Document type

Extracting data from ITIN notices

The IRS assigns an Individual Taxpayer Identification Number to a filer who cannot get a social security number, and confirms it by letter. One agency issues every one, in English and Spanish, across several notice types. Lenders and tax preparers see thousands each season, arriving as phone photographs and portal uploads.

A lender needs the ITIN, the name it was issued to, the date of birth and the notice date. A loan origination system matches those against the application and files them as the borrower's taxpayer identification evidence.

Start for Free Talk to Us

Page 1 of a sample ITIN Number, with extracted values outlined

4,951Markdown characters

99parse blocks

14fields extracted

The problem

What makes ITIN notices hard

Identifiers print more than once

The case reference prints in the header block, again on the tear-off stub and again on the continuation page, identically each time.

Key facts sit in grids

The header block and the personal information table are label-value grids, and a column header can print below the values it labels.

An ITIN has an SSN's shape

The number is nine digits in the same NNN-NN-NNNN pattern as a social security number, so a system has to know which it holds.

How it works

Parse, extract, verify

Parse

Parse turns the document into reading-order Markdown and a hierarchy of blocks (text, tables, figures), and grounds each block to its exact location on the page.

Extract

You define the fields in a JSON schema, and Extract returns their values in a consistent, structured format, each linked back to where it appears in the Markdown.

Verify

Each value links back to where it appears, down to the character range in the Markdown, and the page and bounding box in the source document, so every result can be verified.

Worked example

One ITIN notice, end to end

About this document: The IRS's own published sample of Notice CP565, which confirms an assigned ITIN. Two US Letter pages: the letter and its tear-off contact stub, then a page of instructions. Page 1 is featured. Every value is specimen data the IRS printed for illustration, including the ITIN itself, which it prints as NNN-NN-NNNN.

Processing objective: The objective is to lift the notice's identity and the number it confirms off page 1. The identifiers print more than once on the notice, so each one is taken from a named occurrence and boxed on the line it was read from.

Demonstration fields: ITIN, Notice number, Notice date, Case reference number, Date of birth

The complete example is on GitHubThe sample document, the full extraction schema, the complete parse output and every extracted field.

Open the folder

The document

2 of 2 pages

Page 1shown below
Page 2

Where each value came from

Page 1 of the sample ITIN Number, with every featured field outlined

ITINitin

NNN-NN-NNNN

The region of the sample ITIN Number containing ITIN, with the extracted value outlined

extraction_metadata["itin"]

{
  "value": "NNN-NN-NNNN",
  "page": 1,
  "range": {
    "end": 683,
    "start": 672
  },
  "box": {
    "xmin": 0.09261,
    "ymin": 0.27431,
    "xmax": 0.28729,
    "ymax": 0.29568
  }
}

Notice numbernotice.notice_number

CP565

The region of the sample ITIN Number containing Notice number, with the extracted value outlined

extraction_metadata["notice.notice_number"]

{
  "value": "CP565",
  "page": 1,
  "range": {
    "end": 160,
    "start": 155
  },
  "box": {
    "xmin": 0.74714,
    "ymin": 0.04748,
    "xmax": 0.94217,
    "ymax": 0.06301
  }
}

Notice datenotice.notice_date

January 26, 2017

The region of the sample ITIN Number containing Notice date, with the extracted value outlined

extraction_metadata["notice.notice_date"]

{
  "value": "January 26, 2017",
  "page": 1,
  "range": {
    "end": 215,
    "start": 199
  },
  "box": {
    "xmin": 0.74716,
    "ymin": 0.06301,
    "xmax": 0.94216,
    "ymax": 0.07862
  }
}

Case reference numbernotice.case_reference_number

99999

The region of the sample ITIN Number containing Case reference number, with the extracted value outlined

extraction_metadata["notice.case_reference_number"]

{
  "value": "99999",
  "page": 1,
  "range": {
    "end": 376,
    "start": 371
  },
  "box": {
    "xmin": 0.74724,
    "ymin": 0.12049,
    "xmax": 0.94214,
    "ymax": 0.13606
  }
}

Date of birthrecipient.date_of_birth

January 20, 1990

The region of the sample ITIN Number containing Date of birth, with the extracted value outlined

extraction_metadata["recipient.date_of_birth"]

{
  "value": "January 20, 1990",
  "page": 1,
  "range": {
    "end": 433,
    "start": 417
  },
  "box": {
    "xmin": 0.74726,
    "ymin": 0.13604,
    "xmax": 0.94214,
    "ymax": 0.1515
  }
}

How to try it yourself

Everything on this page is reproducible. The sample, its schema, and the Parse and Extract API outputs are published. You can repeat the same two calls from whichever surface you already build on.

Claude Skilldescribe the job; the agent writes it
CLIno code at all
MCPexpose ADE docs as an agent tool
REST APIany language
Pythonlandingai-ade
TypeScriptlandingai-ade

Run this sample yourself This document on GitHub

Why LandingAI?

Documents into data

Accurate

The best accuracy on the difficult documents other tools get wrong, such as messy scans, complicated tables, checkboxes and handwriting.

Auditable

Every answer cites its sources. Values come back with an audit trail (page, bounding box, character range) so a human can review them.

Agentic

Applies multi-step reasoning, and plans before it acts. Handles documents it has never seen before and adapts to variations.

API-first

Composable APIs that drop into any tech stack and can be called from anywhere. Easy to build on and embed in your pipelines and products.

Affordable

Cost-competitive rates with complete price transparency. Spend controls before you run, and every job reports its exact cost.

Enterprise-grade

Built for production and scale from day one. Fully observable and monitored, running securely in our cloud or yours.

Built for high-volume, high-throughput production

Zero data retention
Priority rate limits
Enhanced support
VPC and on-prem deployment

Available on Team and Enterprise plans.

Compare plans

Capabilities

ITIN Notice processing highlights

English and Spanish notices

Label-value header blocks

Tear-off contact stubs

Multi-page notices

Phone photographs and scans

ITIN and case references

Notice numbers and dates

Names and dates of birth

Agency contact details

Repeated identifier occurrences

Bounding box for every value

Zero data retention available

FAQs

Questions, answered

Every occurrence comes back with its own range and bounding box, so a value printed in three places gives three locations. A field's description can say which printing you want, and each result still points at the text it was read from.

No. A schema names the fields you want. Where they sit on the page is not part of it, so one schema reads a notice it has never seen. A field that notice does not carry comes back null.

Yes. PDFs, scans and photographs all parse the same way, and grounding coordinates come back against the page image either way. A letter photographed on a desk gives the same evidence as the original PDF.

Every value comes back with its page, a bounding box and the character range in the parsed Markdown. The box can be drawn onto the page image, so a reviewer sees the line the number was read from.

An ITIN notice carries a taxpayer identification number, a name, a date of birth and a mailing address. A schema can ask for the notice identifiers alone. Accounts are available with zero data retention, alongside SOC 2 Type II, GDPR and HIPAA compliance.

Start extracting ITIN notice data

Reliable, structured outputs with full traceability in minutes