Image
Workflow
Finance

Invoice intake and exception triage

How Mercury Digital’s agents read, match and code invoices, sort exceptions by type, and route every payment decision to a named approver.

The short answer

Short answer

The best way to use AI for accounts payable is to let an agent do the intake and matching and send only the exceptions to people.

With Mercury Digital, the agent reads each invoice, extracts and normalizes it, matches it to the PO and receipt, and codes it from written rules. Exceptions and payments go to a named approver.

With agents

What the agent does, and what it never does

Does: captures invoices from every inbound channel with their origin recorded, extracts and normalizes fields from PDFs and emails, runs the match, codes from written rules, files the record as work happens, and keeps a queue showing who has each open item.

Never does: approves a payment, changes vendor bank details, or guesses when a rule doesn't cover the case. Those go to a person.

Exception types and where each one goes

The value is in the exceptions. A clean three-way match doesn't need much intelligence; a mismatch does. Each exception arrives with the failed field named and the source documents attached.

ExceptionWhat the agent checksWhat it sends to the personRouted to
Possible duplicateSame vendor, amount, invoice number or date pattern against paid and open invoicesBoth invoices side by side with the matching fields highlightedAP lead
PO mismatch (price or quantity)Invoice lines vs PO linesThe line, the PO value, the invoice value and the differenceBuyer / PO owner
Missing receipt (no goods receipt)Whether a receipt exists for the PO lineThe PO, the invoice and who orderedRequester or receiving
No PO / non-PO invoiceWhether a PO is required under your policySuggested coding from rules, plus the vendor's historyBudget owner
Unknown or changed vendor detailsVendor master vs invoice (name, tax ID, remit-to, bank details)The differences, flagged as a possible fraud risk. Nothing is auto-updatedController (always)
Coding the rules don't coverWhether a written rule appliesThe draft coding left blank, with the question statedAP lead (who then writes the rule)
Above approval thresholdAmount vs your approval matrixThe invoice plus the full match resultNamed approver for that threshold

How it works, step by step

01

Intake with origin

Invoices arrive by email, upload or portal. The agent records where each one came from.
02

Extract and normalize

It reads PDFs, scans and email bodies, extracts header and line fields, and normalizes vendor names, dates and amounts.
03

Match

It runs a two- or three-way match against the PO and receipt. If the match fails, it names the field that failed.
04

Code from written rules

It applies GL coding from rules you've written down. Where no rule covers a case, it asks instead of guessing.
05

Triage exceptions

Each exception gets a type from the table above and goes to the owner for that type.
06

Hold the human gate

Exceptions, and any invoice above the threshold, go to a named approver. Payment approval stays with a person.
07

File as work happens

The invoice, match result, coding and approval are filed together with the approver's name and timestamp. Show queue status by person. Anyone can see what's waiting, and on whom.

Integration

Works with your accounting system (e.g. QuickBooks, NetSuite)

Through its API

The agent works with your accounting system (e.g. QuickBooks, NetSuite) through its API.

Writes only what you approve

It reads the POs, receipts and vendor master the workflow needs and writes only the entries you approve.

Scopes agreed in the audit

The access scopes are agreed during the audit, before anything is built.

Where this fits

Workflow / Automate a Workflow

Usually a Tier 1 engagement (Automate a Workflow). It runs in shadow mode next to your current AP process before any write access is granted.

What to measure. The auto-match rate (share of invoices matched without a person), exception turnaround (time from an exception being raised to its owner resolving it), exceptions by type, and invoices waiting at the payment gate. Mercury sets a baseline for these in week 1 and reports against it.

AP automation software vs a Mercury Digital agent

Tipalti, Stampli and the AP capabilities inside ERPs like NetSuite are licensed products with their own invoice capture, PO matching, approval and payment features. Tipalti describes a finance automation suite covering AP, global payments, procurement and expenses, with AI agents for invoice capture and PO matching (Tipalti). Stampli describes a procure-to-pay platform whose AI assistant, Billy, codes, matches and routes invoices (Stampli). NetSuite's AP describes invoice capture, matching bills to POs and receipts, approval workflows and payment automation inside the ERP (NetSuite Accounts Payable). Mercury Digital builds a governed custom agent around the process you already run, including the exception rules and the approvers you already have.

QuestionAP automation productMercury Digital agent build
What you getA product your team adoptsAn agent workflow built around your current AP process
ExceptionsHandled with the product's workflowTyped, routed to your named owners, and logged with evidence
Payment controlThe product's approval chainA human gate on every payment and on every exception, closed by default
Fits best whenYou're ready to standardize AP on a dedicated product or your ERP's AP moduleYour rules are spread across inboxes, spreadsheets and an existing ERP

Product descriptions checked against the vendors' own pages on Oct 27, 2026. Features and packaging change, so check each vendor's site for current details.

Tier 1 / Prove
How It Works

Questions, answered

Let an agent handle intake, extraction, matching and coding, and send only the exceptions to people. Payment approval stays with a named person.
AI Opportunity Audit

Sort the exception pile into rules and judgement

It starts with the AI Opportunity Audit, which is fixed-price and takes under 20 hours of your team's time.