Intelligent Form Reader
An Intelligent Form Reader is a Salesforce Industries feature that uses optical character recognition (OCR) to pull structured data out of uploaded documents and write it onto Salesforce records.
Definition
An Intelligent Form Reader is a Salesforce Industries feature that uses optical character recognition (OCR) to pull structured data out of uploaded documents and write it onto Salesforce records. It reads PDFs and image files (JPG and PNG), recognizes the fields on a form, and maps each one to a field on a Salesforce object so a person does not have to retype it.
The feature is powered by Amazon Textract and ships inside the Industries Clouds, where Salesforce first released it for Financial Services Cloud in Spring '21. Salesforce has since folded the capability into the broader Intelligent Document Reader, which extends the same OCR engine to more clouds. If you set this up today, you enable it under the Intelligent Document Reader settings, but the Form Reader name still appears in Financial Services Cloud and in the underlying APIs.
How Intelligent Form Reader turns a scanned form into a Salesforce record
Amazon Textract does the reading
The OCR engine behind the feature is Amazon Textract, an AWS machine-learning service. Textract goes past plain character recognition. It detects key-value pairs (the label and its answer), tables, and signatures, and it can read both printed text and handwriting. Salesforce calls Textract through a managed connection, so an admin never touches an AWS console or manages AWS credentials directly. You enable the integration inside Setup, and the platform handles the round trip to Textract for each uploaded document. Two Textract options are worth knowing. Analyze ID reads identity documents such as a driver's license or passport and returns the standard fields on them. The Tables option pulls rows and columns out of less structured documents. You turn these on in the global content extraction settings when the form types you handle need them. Because Textract is doing the heavy lifting, the quality of the source scan matters. A clean, high-resolution upload extracts far more reliably than a faxed, skewed, or low-contrast page, so input quality is part of the design.
Document types, templates, and field mapping
The feature does not guess at a random page. You first create a document type for each kind of form your org receives, such as a loan application, an insurance claim, or a patient intake sheet. For each document type you build a template that lists the fields on the form and ties every field to a target object and field in Salesforce. That mapping is the core of the configuration. When a user uploads a form of that type, the extracted values land on the mapped fields, ready to create a new record, update an existing one, or verify data already in the org. The mapping is declarative, so the common path needs no Apex. A template can cover a form with a fixed layout, and Textract's machine learning also copes with forms whose layout shifts a little from one document to the next. When a form design changes in a meaningful way, the template usually needs a refresh, because fields that move can map to the wrong place. Treat template maintenance as part of your release cycle rather than a one-time task.
Confidence scores and the human review step
Textract returns every extracted value with a confidence score. Intelligent Document Reader expresses that score on a 0 to 100 scale and surfaces it in Salesforce. During enablement you set a Confidence Score Threshold, and any field that scores below the threshold gets highlighted for a reviewer. The point is to keep a person in the loop without making them retype the whole form. High-confidence fields can flow through, and the reviewer focuses attention on the values the AI was unsure about. This matters most in regulated work. A misread Social Security number or a wrong date of birth on a benefits application is not a small error, so the threshold gives you a deliberate control over how much you trust automated extraction. Set the threshold too high and reviewers check almost everything, which erodes the time savings. Set it too low and bad values slip onto records. The right number depends on the specific form, so tune per document type and watch the manual-correction rate as you go.
Setup objects and permissions
Enabling the feature touches a handful of Industries objects, and users need the right access to them. The setup grants create, read, edit, and delete permissions on objects including Document Checklist Item, Location, Received Document, and OCR Document Scan Result. The Received Document and OCR Document Scan Result records hold the uploaded file reference and the structured output of the scan. A Document Checklist Item often represents a required document in an intake process, which is why these features are frequently used together. Two more settings round out a basic enablement. You set a maximum number of pages to extract, which defaults to five consecutive pages, so very long documents do not run away with processing. You also enable alternative file previews in Salesforce Files settings so the image preview works. All of this lives under the Intelligent Document Reader area of Setup, and the feature requires the Intelligent Document Reader add-on license on top of the Industries Cloud you run.
Where it fits in the Industries Clouds
Form Reader started life in Financial Services Cloud, aimed at the paperwork that runs through banking and insurance. Loan applications, account opening packets, and claims are classic examples, because each one is a form a customer fills out and someone then keys into a system. As Intelligent Document Reader, the same OCR engine now reaches more clouds, including Health Cloud, Public Sector Solutions, and others across the Industries portfolio. Health Cloud uses it for patient intake, referrals, and prior authorization documents. Public Sector teams use it for permits, benefit applications, and identity verification. The common thread is volume. These are industries where forms arrive in large numbers and where manual transcription is slow, costly, and a source of data-entry mistakes. Salesforce ships starter templates for several common document types, which gives teams a head start instead of mapping every field from scratch. The feature pairs naturally with case management and onboarding flows already built in those clouds, so the extracted data feeds straight into the process that needs it.
Driving extraction from OmniStudio and flows
The reader is rarely a standalone screen. It is usually invoked from a process the user already follows. An OmniScript can collect the file upload and trigger extraction, and an Integration Procedure can orchestrate the call when you want it to run as a service rather than a guided screen. Salesforce also exposes flow actions, so a record-triggered or screen flow can kick off extraction without code. The pattern is consistent. A customer or an agent uploads the document, the platform sends it to Textract, the extracted values come back onto the mapped fields, and the user reviews anything flagged below the confidence threshold before the data is saved. Because the orchestration is declarative, most teams build the whole loop without writing Apex. For programmatic needs, the Intelligent Document Reader Business APIs let you read the mapping details and the extracted output from the OCR Document Scan Result object, which is useful when you integrate with an external system or build a custom review experience on top of the standard one.
Languages, handwriting, and accuracy planning
The OCR supports six languages for printed text: English, Spanish, German, Italian, Portuguese, and French. Handwriting recognition is available for English, and it typically reads with somewhat lower accuracy than clean printed text. Forms that mix printed and handwritten fields work, but the handwritten sections tend to need more reviewer attention, so plan for that when you set thresholds. Accuracy is not a single fixed number. It depends on scan quality, the form layout, the language, and whether the content is printed or handwritten. The practical approach is to measure rather than assume. Track how often reviewers correct the AI-extracted values for each document type. A high correction rate points to a template that needs work, a threshold set wrong, or source documents that are too poor to read reliably. Storage is the other planning item. Every uploaded document is retained and linked from the extracted record back to its source file, which supports audits but accumulates quickly in a high-volume intake process. Size your file storage for the document throughput you expect.
Enable and configure Intelligent Form Reader
Enabling Intelligent Form Reader (configured today under Intelligent Document Reader) takes a short setup pass plus a template for each form type. You need a System Administrator profile, Lightning Experience, an Industries Cloud such as Financial Services Cloud, and the add-on license. These are the main steps.
- Turn on the feature
In Setup, search for Intelligent Form Reader or Intelligent Document Reader and enable it. This exposes the global content extraction settings and the related objects.
- Set global extraction settings
Edit the global content extraction settings to set the maximum pages to extract (default five) and the Confidence Score Threshold. Turn on Amazon Textract Analyze ID for identity documents and the Tables option if your forms need them.
- Grant object permissions and previews
Give users create, read, edit, and delete access to Document Checklist Item, Location, Received Document, and OCR Document Scan Result. In Salesforce Files settings, enable alternative file previews so image previews render.
- Create document types and templates
For each form you handle, create a document type, then build a template that maps each field on the form to a target object and field in Salesforce. Test the mapping with a real scanned sample before going live.
A 0 to 100 value. Fields scoring below it are highlighted for review. Tune it per document type rather than using one number everywhere.
Caps how many consecutive pages are extracted from a document. Defaults to five, which keeps very long files from over-processing.
Enables reading of identity documents such as licenses and passports, returning their standard fields.
Extracts rows and columns from documents that hold tabular or less structured data.
- The feature needs the Intelligent Document Reader add-on license on top of your Industries Cloud. It is not part of the base edition.
- Poor scan quality (skewed, low-contrast, or faxed pages) drops extraction accuracy no matter how the template is configured.
- When a form's layout changes, refresh its template. Moved fields can map to the wrong target and silently corrupt records.
- Retained source documents accumulate fast in high-volume intake, so size your file storage for the expected throughput.
Prefer this walkthrough as its own page? How to Intelligent Form Reader in Salesforce, step by step
Trust & references
Cross-checked against the following references.
- Intelligent Document ReaderSalesforce
- Enable Intelligent Form Reader and Configure SettingsSalesforce
Straight from the source - Salesforce's reference material on Intelligent Form Reader.
- Intelligent Form ReaderSalesforce
- Intelligent Form Reader Business APIsSalesforce
Hands-on resources to go deeper on Intelligent Form Reader.
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
Q1. Which underlying engine does Intelligent Form Reader use to extract data from scanned forms?
Q2. How does Intelligent Form Reader signal which extracted fields a human reviewer should verify?
Q3. Which OmniStudio tools commonly pair with Intelligent Form Reader for an end-to-end intake flow?
Discussion
Loading discussion…