Every deduction you code has a story behind it, and usually a document that tells it — a chargeback notice, a deduction invoice, a spreadsheet of retail-location detail, sometimes just a scanned page someone forwarded from a distributor portal. Figuring out which remittance line that document backs up, and what it actually says caused the deduction, is the part of coding that doesn't get faster no matter how many times you do it. It's manual, it's the same few minutes every time, and it's the reason a stack of PDFs can sit untouched for a week even when the rest of your remittance is already coded.
AI Document Coding is built for exactly that part. Upload the backup you already have, and RemitParse reads it, matches it to the right line, and proposes the coding — Type, Context, and Period — based on what the document actually says.
TL;DR
- Upload a chargeback notice, deduction invoice, spreadsheet, or scanned image, and AI reads it, matches it to the correct remittance line by reference number and amount, and proposes Type, Context, and Period.
- Anything the AI isn't confident about — a low-confidence field, a document that couldn't be matched, a line with no document, or a deduction over a threshold you set — gets flagged for review, not guessed.
- Works alongside Auto-Coding: a document match takes priority over a history-based guess when both apply, and every confirmed coding feeds the same shared history either way.
- Available on the Pro plan. Handles PDF, scanned images, Excel, Word, and CSV.
How it works
The workflow sits right after you upload your remittance — it's an optional second step, not a separate tool:
- Upload your backup documentation. Drop in whatever you already have for a deduction, in whatever format it came in. There's no reformatting or re-typing required first.
- AI reads it and matches it to your remittance. The document gets matched to the correct line by reference number and amount, and Type, Context, and Period get proposed based on what the document actually says — not a guess based on the invoice number's format.
- Review what's flagged, confirm the rest. A short summary appears once processing finishes, and a guided walkthrough steps you through anything that needs a second look before you export.
What "not a guess" actually means
A lot of deduction coding today works by pattern — an invoice number format, a prefix code, a description that vaguely resembles something coded before. That works most of the time and produces confidently wrong answers the rest of the time, because a pattern isn't evidence, it's an inference. AI Document Coding reads the document itself: a chargeback notice that says "Shelf Activity Fee," a spreadsheet showing a per-unit rate over a promotional date range, a scanned invoice with a program type printed at the top. The coding it proposes is grounded in what's actually written down for that specific transaction, not in what similar-looking transactions were coded as before.
That distinction matters most exactly when it's easy to get wrong — a new distributor program, an unfamiliar deduction code, a document with an invoice number that doesn't follow the usual format. Those are the cases where pattern-matching has nothing reliable to match against. They're also the cases a real document resolves cleanly, because the document was never relying on the pattern in the first place.
Nothing gets applied silently
Reading a document correctly and being confident about it are two different things, and RemitParse treats them that way. Every field comes with its own confidence tier, and low-confidence fields get flagged instead of applied. On top of that:
- A document that couldn't be confidently matched to any line is flagged as unmatched — nothing gets guessed onto a row it might not belong to.
- A remittance line with no matching document, once you've uploaded a batch, is flagged so it doesn't get missed.
- A deduction over a dollar amount you configure gets flagged for a manual look regardless of confidence — a large deduction deserves a human glance even when the AI is sure about it.
Everything flagged goes into a guided walkthrough — one item at a time, with the reason it's flagged shown up front, and the source document right there to check it against. Confirming an item, even without changing anything, is how you tell RemitParse "I looked, this is fine." Nothing exports until you've been through the queue.
You can also see the AI's reasoning. Every matched document shows why it landed on the Type it chose, why it identified that Context, and where in the document it found the Period — not just a confidence score, but the actual reasoning, so you're never taking the coding on faith.
How this fits with Auto-Coding
If you're also using Auto-Coding, the two don't compete for the same row. Auto-Coding fills first, from patterns in your own coding history, the moment a remittance uploads. If you later match a document to that same line, the document's answer takes priority — it's actual evidence for this specific transaction, not a pattern guess, so it should win.
The two features do reinforce each other, though. Every coding you confirm — typed by hand, pre-filled by Auto-Coding, or matched from a document — is saved to the same shared history. The more of your coding that comes from an actual source document instead of an inference, the more that shared history reflects verified answers. Auto-Coding's next guess is only as good as the history it's drawing from, so using AI Document Coding today makes Auto-Coding quietly more accurate on every upload after it.
A note on how this compares to autonomous deduction recovery tools
If you've read our CPG deduction software comparison, you'll know Glimpse takes AI further in one specific direction: its agents log into retailer and distributor portals themselves, retrieve backup documentation without anyone uploading it, and file disputes on invalid deductions autonomously. That's a genuinely different, more ambitious scope than what AI Document Coding does.
AI Document Coding works on documents you already have — it reads and codes what you upload, it doesn't go get documents from portals, and it doesn't dispute anything on your behalf. The two aren't solving the same problem end to end. If you're already using an autonomous recovery service for invalid deductions, AI Document Coding is what handles the valid, accepted ones — reading the backup you already receive and getting them coded and posted to QuickBooks correctly, without turning that into a second manual workflow.
Supported file formats for deduction backup documents
| Format | Notes |
|---|---|
| Both text-based and scanned/image-only PDFs | |
| Images (PNG, JPG) | Photos or scans of a printed document |
| Excel (.xlsx) | Including multi-sheet workbooks and aggregate spreadsheets covering many line items |
| Word (.docx) | Agreements or letters, including any tables within them |
| CSV | Plain tabular exports |
Getting started
AI Document Coding is available on the Pro plan. It's opt-in — turn it on per profile from Settings, and an optional "Upload Deduction Support" step will appear after each remit upload from then on. If you're on Starter or Growth and want to try it, you can upgrade from your account settings; existing coding history carries over, so there's nothing to rebuild.
Related guides
- CPG Deduction Software Compared — where AI Document Coding fits alongside Glimpse, Confido, and TPM platforms
- Confido vs. RemitParse — a detailed comparison of two cash application approaches for CPG brands
- Is Vividly Worth It for Your CPG Brand? — why clean, consistent deduction coding matters before a TPM platform is worth it
- How RemitParse Works — the full remittance-to-QuickBooks workflow
Try it on your own backup documents
Upload a remittance, drop in a chargeback notice or deduction invoice, and see what AI Document Coding proposes — 30-day free trial on Pro, no credit card required.
Start Free Trial →