Physical and scanned documents are a persistent source of friction in CRM workflows. Business cards collected at events, signed forms received by email, supplier invoices, customer ID documents, and shipping labels all contain structured data — but extracting that data into CRM fields typically means manual keying, which is slow, error-prone, and unscalable as volumes grow.

Zia's Intelligent Character Recognition (ICR) feature eliminates that friction by reading the content of scanned or photographed documents and mapping the extracted data directly into the corresponding fields on your CRM records. The process is triggered when an image is uploaded to a designated field, making document processing a seamless part of the record creation or update workflow rather than a separate step.

Supported Document Types

Zia ICR is designed to handle the document types most commonly encountered in sales, support, and operations workflows:

• Business Cards — Extracts contact name, job title, company, phone numbers, email addresses, and website from cards photographed or scanned after events or meetings.
• Forms — Reads completed form fields — both printed and handwritten — and maps values to the corresponding CRM fields using a configured template.
• Invoices — Extracts line items, totals, vendor details, invoice numbers, and dates from supplier or customer invoices, reducing manual entry in accounts and purchase workflows.
• ID Documents — Reads structured fields from identification documents such as passports and driving licences, useful in onboarding, compliance, or verification workflows.
• Shipping Labels — Extracts recipient details, reference numbers, postal codes, and carrier information from shipping labels, supporting logistics and fulfilment tracking within CRM.

Two Extraction Methods

Zia ICR offers two distinct approaches to data extraction, each suited to different document types and workflow needs:

Template-Driven Extraction

You define a document template that maps regions of the document to specific CRM fields. Zia uses the template to locate and extract the correct value from each region when a matching document is uploaded. Best suited to standardised documents with consistent layouts, such as your own internal forms, partner invoices with a fixed format, or ID documents with known field positions.

• Requires template configuration per document type
• High accuracy on consistent, structured layouts
• Cannot extract data from table structures within documents

Zero-Shot Extraction

Rather than defining a template, you describe which fields you want Zia to extract using plain language prompts. Zia uses its language model to identify and extract those values without needing to be trained on the document layout. Best suited to variable document types — business cards, ad-hoc forms, supplier invoices with differing layouts — where a fixed template would be impractical to maintain.

• No template configuration required
• Handles variable and unstructured document layouts
• Cannot capture embedded images within documents

Creating One Record from Multiple Images

Zia ICR can now create a single record from up to three images uploaded together, such as an ID, a passport, and a business card. Zia extracts the relevant information from all of them and maps it to the correct CRM fields, giving a complete and accurate record before it's even created.

Picture a sales rep at a trade show who collects a business card, its reverse side, and a printed spec sheet for one prospect. Previously, the rep had to upload and process each image one at a time: upload the card, wait for Zia to extract the fields, verify and save the record, then repeat the whole cycle for the back of the card and again for the spec sheet. Each new upload risked overwriting data from the previous one, increasing the chance of duplicate or lost information.

Now, the rep uploads all three images together in a single batch. Zia ICR extracts and reconciles the data from all three at once and maps everything to the right CRM fields for one clean review and save, a complete, accurate record in a single pass instead of three.

What's New

• Upload multiple images. Create a single record from up to three images in one go. Zia extracts each one and fills in the fields with the corresponding values, based on the rules defined in the relevant ICR rule.
• Add images during extraction. At any point after uploading images and before creating the record, more images can be added to the same batch.
• Choose the best value among duplicates. When two or three images contain different values for the same field, such as two phone numbers or two spellings of a name, Zia lists every value it found under that field, letting the user pick which one to keep.

Setting Up ICR

ICR rules are configured per image field and per module. Navigate to Setup > Zia > Vision > Intelligent Character Recognition to create and manage your rules. For each rule, select the target module, the image field that will receive the document upload, the document type, and the extraction method. For template-driven rules, define the field mapping using the template builder. For zero-shot rules, specify which CRM fields to extract and provide plain-language descriptions of what each field contains.

Known Limitations

• Template-driven extraction cannot read table data. Documents containing tabular content (e.g., itemised invoice line tables) cannot have those tables extracted via a template rule — field-level extraction outside tables works as expected.
• Zero-shot extraction cannot capture embedded images. If a document contains photographs, logos, or other image elements, zero-shot extraction will not retrieve them — only text-based field values are extracted.
• ICR recognises human faces, not general objects. When processing identification documents containing photographs, Zia can recognise and handle human faces — but it is not a general object recognition system and should not be used for image classification tasks better suited to the Image Validation feature.
• Multi-image upload is capped at three images per record. A fourth image cannot be added to the same batch; a separate record or a separate upload is needed instead.

Availability