A product of Infiqon
Your invoices, keyed by a machine.
Checked by your own rules.
Docqon reads vendor invoices from your inbox, Salesforce or a scanner, extracts every field and line item, and proves the arithmetic before anything reaches your ledger. What passes posts itself. What doesn't lands in a review queue built for speed.
No card. No setup call. Upload an invoice and watch it come back structured.
Most invoice tools stop at "the AI read it"
That's the easy half. The hard half is knowing when the AI is wrong — before a wrong number becomes a wrong payment.
Docqon re-does the maths on every invoice it reads. Line items must sum to the subtotal. Subtotal plus CGST, SGST and IGST must equal the grand total. A GSTIN must pass its checksum, not merely look like one. A date must be plausible. A vendor must exist in your ERP's vendor master.
A field that fails one of those checks is marked low-confidence no matter how certain the model claims to be — and the document is held for a human. That single rule is why a wrong total gets caught instead of paid.
How it works
Four stages. You only touch the third, and only when something is off.
-
1
Collect
Poll an IMAP or Microsoft 365 mailbox, pull attachments from Salesforce, or drag files in. Duplicates are caught on message ID and file hash, so the same invoice never enters twice.
-
2
Extract & verify
A vision model reads every page against your field schema — no fixed templates. Then deterministic checks re-compute the totals and validate formats. Multi-invoice files are split into one document per invoice.
-
3
Review only the exceptions
Clean invoices skip straight through. The rest open in a side-by-side reviewer: fields needing attention sorted to the top, click a field to highlight it on the page, Enter to confirm and jump to the next. Corrections are stored as ground truth.
-
4
Post to your ERP
Approved invoices become vendor invoices in Dynamics 365 over OData, with the vendor fuzzy-matched against your master. Retries never duplicate.
We publish our accuracy, and how we measure it
Vendor accuracy claims are usually unfalsifiable. Ours comes from a fixed, public corpus of 72 real invoices and receipts — vendor PDFs, phone photos, scans, e-invoices, and deliberate traps — with hand-verified ground truth.
What it gets right
95.2% of hand-checked fields, including every invoice number and date in the golden set. Vendor names, PO numbers and line-item counts matched in full.
Where it slips
Faded thermal receipts, handwriting, and unfamiliar locales are measurably weaker. We publish those numbers to ourselves rather than hide them — the review queue exists precisely for those documents.
Why it still posts safely
Every mistake in our benchmark was caught by an arithmetic check and routed to a reviewer. The model being wrong is expected; the ledger being wrong is not.
Every change to our extraction pipeline is re-run against that corpus before it ships, and any regression against ground truth blocks the release.
Fits the stack you already run
Sources
IMAP, Microsoft 365 (Graph), Salesforce attachments, and direct upload. PDF, PNG, JPEG and TIFF — including faxed .tif and camera photos.
Destination
Microsoft Dynamics 365 Finance & Operations via OData, with idempotent posting and fuzzy vendor matching against your vendor master.
Models
Qwen-VL on private GPUs, Google Gemini, or OpenAI — chosen per organization, with automatic fallback if a provider fails.
Need a different ERP or source? The connectors are pluggable interfaces — tell us what you run.
Pricing
Start free today. Move up when the volume is real.
Free
$0
2 invoices per day
- Full extraction and validation
- Complete review queue
- Document splitting
- No card required
Growth
Talk to us
Volume-based daily limits
- Everything in Free
- Mailbox and Salesforce ingestion
- Dynamics 365 posting
- Accuracy reporting per field
Enterprise
Talk to us
Unlimited, with your own models
- Private GPU deployment (Qwen-VL)
- Custom field schemas and checks
- SSO and audit requirements
- Implementation by Infiqon
Questions worth asking
How accurate is Docqon at reading invoices?
On a 72-document public benchmark of real invoices and receipts, Docqon matched hand-verified ground truth on 95.2% of fields. More importantly, the errors it did make were caught by deterministic arithmetic checks and sent to a human instead of reaching the ledger.
Do I have to check every field by hand?
No. Invoices that pass every arithmetic, tax and format check above your confidence threshold post automatically. Only documents that fail a check, fall below the threshold, or have an unmatched vendor reach a reviewer — and those open with the problem already highlighted.
Which file types and languages are supported?
PDF, PNG, JPEG and TIFF, including phone photos and scans. Tested on English, French, German, Dutch and Indonesian invoices in USD, EUR, GBP, INR, IDR and CAD. Indian GSTINs are validated including the checksum digit, and CGST/SGST/IGST are reconciled against the grand total.
What happens when one file contains several invoices?
Docqon detects the boundaries and splits the file into one document per invoice. Supporting pages — delivery notes, card slips, terms — are kept as attachments rather than billed as invoices. You can require a human to confirm each split, or split mechanically one document per page for scanner batches.
Where does the extracted data go?
Approved invoices are posted to Microsoft Dynamics 365 Finance & Operations over OData. Every push carries an idempotency key, so retrying a failed post never creates a duplicate vendor invoice.
Is the free plan really free?
Yes — two invoices per day, no card. The quota resets at midnight UTC. When you need more, we raise the limit on your organization.
Put one invoice through it
That's the honest demo. Upload something awkward — a photo, a foreign-currency bill, a scan with three invoices in it — and see what comes back.
Start free — 2 invoices a day