Step 1
1. You send samples
An empty form and several filled-in copies without personal data or with fictitious data. Scans and photos in the quality in which documents reach your work are best.
Data extraction from contracts, forms and blanks
NeuroVision turns forms, applications, blanks and standard contracts into data: names, dates, numbers, amounts and details reach your system as ready-made JSON fields. The template is set up for your form, and fields the model is not sure about go to a person for review.
Recognition result Insurance claim form
- Applicant high TIKHONOVA ZOYA LVOVNA
- Policy number high 514 2086379
- Event date high 01.10.2026
- Phone high +7 915 207-43-18
- Event description review Flat flooded from the apartment above
- Template for your form
- Confidence for every field
- A free month on your own samples
Data is stored in Russia Russian Software Registry No. 19972 Compliance with 152-FZ
Data extraction from documents: how IDP differs from OCR
IDP documents: from paper to fields
What changes for the operator when a form goes through data extraction. Recognition of identity documents — passports, driving licences, ID cards — works without any setup: the model knows their layouts. More on IDP for identity documents is on the page document recognition. This page is about documents that every company has its own.
- 01
OCR is text
Optical character recognition turns an image into lines of text. OCR does not know which line is the surname and which is the policy number: a person or your code has to parse the text and copy it into the customer record.
- 02
IDP is fields
Intelligent document processing answers the question “what is written here and which field does it go into”. The system finds the needed zones on the blank, reads each one and returns “field name — value” pairs. That is why IDP is also called document parsing and automatic data capture from a document.
- 03
The form template is the basis of accuracy
A bank questionnaire, a payout claim and an employee personal card are built differently. A template is set up once for each form: where the name is, where the date is, where the amount is and what format the value has. After that the service finds the fields on its own on any filled-in copy.
- 04
Confidence and a human in the loop
For every field, a system that extracts data from documents returns a confidence score. Confident values go straight to the accounting system, doubtful ones go to an operator. Manual entry is replaced by checking a few fields.
- JSON fields instead of a solid block of text
- 1 template per form, then no manual markup
- conf a confidence score for every field
Recognition of forms, applications and contracts: what we process
Documents with a fixed structure work best: the fields stay in place and only the values change. A separate template is set up for each form. We accept JPEG and PNG images; send a multi-page document or a PDF contract page by page in one session.
-
Customer forms
Forms of a bank, insurer, fund or clinic. We extract full name, date and place of birth, passport details, address, phone number, e-mail and yes / no checkbox marks.
-
Applications
Claims for an insured event and for payout, applications for admission to a school, for transfer of savings, for a service. OCR for applications immediately sorts the applicant's data and the substance of the request into fields.
-
Blanks and HR forms
Strictly structured blanks and unified forms, for example the Form T‑2 employee personal card. Tables of a fixed layout are read row by row.
-
Standard contracts
Contracts built on your template: number and date, parties, passport details and requisites, amounts and terms in fixed fields. We do not extract the meaning of clauses written in free wording — details in the section below. What we extract from a contract
-
Attached identity documents
A passport or driving licence that the customer attached to an application is recognised by the ready-made model with no setup: 16 000+ document types from 195+ countries and territories. The data can be cross-checked against the form. Passport recognition
-
Handwritten fields
Forms and applications filled in by hand in block letters — in boxes or on the lines of a blank. We check cursive handwriting on your samples. Handwritten fields
Form, blank or contract: what is extracted automatically
-
A form or blank with fields
Example. A form with boxes, an application blank, a Form T‑2 personal card.
How fields are found. By the coordinates and labels of the template fields.
What setup needs. An empty form and several filled-in samples.
Where a person is needed. In fields with low confidence.
NeuroVision. Yes, by template. -
A standard document on your template
Example. A contract on the company template, a free-form application on a sample.
How fields are found. By template: details, parties and amounts in fixed places.
What setup needs. Samples in several versions.
Where a person is needed. In fields with low confidence and when the sample version changes.
NeuroVision. Yes, after a test on your samples. -
A contract in free wording
Example. A counterparty's contract, individual terms, correspondence.
How fields are found. Field positions differ every time.
What setup needs. A template cannot be built.
Where a person is needed. All the work is with the text.
NeuroVision. No: we do not extract the meaning of terms.
The more constant the structure of a document, the less setup and manual checking it needs. This rule holds for any IDP system, not only for NeuroVision.
Contract recognition: which fields we extract and which we do not
- Contract number and date, place of signing
- Parties: full name, organisation name, signatory's position
- Passport details of an individual party in the preamble
- Tax number (INN), OGRN, KPP and bank details in the “Details of the parties” section — with a format check and a check of the INN checksum
- Price, term, property address — if they sit in fixed fields of the sample
- Signature and stamp on the page — their presence is recorded, the signature is not “read”
- A copy of a party's passport or driving licence
- A customer form attached to the contract
- A specification, act, payment schedule — tables of a fixed layout
- Terms written in the text of the clauses: penalties, payment procedure, grounds for termination
- Searching for risky terms, legal assessment of a contract
- Comparing versions and protocols of disagreements
- A counterparty's contract in any form, without your sample
- by template — the field is extracted once the template for your sample has been set up and tested.
- ready-made model — the document is recognised without setup, as in KYC.
- to be confirmed — we check on your samples before launch.
- not extracted — this is a text analysis task, not recognition: it needs a lawyer or a separate contract analysis system.
- Contract recognition in NeuroVision turns the details and party data into fields of your accounting system so that the operator does not retype them from a scan. You can check the counterparty by the INN from the contract right after recognition — counterparty check by tax number.
How the template is set up for your forms and blanks
The NeuroVision team sets up the template for a form. From you — samples and rules on which fields of your system the data goes into.
Step 2
2. We set up the template
We mark the fields: where the name is, where the date is, where the amount is, which checkbox marks there are. For each field we set a format: a date as a date, the INN with a checksum, a phone number by a mask.
Step 3
3. We test on your samples
We run a test batch and show which fields were recognised with confidence and which would have gone for review. From this data you decide whether the form suits automatic processing.
Step 4
4. We connect to your system
Documents are sent over the REST API, and the result comes as JSON in the response or as a webhook. The fields are written to a CRM, core banking, HR or accounting system through your backend.
Step 5
5. You work and check the disputed
Confident values go on automatically, doubtful ones go to your operator. A new version of the blank is a reason to update the template.
Confidence for every field and human review
-
Field confidence level
In the response every field has a name, a value and
conf— the model's confidence in the reading. Separate flags report that a field cannot be read or does not match another source. -
Format check
The value is checked against the field rule: in a date, a real date; in the INN, a valid checksum; in a phone number, the right number of digits. A format error lowers trust in the field even if the characters are read clearly.
-
The operator sees only the disputed
You set the rule: which fields to accept automatically at what confidence and which to show to an employee. They compare the value with the scan and correct it in your system — there is no need to retype the whole form.
Accuracy depends on the form, the handwriting and the quality of the scan, so we do not promise a percentage in advance. Measure it on your own documents during the test month.
Forms and applications by industry: where data extraction replaces manual entry
Wherever a document arrives as a scan or a photo and an employee copies it into the system field by field.
-
Insurance
- Claims for an insured event and for payout
- Applicant, policy and event data — into the claims settlement system
- An attached passport and driving licence — without setup
-
Banks
- Customer forms and loan applications filled in at a branch or by an agent
- Data in the core banking system without retyping
- Checking a form against the passport Learn more
-
Private pension fund
- Applications for transfer of savings and for payout
- Documents of legal successors
-
Notaries
- Applications and forms of people who contacted the firm
- Party data from passports and standard contracts — into the case file Learn more
-
HR departments
- Employee personal cards in Form T‑2 and candidate forms
- Digitising paper personal files
- Passport and documents of foreign employees
-
Education
- Applicants' applications and consents to data processing
- Applicants' data — into the admissions office system
- The applicant's passport — without setup
Connection: REST API, Web SDK or installation inside your perimeter
Scans from an archive and streaming applications — through the API, photographing a blank by an employee or a customer — through the Web SDK, documents with sensitive data — inside your perimeter.
- Fields in JSON
- Confidence for every field
- Templates for your forms
- Webhooks
Terms of connection
A free one-month test on your own samples, a personal manager, payment only for processed documents.
Personal data in forms and contracts: what Federal Law No. 152‑FZ requires
Data in Russia
The NeuroVision cloud runs on servers in Russia
When collecting the personal data of Russian citizens, recording, systematisation, accumulation and storage must take place in databases located in Russia (Art. 18, Part 5 of Federal Law No. 152‑FZ). Cloud processing does not breach this requirement.
Processing on instruction
You are the operator, NeuroVision processes the data on your instruction
When personal data processing is entrusted to another party, the instruction states the purposes, the list of actions and the protection requirements (Art. 6, Part 3 of Federal Law No. 152‑FZ).
Retention periods
You decide how long to keep it
The retention period for results is set in the scenario; document images are deleted after 6 months. In your perimeter you define the periods.
A webhook without personal data
Statuses only, if that is what you need
The webhook can be encrypted with a secret key or sent without personal data: your system will fetch the fields with a separate request.
Isolated environment
Nothing sent outside
With an on-premises installation, forms, contracts and results are processed and stored only on your servers.
152‑FZ
152‑FZ — a Russian developer, software in the Russian Software Registry (entry No. 19972)
How much data extraction from forms and contracts costs
The price depends on the number of forms, the volume of documents and the installation method. The first month is a free test on your own samples.
Calculate the cost- 01
Pay for processed documents, no subscription fee
- 02
Template setup for your forms
- 03
Cloud in Russia or installation in your perimeter
Glossary: IDP, form template, confidence level
Terms you will meet in commercial proposals and terms of reference for document recognition.
- IDP
- Intelligent document processing: text recognition, finding the needed fields, checking values and passing them to the accounting system. Unlike OCR, the result is fields, not a solid block of text.
- Intelligent document processing
- OCR
- Optical character recognition: converting an image of printed text into machine-readable text. The basis of IDP, but not a replacement for it.
- Optical Character Recognition
- ICR
- Recognition of handwritten characters written separately — for example, in block letters in the boxes of a form.
- Intelligent Character Recognition
- Form template
- A description of a document for the system: where the fields are, what they are called and what format the values have. Set up once per form.
- Structured document
- A document whose fields always sit in the same places: a form, a blank, a unified form. Best suited to automatic extraction.
- Semi-structured document
- A document with a constant set of data but a floating layout: a standard contract, a free-form application on a sample.
- Confidence level
- The model's estimate of how reliably a field was read. It is used to decide whether to accept the value automatically or send it to an operator. In the API response it is the field
conf. - conf
- Human review
- A scheme in which an employee checks only the low-confidence fields, not the whole document.
- human-in-the-loop
- Document classification
- Detecting the document type from the image: passport, driving licence, bank form, application. It works automatically for identity documents.
- Field validation
- Checking a value against a rule: date format, phone mask, INN checksum.
Licences and certificates
The project «Development of a multimodal system using artificial intelligence "Neuromed" for managing the diagnostic processes of a clinic on the basis of the "Neurovision" computer vision platform» is supported by the Innovation Promotion Foundation within the federal project «Domestic Solutions»
The Neurovision computer vision platform is registered with Rospatent (certificate No. 2023661508) and included in the Russian Software Registry (entry No. 19972 of 13.11.2023)
The right to use the Platform is granted under a licence agreement concluded by Neurovision with its customers
Send us samples — we will set up recognition for your form
Tell us which forms, applications or contracts you need to process and in what volume. A manager will ask for samples without personal data or with fictitious data after your request — do not attach them to the form.
Can't put your request into words? Book a call and our team will contact you at a convenient time
Book a call
The company is included in the register of IT companies and the register of personal data operators of the Russian Ministry of Digital Development.
Address: Moscow, Presnenskaya Embankment 12 — Moscow City business centre, Federation Tower. Nearest metro stations: Delovoy Tsentr and Mezhdunarodnaya.
Read also
How to improve recognition quality and protect customer data — in our blog.
353 23
KYC Document Verification
AI-OCR in KYC: how automatic document recognition reduces manual review
399 10
AI OCR, IDP and Document Recognition
How to improve the accuracy of AI OCR and MRZ verification in KYC: metrics, errors and working solutions
392 21
KYC Document Verification
Why users retake documents: how to reduce the number of repeat attempts during a KYC check
356 16
KYC, AML and Digital Identification
How to ensure the protection of personal and biometric data in KYC processes
Frequently asked questions about data extraction from forms and contracts
Answers about documents, templates, accuracy, integration and cost
It is turning a scan or photo of a document into separate fields: names, dates, numbers, amounts, details. It is also called document parsing or IDP. Unlike plain OCR, the result is not a solid block of text but data that can be written straight into the accounting system.
OCR recognises characters and returns text. IDP finds the needed fields in a document, reads each one, checks the format and returns “field — value” pairs with a confidence score. OCR is a part of IDP.
Forms, applications, blanks, unified forms and standard contracts on your template — after the template is set up. Passports, driving licences, ID cards and other identity documents are recognised without setup: 16 000+ document types from 195+ countries and territories.
Yes, if the contract is drawn up on your template: we extract the number and date, the parties, passport details, requisites, amounts and terms in fixed fields. We do not analyse the meaning of terms in the text of clauses, risky wording or counterparties' free-form contracts.
Yes, one template per form. The NeuroVision team sets it up from your samples: an empty form and several filled-in copies without personal data. The template is then tested on a test batch.
A confidence score is returned for every field. You set the rule: confident values are written automatically, doubtful ones are shown to an operator, who compares them with the scan.
It depends on the form, the handwriting and the quality of the scan, so we do not name a percentage in advance. During the test month accuracy is measured on your documents, and the confidence score shows which fields need checking.
The service accepts JPEG and PNG images; PDF files are rejected. Send a multi-page contract or form page by page — as several images in one session.
Yes, if the fields are filled in block letters — in boxes or on the lines of a blank, as well as checkbox marks. We check cursive handwriting on your samples. More on the page on handwritten text recognition.
Yes, through your backend: the service returns the fields as JSON in the response to a request or as a webhook, and your system writes them where needed. We do not claim ready-made connectors to 1C and document management systems.
In the cloud — on NeuroVision servers in Russia; with installation in your perimeter — only on your servers. Images in the cloud are deleted after 6 months, and the retention period for results is set in the scenario.
Payment is for processed documents; the price depends on the number of forms and the volume. The first month is a free test on your own samples. We will calculate the exact cost after reviewing the forms.
