and telling those apart from the expensive ones is the job.
Timing at period cutoff produces the largest share of those differences. They invoiced on the 30th, you received it on the 3rd, and both records are correct. These resolve themselves, so identify them mechanically.
Then come the differences that do not resolve themselves. The supplier issued invoices that never reached you, lost to a wrong address or a silent portal upload failure. They raised credit notes you never applied, so the credit shows on their statement as a reduction and on yours as nothing. Disputed items sit open because nobody closed them. Partial payments landed against the wrong invoice, so both records show the right total and the wrong composition. Your ledger carries duplicate postings, and currency or tax differences leave their gross different from your net.
The Washington State Auditor put duplicate payments at 0.8% to 2% of total payments in 2022 research. Statement reconciliation surfaces them, because the supplier's records still show the credit.
Ardent Partners reported in its State of ePayables 2025 benchmarks, published January 2026, that 21.9% of AP staff time goes to supplier inquiries. Unreconciled accounts generate much of that queue.
Treating this as a matching exercise is the mistake. Two ideas make it tractable.
First, every unmatched item belongs to exactly one class, and that class tells you who resolves it and what evidence closes it. Without a taxonomy, each variance becomes a fresh investigation.
Second, direction matters. Items the supplier shows that you do not are a liability you may be understating. Items you show that they do not are frequently your own duplicate postings or an unapplied credit, and those are recoverable cash. Sort by direction before size, because the small item in your favor is often worth more than the large one you already know about.
Set a materiality floor and write it down. Chasing a $14 rounding difference across three emails costs more than the difference. Net everything below the floor into one write-off line per supplier per period, and set the floor by the loaded cost of an analyst hour.
Request statements on a fixed cycle rather than waiting for them to arrive, and specify a cutoff date so you compare to a known point. Prioritize by exposure, meaning open balance and volume. Keep the request, the statement, and its arrival date together, because arrival date determines which timing differences are legitimate.
Convert whatever arrived into one structure with supplier reference, document number, date, type, currency, and amount. Strip the formatting that breaks matching, meaning leading zeros, prefixes, and punctuation inside document numbers, while keeping the original string. Normalize sign conventions, since credits appear as negatives on some statements and in a separate column elsewhere.
Run three passes and stop at the first hit. Pass one matches on normalized invoice number. Pass two matches on amount plus date inside a tolerance window, catching the reference mutations that defeat pass one. Pass three matches on amount alone within the supplier, which needs review because suppliers bill identical amounts monthly. Record which pass matched, since pass three is weak evidence.
Assign every unmatched line a class from the taxonomy below, including the residue, which goes to unidentified rather than blank. An unclassified variance stays open next month.
Send each class to its owner as a batch with evidence attached. Routing single items to single people is what makes this expensive, because context switching costs more than the analysis.
Close each item against the evidence its class requires, set a standing deadline, and escalate whatever is open at the next cutoff.
Post the adjustments, apply the credits, write off what fell below materiality, and carry the remainder forward with an aging tag. A reconciliation ending in a spreadsheet rather than a journal entry has not finished.
Eight classes cover everything. Each entry names what it is, who resolves it, and the evidence that closes it.
Timing. The item was booked before cutoff on one side and after it on the other. AP resolves it mechanically. It closes when the item appears next period, or when a remittance postdates the statement.
Missing invoice on your side. They issued the invoice and you never received it. AP resolves it with the supplier. It closes on a copy of the invoice plus a posting reference.
Missing payment on their side. You paid the item and they have not yet applied it. The supplier resolves it, prompted by AP. It closes on remittance advice showing the reference and clearing date.
Unapplied credit. The supplier raised a credit note that was never applied to an invoice. AP resolves it. It closes on the credit note plus an application entry against a named invoice.
Disputed item. Someone contested the quantity, the price, or the delivery. The requester or category owner resolves it. It closes on a written outcome, either a credit note or an agreement to pay.
Duplicate posting. Your ledger records the same invoice twice. AP resolves it. It closes on both posting references plus the reversal entry.
Price or tax difference. Both sides show the same structure and a different value. AP resolves it, escalating to tax. It closes on the contract price, or on the tax treatment for that jurisdiction.
Unidentified. This class holds whatever is left after the seven above. The AP supervisor resolves it. It closes on reclassification into a real class, or on an approved write-off.
Watch the unidentified bucket. It measures whether normalization and matching are working, and it should shrink each month.
Published advice assumes you have the statements. Getting them is most of the work.
At enterprise scale you ask thousands of suppliers for a document every month, with no standard for how it arrives. Some send a PDF. Some send a spreadsheet with merged header rows and a subtotal band every fifteen lines. A meaningful share paste the statement into an email body as an untagged table, which defeats most automation because there is no attachment to detect.
Then there is chasing the suppliers who never sent one. An unanswered request needs a follow-up that references the first one, and that loop is why statement reconciliation stays an exception process rather than a standing control.
Statement reconciliation is the strongest opening use case for agentic AI in accounts payable, and the argument is structural.
Volume is high and the work per instance small, the shape where automation returns the most value. The input is unstructured and variable, which deterministic rules handle badly, so it is a fair test rather than a task scripting would have solved. The steps are describable as a standard operating procedure (SOP), so the behavior can be specified and audited.
Blast radius matters most. The output of a reconciliation is a proposed adjustment, a classified variance, and a chase email. None of those moves money. A mistake produces a wrong classification that a human rejects, costing minutes. Compare that with autonomous payment approval, where a wrong decision sends money out of the bank. When a finance team decides where an agent may act with limited supervision first, the right answer is the process with the highest volume and the lowest consequence per error.
Start in proposal mode, where the agent classifies each variance and drafts the chase email while a human approves every batch, then widen autonomy class by class.
Reconciliation is detective. It finds money after it has left, so it recovers losses rather than preventing them, and recovery is never complete. A duplicate payment found eight months later becomes a credit request against a supplier who may have applied it elsewhere. Prevention lives upstream, in duplicate gatekeeping at intake and a clean vendor master.
Reconciliation is also only as good as the statement, which is the supplier's assertion rather than independent evidence.
Coverage is the limit teams never state. Reconcile your top 50 suppliers monthly and everything else annually, and the long tail stays unexamined. That is where dormant credits accumulate, because small suppliers rarely chase money you owe.
Our platform treats a statement as another document arriving in the AP mailbox. Inside the AP Inbox Service Center, our AI recognizes a statement whether it arrived as a PDF, a spreadsheet, or a table pasted into an email body. It normalizes the statement and matches it against your ledger through the enterprise resource planning (ERP) connection.
What comes back is a classified variance list rather than unmatched lines, each item routed by class with the chase correspondence drafted. Your team approves, adjusts, or rejects each item. Because our Agents are built from an SOP, the classification logic is the one your controller wrote, and every decision leaves an audit trail. Our webinar on AI Agents for your AP inbox covers how teams phase this in without handing over approval authority on day one.
Take your twenty highest-exposure suppliers, run the seven steps by hand, and count how many variances you assign to each class. That distribution tells you which class to automate first, and it usually says something uncomfortable about unapplied credits. We cover the upstream half in vendor invoice management.
Vendor statement reconciliation compares a supplier's statement of account against your AP subledger for the same period, classifies every difference, and resolves each to a documented outcome. The result is journal adjustments, applied credits, and a residual list carried forward.
Reconcile monthly for suppliers with high balances or volume, quarterly for the middle tier, and at least annually for the long tail. That tail is where unapplied credits accumulate, because small suppliers rarely chase a balance in your favor.
An unapplied credit is a credit note the supplier raised that was never matched against an invoice on your ledger. It shows as a reduction on their statement and nothing on yours, so you keep paying gross while holding a balance in your favor. These are the fastest cash a reconciliation recovers.
Yes, and it is one of the more reliable detective controls, because the supplier's records show the second payment as a credit. The Washington State Auditor put duplicate payments at 0.8% to 2% of total payments in 2022 research.