Products › Intelligent Document Management
Turn documents into
structured data.
Contact and receipt intelligence
AI extraction reads business cards and receipts in any language, pulls out every field, converts currencies and flags tampering — with local OCR as an automatic offline fallback, and a person confirming before anything is saved.
Nothing reaches your library until you have looked at it.
Photograph a card, check the flagged fields, confirm — and it is a record.
It tells you what it
is not sure about.
Extraction is the easy half. The hard half is knowing which of the twelve fields it just filled in you should actually check — so the Intelligent Document Management marks them, next to the original photograph, before you commit anything.
Per-field confidence, not one score for the document. Fields below the threshold are highlighted individually, so you verify the two that are doubtful rather than re-reading all twelve.
The photo sits beside the form. You check the flagged value against the paper without hunting for the original.
Warnings arrive before the confirm, not after. A receipt carries a tamper-risk flag, a blurry-image flag and a duplicate check — because a receipt becomes money against a cost centre, and "this image may have been altered" is not something to learn later.
It refuses to invent a person. A card whose name could not actually be read is rejected rather than saved with a placeholder in the name field — added after one came back reading IMAGE NOT INCLUDED at high confidence and the review pane presented it as somebody's name.
What you correct is what gets saved. Your edits win, and the original extraction is kept underneath rather than overwritten.
2 fields below the confidence threshold · shown beside the original scan · nothing written until you confirm
Photograph it. Then check five fields.
Five stages, and only one of them needs you. The record does not exist until the fourth.
What the Intelligent Document Management does
Two document types, done properly
Business cards and receipts. Nothing else — not invoices, purchase orders, contracts or identity documents. From a card you get the full name, company, job title, department, the primary email and every other email found, the primary phone and every other phone, the address broken into city, state, postal code and country, a LinkedIn URL and notes. From a receipt: merchant name and address, receipt number, transaction date, currency, total, tax, discount, payment method, expense category and the individual line items.
Any language, any currency
The reader is not tuned to one script or one locale. It reads the document in whatever language it was printed in and translates as it extracts, so a receipt in Malay, Mandarin or Tamil comes back as the same structured fields as an English one. Amounts are converted and the base currency is recorded alongside the original, so a figure is never ambiguous about which currency it is in.
Recovering from a bad scan without retaking it
Read again re-extracts the image you already uploaded, so a poor first pass does not mean finding the card again. Merge folds a duplicate into the original — gaps filled, emails and phone numbers combined rather than one overwriting the other. Link later attaches a capture to a record after the fact. And a manual receipt lets you type one that never had a photograph, so a cash fare sits beside the scanned costs instead of in a separate note.
It works when the connection does not
Extraction runs on AI vision where it can and falls back to local optical character recognition automatically when it cannot — so capture in a basement car park or on a plane still produces something to review later, rather than an error and a lost receipt.
The original extraction is never thrown away
Every capture keeps what the model originally read alongside what you confirmed, with the edit history attached. That matters twice: you can see what was changed and by whom, and it is the honest basis for measuring how often review actually corrects something — which is a real number about this product rather than a claimed accuracy figure.
Taking the data out
Cards export to CSV, Excel, JSON and vCard, so a stack of business cards becomes contacts your phone can import directly. Receipts export to CSV, Excel, JSON and PDF. There is no lock-in step and nothing needs an integration to be useful — the export is the integration.
Extraction is automatic.
The write is not.
A document intelligence tool that saves silently is a data-entry tool with extra steps. Three things make the difference.
A person confirms every record
A capture moves through processing, then waits in review. It becomes a contact or a cost only when somebody has read it and said yes — and it can be discarded instead. There is no automatic write path.
Your library is only yours
Every query is scoped to the signed-in account. Nothing in the product exposes one person's captures to another, and a scanned business card is by definition somebody else's personal data being held on your behalf.
Warnings before the decision
Tamper risk, blur, duplicates, missing fields and per-field confidence are all surfaced at the moment you are deciding — not recorded quietly for an audit nobody runs.
And what it does not do. One account, one library. There are no teams, no sharing and no delegation anywhere in the product — you cannot give your accountant access, and a bookkeeper cannot hold a client's library on their behalf. Two document types only, images only. Capture by emailing or messaging a document in is built but not switched on, so today it is upload and camera. Reconciling receipts against a bank statement is in the product, but the suggestion queue is not yet available. And we publish no accuracy figure, because nothing in the product measures accuracy against a ground truth — what we can show you instead is how often a human corrected a field, which is a real number from your own library rather than a claim about ours.
Frequently asked questions
Can this document scanning tool also read invoices, purchase orders or contracts?
No. It handles business cards and receipts only, and nothing else — not invoices, purchase orders, contracts or identity documents. If your paperwork is mainly those other types, this is the wrong tool.
Does the extracted data get saved automatically, or does someone have to approve each record?
Someone has to approve it. Nothing is written to your library until you confirm the capture. The page describes five stages, only one of which needs a person, and the record does not exist until the fourth stage.
Will it read receipts printed in Malay, Mandarin or Tamil, and what about foreign currencies?
Yes. It reads the document in whatever language it was printed in and translates as it extracts, so a Malay, Mandarin or Tamil receipt comes back as the same structured fields as an English one. Amounts are converted and the base currency is recorded alongside the original figure.
How do I get contacts and receipts out of it — does it need an integration with my accounting or contacts software?
No integration is needed; the export is the integration. Business cards export to CSV, Excel, JSON and vCard so your phone can import them as contacts directly, and receipts export to CSV, Excel, JSON and PDF.
Can I give my accountant or bookkeeper access to the same receipt library?
No. It is one account, one library — there are no teams, sharing or delegation anywhere in the product, so you cannot give your accountant access and a bookkeeper cannot hold a client's library on their behalf.
How accurate is the receipt and business card OCR?
TalbotIQ publishes no accuracy figure, because nothing in the product measures accuracy against a ground truth. What it shows instead is how often a human corrected a field in your own library, which is a real number from your data rather than a claim about the product.
Bring a card and a receipt.
Book a 30-minute demo and we will scan both live — including a deliberately bad photo, so you can see what it does when it is unsure.
Malaysia-based team · Response within one business day · No obligation