How Can I Automate Matching Remittance Advice From Emails and PDFs to Customer Payments?
Automating remittance matching starts with collecting all remittance advice in one place, where software can read the invoice numbers and amounts out of each email or PDF. Matching only works once those fields are cleaned up, for example by mapping a payer’s different names to one customer.

Automating remittance matching starts with collecting all remittance advice in one place, where software can read the invoice numbers and amounts out of each email or PDF. Matching only works once those fields are cleaned up, for example by mapping a payer’s different names to one customer.
In a typical case, the bank shows a lump-sum credit while the invoice detail sits in a PDF that arrived two days later, in an inbox nobody has opened yet.
Which remittance formats are hardest to extract?
Most B2B teams receive remittance through several routes at once, and each format fails in its own way.
| Source | What it usually contains | Where extraction goes wrong |
|---|---|---|
| Email with a PDF attached | A table of invoices with the amount paid on each | Merged cells and column headers that change between payers |
| Text in the body of an email | A short list of invoices, sometimes with a deduction note | Free-form wording, with invoice numbers mixed in among order or PO numbers |
| Automated email from an AP platform | A structured notice with a payment ID and an invoice list | The sender is the platform, so the email doesn’t name your customer directly |
| Spreadsheet attachment | One row per invoice, often with the buyer’s voucher numbers | Column names differ from payer to payer, and totals rows get read as invoices |
| Scanned check stub or lockbox image | Handwritten or printed invoice references | Handwriting and stubs that run over several pages |
| Bank file such as BAI2 or EDI 820 | Structured payment detail, sometimes with addenda | Addenda dropped by summary feeds, or references truncated by the bank |
Our guide to what remittance matching is covers the common formats in more detail.
How do you get every remittance into one place?
Give customers one address for remittance and print it on every invoice and in your payment instructions. Some customers will keep sending remittance elsewhere for a while, often as a reply to whoever chased the invoice last. Those replies sit in one person’s mailbox where nobody else can see them, so forwarding rules are needed to close the gap.
Base those rules on what remittance emails tend to look like. Subject lines often include phrases like “remittance advice” or “payment advice.” Many customers pay through AP platforms such as Tipalti or Bill.com, which send automated remittance emails from their own domains, so add those sender domains to the rules as well.
Someone also needs to own the shared inbox, since an address nobody checks only moves the backlog somewhere else.
Monk reads remittance from customer emails and AP portals. Anything that arrives another way, such as a lockbox file or a PDF a customer sent to one person, can be uploaded to Monk directly. Capture from portals and bank files is covered in more depth in our cash application guide.
How do you tell remittance apart from everything else in the inbox?
A shared AR inbox holds far more than remittance. Statements, invoice copies, W-9 requests, out-of-office replies and dispute emails all land there, but extraction should only run on documents that describe a payment.
Statements cause the most confusion, because they also list invoices with amounts. The quickest test is whether the document refers to a payment that has already been made. Remittance usually carries a payment date or a total that ties to a bank credit, and a statement has neither.
Classify each email and attachment before extracting anything, so dispute emails and document requests go to collections instead of into the extraction queue.
What should extraction pull from each email and PDF?
At a minimum, each remittance should yield:
- the payer name as written
- invoice numbers, or the buyer’s own voucher or PO references
- the amount paid against each line
- the payment date and currency
- any deduction or short-pay reason
Text in born-digital PDFs and spreadsheets can be read directly. Scanned check stubs and photographed remittances depend on OCR or a model that reads images, so their error rates run higher.
Tables that continue onto a second page are a common failure, because the header row only appears on the first one. Our comparison of invoice scanning tools explains how OCR differs from LLM-based extraction.
What cleanup has to happen before matching?
Extracted fields rarely line up with your ledger exactly. Before matching, the software should:
- strip prefixes and leading zeros from invoice numbers
- map each variation of a payer’s name, such as ACME CORPORATION and Acme Inc., to one customer
- bring amounts to the same precision and currency
- convert dates to one format
The payer alias list does the most work here. Without it, a subsidiary that pays under a slightly different name looks like an unknown customer every time. The list should also include the AP platforms that pay on your customers’ behalf, mapped to the customer each payment comes from.
How does matching work once the fields are clean?
Most payments then match on invoice number and amount. In Monk, a payment posts automatically only when it matches exactly and passes deterministic checks, so anything else goes to someone on your team for review, usually with a suggested match. About 80% of payments match automatically, rising to 95% once suggested rules are accepted.
For payments that cover several invoices, see our guide to auto-applying one payment to multiple invoices. The post on explainable cash application covers why some payments go to review.
What does one payment look like from email to match?
Take Northwind Freight, a fictional customer. An ACH credit for $41,615.00 arrives on November 3 with the bank description “NWF LOGISTICS LLC” and no addenda. That afternoon, an email from Northwind’s AP platform reaches the remittance inbox with a PDF attached.
The PDF lists four invoices:
| Invoice as written in the PDF | Amount | Note |
|---|---|---|
| 00412 | $12,400.00 | |
| 00418 | $9,850.00 | |
| 00431 | $11,200.00 | |
| 00437 | $8,165.00 | $435 deducted, damaged pallet, claim 7781 |
| Total | $41,615.00 |
Here’s how each step handles it:
- Classification. The email comes from a known AP platform domain and the PDF shows a payment date and total, so it’s treated as remittance.
- Extraction. The table yields four invoice references with their amounts, plus the note explaining the deduction.
- Cleanup. “00412” is normalized to 412, which matches INV-412 in your ledger, and “NWF LOGISTICS LLC” maps to Northwind Freight through the alias list.
- Matching. Invoices 412, 418 and 431 match at their exact open amounts. Invoice 437 is open for $8,600, so the $435 difference goes to a deduction case with the claim number attached, and the invoice stays open for that amount.
The PDF total ties to the bank credit, so the only item left for a person is the $435 deduction.
How should you roll out remittance automation?
Run the rollout in stages, and don’t move on until each stage meets its target.
| Stage | What happens | Example target before moving on |
|---|---|---|
| Weeks 1 to 2 | Publish the remittance address and set up forwarding rules, with one person owning the inbox | Remittance for most of last month’s payments can be found in the shared inbox |
| Weeks 3 to 4 | Run classification and extraction, with a person checking every result | Few corrections needed on digital PDFs |
| Weeks 5 to 6 | Match in review mode, so every suggested match needs approval | Suggestions accepted without changes most of the time |
| Week 7 onward | Let exact matches post automatically and preview ERP write-backs | No unexpected entries in the ERP preview |
Scanned images usually need longer in the review stage than digital PDFs. It’s fine to automate the PDFs first and keep the images under review until their error rate comes down.
How does the morning routine change?
Many AR teams start the morning with the bank portal and the remittance inbox open side by side. Someone downloads each attachment and keys the invoice numbers into the ERP by hand, while payments with no remittance wait until last.
Once extraction and matching are automated, most payments are already applied before anyone logs in. The morning is spent on the review queue, which holds suggested matches to approve and deductions to send to collections.
Corrections also become part of the routine. When someone fixes an extracted field or adds a payer name to the alias list, the next remittance from that customer needs less attention.
Who owns each part of the process?
| Task | Usual owner |
|---|---|
| Remittance inbox and forwarding rules | AR or cash application lead |
| Mailbox access and data retention | IT or security |
| Extraction errors and the payer alias list | Cash application team |
| Deductions and disputes found in remittance | Collections |
| Approving write-back to the ERP | Controller |
The alias list belongs with the cash application team, since they’re the first to see a new payer name and can add it before the next payment arrives.
How do you test it before go-live?
Before you rely on any setup, whether it’s built in-house or bought, put together a test set from your hardest recent payments. Include:
- a scanned check with a handwritten memo
- an email with several PDF attachments and notes in the body
- a wire that cites only a PO
- a partial payment explained in the email body
- a remittance table with merged cells
- the same remittance sent twice
- a payment in a foreign currency
- a customer statement, to check that it isn’t treated as remittance
For each case, check that the extracted fields are correct and that the match, or the reason it went to review, makes sense. Run the set again whenever a payer changes its format or a new one starts paying you.
Which metrics show it’s working?
| Metric | What it tells you |
|---|---|
| Remittance capture rate | The share of payments whose remittance reached the shared inbox |
| Time from payment to remittance | How long cash sits before anyone knows what it’s for |
| Extraction errors by source | Which formats or payers need attention |
| Manual corrections per week | Whether the alias list and rules are keeping up |
| Age of unapplied cash | Whether payments without remittance are being resolved |
Watch capture rate first, since extraction can’t help with remittance that never reaches the shared inbox.
What should you ask a vendor about email and PDF remittance?
These questions go further than a general AR checklist:
- Does the software read remittance directly from your mailbox, or do files have to be uploaded?
- Can it read remittance written in the body of an email, as well as attachments?
- Which file types does it accept, and how does it handle scanned images?
- How does it tell a statement apart from a remittance?
- When an extracted field is wrong, how does your team correct it, and does the correction carry over to that payer’s next remittance?
- How long are emails and documents stored, and who can see them?
Ask the vendor to run your test set during the evaluation and show the extracted fields for each case.
What controls should finance keep?
Mailbox access should be limited to the remittance and AR inboxes the software needs. Every match should keep a link to the email or document it came from, so an auditor can trace a posting back to its source. Write-backs to the ERP should be reversible, so a wrong match can be undone without a manual journal entry.
Monk is SOC 2 Type II and ISO/IEC 27001:2022 compliant. To see how it would handle the remittance in your inbox, book a demo.
Frequently Asked Questions
Can software read remittance from scanned checks?
Yes. Handwritten memos are the hardest part, so keep a person checking them for longer than printed stubs.
How should an email with several PDF attachments be handled?
Treat each attachment as a separate document but keep them linked to the same email. One may be the remittance and another a statement or a copy of the invoice.
Can the same setup handle EDI 820 files?
Yes. EDI 820 files are already structured, so the fields can be parsed directly. They usually match more reliably than emailed PDFs.
How do we protect payment data that arrives by email?
Bank details pulled from remittance should be encrypted and hidden in logs from anyone who doesn’t need them. Agree a retention period for stored emails with your compliance team before the software goes live.
What happens when a payer changes its remittance format?
Extraction accuracy for that payer usually drops. Monitor errors by payer, and review the first few remittances in the new format by hand.
What if the remittance total doesn’t match the bank credit?
Match the lines you can, then look for a bank fee or a deduction the remittance doesn’t mention. Whatever is left should go to someone with both the remittance and the bank record attached.
Is it worth building this in-house?
Extraction and alias maintenance need attention every time a payer changes format, so factor in ongoing engineering time. Our build vs. buy guide covers the trade-offs.