Cash application is one of those finance tasks that sounds simple until you're the one doing it. You receive a remittance advice from a customer, figure out which invoices they're paying, match the amounts, account for any early payment discounts, and post it to your ERP. Multiply that by 50, 100, or 300 remittances a week, many arriving as PDFs across different inboxes, and you have a full-time job that's almost entirely manual data entry.
This post walks through how AR teams are using AI document extraction to automate that process. The tool we'll reference is Lido, an AI-powered document data extraction platform that reads any document format without templates and outputs structured data your ERP can import directly.
The honest answer is that remittance advices are a nightmare to standardize. Unlike purchase orders or invoices, which at least trend toward some predictable structure, remittance advices are produced by whatever AP system your customer happens to run. One customer sends a clean PDF with a table. Another pastes the data into the body of an email. Some send multi-page packets where the actual remittance is buried behind a debit memo and two pages of terms.
Template-based OCR tools, which have been around for years, require you to define a layout for each document type. You specify that the invoice number lives in column three, row four, that the vendor name is in the top-right corner, and so on. That works fine if you have one customer who never changes their format. In practice, you might have 40 customers each with different formats, and when any one of them changes their AP system, your template breaks.
So most companies fall back to the thing that always works: a human reads the document and types the data into the ERP one field at a time.
We talked to a manufacturing company running Global Shop Solutions that had exactly this workflow before their new CFO arrived. Someone on the AR team would receive remittance advices, print them out, key each payment into the ERP, and then file the physical paper in a cabinet. This wasn't a small operation. They were processing enough volume that this was essentially one person's entire job. The CFO's first question was why the printer was running all morning.
If you've done cash application, you know the drill. For each remittance advice, you're extracting:
- Vendor or customer name
- Their vendor number (if it even appears on the document, which it often doesn't)
- One or more invoice numbers being paid
- The original invoice amount for each line
- Any early payment discount taken
- The amount actually paid per line
- The check number or ACH reference
- The check date or payment date
Then you're typing all of that into whatever your ERP's cash receipts or payment posting screen looks like. If you have 15 invoices on one remittance, that's 15 rows of data entry. If the vendor name on the document is "iTool" but your ERP has them as "iTool Co." or "I-Tool Company", you're also doing a lookup to figure out who this actually is before you can post anything.
Vendor name mismatches are more common than you'd think. The same company might appear differently across documents depending on how their AP team entered their own company name, whether they're using a DBA, or just whether someone abbreviated it. None of these trigger an obvious error. You just have to know.
The approach that solves this is AI-based extraction that reads the document the way a person would, understanding context rather than matching coordinates. Lido uses this approach, which means it can pull the right data from a remittance advice regardless of whether the layout matches anything it's seen before.
You upload or route your remittance advices into Lido and configure what fields you want extracted. For cash application, that's typically: invoice number, invoice date, original amount, discount amount, amount paid, check number, check date, and vendor name. Lido's auto-field detection can identify these from your documents automatically, or you define them yourself.
Where it gets useful is the AI column feature. Not every remittance advice uses the same terminology. One might say "invoice number," another says "reference number," another says "our ref." An AI column lets you write an instruction in plain language, something like "extract the invoice or reference number being paid on this line," and Lido figures out which field maps to that across any format. You're not writing regex or building field mappings per vendor. You write the instruction once.
The output is a structured CSV with one row per invoice line, with columns you define. You format that CSV to match exactly what your ERP's import template expects: same column names, same order, same date format. In Global Shop Solutions, for example, the cash receipts import has specific column requirements. Once you've built that output template in Lido, every subsequent extraction produces a file you can import directly without touching it. The same approach works for NetSuite, SAP Business One, QuickBooks, and Sage Intacct. Any ERP that accepts a CSV cash receipts import can receive extracted data this way.
For a deeper look at how AI-based extraction differs from traditional OCR, see our post on what is OCR data extraction.
The vendor name mismatch issue has a specific solution in Lido: context documents. You upload your vendor master list, a simple CSV with vendor names and vendor numbers, and Lido references it during extraction. When it pulls "iTool" from a remittance, it checks that against your vendor master and returns the matched vendor number and canonical name instead.
This matters because your ERP's import process likely requires a vendor number, not a name. If that field is blank or wrong, the import fails and someone has to fix it manually. Context documents eliminate that step. The vendor number comes through pre-matched in the extracted output.
Vendor numbers are also frequently absent from remittance advices entirely. Customers don't know your internal vendor numbering. They know their own. Context documents solve this too. Lido does the lookup, so you don't have to.
Real remittance advices often have pages you don't want. Debit memos, credit adjustments, terms and conditions attachments. These show up in the same PDF as the actual payment data. If you're paying per page or per extraction, processing these wastes money and introduces noise into your output.
Lido's exclude pages feature lets you mark specific pages to skip. You can do this at the document level or build rules to exclude pages that match certain patterns. The result is that only the actual remittance data gets processed, the debit memo on page two doesn't generate phantom rows in your CSV, and you're not paying to extract pages that serve no purpose.
A significant portion of remittance advices arrive via email, either as PDF attachments or as data pasted directly into the email body. If your process requires someone to download attachments and upload them manually to an extraction tool, you've only automated half the problem.
Lido supports email automation through auto-forwarding rules. You set up a forwarding rule in your email client or inbox (this takes about two minutes in Gmail or Outlook) so that any incoming remittance email gets forwarded to a Lido-assigned address. From there, extraction runs automatically and the output lands wherever you've configured it: an email back to your team, a shared folder, or a direct feed into your workflow.
The end state is that remittance advices arrive in your inbox, get extracted without anyone touching them, and a clean import-ready CSV shows up for review. Someone still looks at it before importing, and that's the right approach, but the data entry work is gone.
We've written a detailed walkthrough of this setup in our post on how to set up automated email invoice processing.
Accounts receivable automation changes the cash application role significantly. Instead of spending 6 hours a day reading PDFs and typing, that person spends an hour reviewing extracted outputs before import. They're looking for anything that didn't match cleanly: an invoice number that looks off, a discount amount that seems wrong, a vendor the system flagged as unmatched. Exception handling, not data entry.
Industry figures on this vary, but teams that have implemented AI-assisted cash application consistently report cycle time reductions of 70-80%. For a company processing 200 remittances a week, that's the difference between a full-time AR role and a part-time review task that gets absorbed into someone's existing responsibilities.
The manufacturing company mentioned earlier went from printing and filing paper to a workflow where remittance PDFs are extracted automatically and the AR team imports a file. The physical filing cabinet is still there. It just stopped getting new additions.
If you want to understand how automated extraction applies to AP-side workflows as well, our post on automated invoice processing covers the incoming document side of the same problem.
Not all document extraction tools are the same. Template-based tools will struggle the moment a customer changes their remittance format. For cash application specifically, you want a few things:
No-template extraction. The tool should work on documents it's never seen before. If setup requires mapping each vendor's format individually, that work compounds every time you add a new customer.
Line-item accuracy. Remittance advices have multiple invoice lines per document. The tool needs to extract rows correctly, not just header-level data. A tool that pulls the check total but misses individual invoice line amounts is only marginally useful for posting.
ERP format flexibility. Your import file needs to match your ERP's template exactly. The tool should let you define output column names, formats, and ordering rather than just exporting to a generic spreadsheet.
Vendor master integration. Matching vendor names to internal vendor numbers is a core part of cash application. A tool that supports lookup against a reference file saves significant time.
Email processing. If remittances come in via email, auto-forwarding support means the intake step is fully automated, not just the extraction step.
The fastest way to test whether this works for your situation is to run a sample of your actual remittance advices through an extraction tool. Take 20 RAs from different customers, pick ones that represent the format variety you deal with, and see what comes out. If the line-item data is accurate and the output matches what your ERP import needs, you have your answer.
The setup time for a basic Lido workflow is measured in hours, not weeks. You define your fields, configure your output format, set up email forwarding if needed, and upload your vendor master as a context document. From there, it's running.
For a team that's currently doing this manually, the calculation is straightforward: how many hours per week goes into reading remittance advices and keying data, and what's that worth. For most AR teams, the answer justifies the tool cost within the first month.
Cash application automation uses AI and software to match incoming customer payments to open invoices automatically, replacing the manual process of reading remittance advices and typing payment data into an ERP system. It covers extracting payment details from remittance documents, matching them to accounts receivable records, and posting the results.
A basic extraction template for remittance advices can be set up in under an hour. You define your output fields, configure formatting to match your ERP import requirements, and optionally upload a vendor master for automatic ID lookup. Most teams are processing their first batch of real remittances within a few hours of starting setup.
Yes. AI-powered extraction tools like Lido read documents contextually rather than matching fixed coordinates. A single extraction template handles remittance advices from different customers regardless of layout, column names, or formatting differences. The AI maps fields like Reference Number or Our Ref to your Invoice Number column automatically.
The standard fields for cash application are invoice number, invoice date, payment amount, discount amount, check number, check amount, check date, and customer or vendor name. Your specific ERP import template may require additional fields like vendor ID or external reference numbers.
Use a context document containing your vendor master list with canonical names and vendor IDs. The extraction tool cross-references customer names from remittance advices against your master list using fuzzy matching, returning the correct vendor ID even when the name on the document differs from what is in your system.
Any ERP that accepts CSV or Excel imports for cash receipts can work with automated extraction output. This includes NetSuite, SAP Business One, QuickBooks Desktop, Sage Intacct, Global Shop Solutions, Oracle, and Microsoft Dynamics. The key is configuring your extraction output columns to match the specific import template your ERP uses.
Teams that implement AI-assisted cash application typically report cycle time reductions of 70-80%. For a company processing 200 remittances per week, this reduces a full-time data entry role to roughly one hour of daily exception review. Most teams see positive ROI within the first month based on time savings alone.
Yes. Configure your extraction template to include a discount amount column. The AI extracts discount values from remittance advices where customers have taken early payment discounts, and outputs both the original invoice amount and the discount amount as separate fields for your ERP import.
%20(1).svg)
