Skip to main content

Zonal Extraction (Template Zones)

What this page is — marking fixed areas on a document type's form, so that values sitting in the same place on every incoming copy are read automatically in the Capture Inbox.

What it is for — so key values on identically laid-out forms are read automatically, and duplicates are caught before they become records.

The problem it solves — every captured invoice or application otherwise has its values retyped by hand, and a resubmitted invoice is only caught if someone remembers it.

Route: /org/papers/designer/<type> → Zones tab, and /org/papers/capture · Permission: Create/edit/retire project types and template versions. (or Manage org-canonical types/templates (projects inherit). for org-wide types) to mark zones; View this project's document types. to read them


1. What it is​

A zone is a rectangle on a page of the form, stored as percentages of the page's width and height, with a target field. When a captured file is classified as that type, each active zone is read and its value is carried into the new document.

Zonal extractionWhole-document AI classificationSmart fields
ReadsOnly the marked rectanglesThe whole fileValues already in Orbit
Needs a fixed layoutYesNo—
Cost on digital PDFsNone — reads the text layerAI creditsNone
Cost on scansAI credits per zoneAI credits—
Duplicate detectionOptional per zoneNoNo

Zones belong to the type, not a template version. They describe the incoming paper form, which does not change when you reword your own template.


2. Why you would use it​

  • Nobody retypes invoice numbers. At fifty supplier invoices a day, reading six values each is an hour of keying — and a source of transposed digits.
  • Duplicates are caught at the door. A zone checked against existing records flags a resubmitted invoice before it becomes a second payable.
  • Exact beats approximate. On digital PDFs the value comes from the text under the rectangle, with no model guessing, so no credits and no hallucinated numbers.
  • Nothing breaks if it fails. An unreadable zone falls back to normal AI classification, so adding zones never makes capture worse.

3. Step by step​

Marking zones​

  1. Open the type in the Type Designer → Zones tab. With no sample loaded, the tab says "Upload a sample to mark zones".
  2. Click Upload a sample of this form (later Replace sample). It is rendered for drawing only — never stored and never attached to the type.
  3. Drag a rectangle over the spot where the value always sits. Use the page arrows for multi-page forms.
  4. In the zone card:
    • Choose the Field — the type's fields are offered, or type a key.
    • Add a Label.
    • Optionally add a Cleanup pattern.
    • Set the existing-records check.
  5. Click Save zones ("Zones saved").
  6. Pause a zone without deleting it with its badge. The tooltip reads "Zone is read on capture — click to pause it" or "Zone is paused — click to activate".

How a capture is read​


4. Field reference​

FieldRequiredValidation — exact messages
FieldYeszones[n]: field_key is required
LabelNoFree text, shown on the capture chips
RectangleYeszones[n]: the zone rectangle has no area; zones[n]: coordinates must stay within the page (0-100%)
PageYeszones[n]: page_number must be >= 1
Cleanup pattern (regex, optional)Nozones[n]: extract pattern is not a valid regular expression. The first bracketed group is kept, otherwise the whole match. No match reports an error, never a wrong value
Existing-records checkNoLooks for documents of the same type whose field already holds this value. Off reads "Not checked against existing records"

Sample upload rules​

MessageCause
"the sample must be a PDF, PNG, JPG or TIFF file"Wrong file type
"the sample file exceeds the 20MB limit"Too large
"this file could not be opened — upload a PDF or image sample of the form"Corrupt or password-protected

In the Capture Inbox​

You seeMeaning
Green chip, e.g. Invoice number: INV-2026-041Read from the zone; hover shows text layer or vision
Amber chip with ⚠Already exists on another document of this type — possible duplicate
zones unreadable — AI text usedThe zones could not be read; normal classification applies
Read zones for this type (Convert drawer)Re-reads the file against a different type you picked

Re-reading refuses with "document type "…" has no active zones — mark them in the Type Designer's Zones tab first" when that type has none.


5. Drawing zones that survive real paper​

DoWhy
Draw a little wider than the valueScanned copies shift slightly. Scans get a 0.5% margin, but only that
Use a cleanup pattern for anything with a formatINV-([0-9]+) keeps 041 from Invoice No. INV-041 (copy)
Put one value in one zoneTwo values in one rectangle come back joined
Turn on the existing-records check only for identifiersChecking dates or amounts flags everything
Test with a scanned copy as well as a digital oneScans take the vision path and cost credits

Zones assume the same layout. A supplier who redesigns their invoice needs a new type or re-drawn zones.


6. Worked example​

Gatiro's accounts team receives about 40 invoices a week from Northwind Supplies, always on the same template.

  1. They create the type Supplier invoice — Northwind with fields invoice_number, invoice_date, total_amount and po_reference.

  2. On Zones they upload last month's invoice and draw four rectangles:

    LabelFieldPatternCheck existing
    Invoice numberinvoice_number(NW-[0-9]{6})✓
    Invoice dateinvoice_date——
    Totaltotal_amount([0-9,]+\.[0-9]{2})—
    POpo_reference(PO-[0-9]+)—
  3. Monday's emailed PDF arrives in the Capture Inbox. Classify suggests the Northwind type, and four green chips appear: NW-004417, 03/03/2026, 1,24,500.00, PO-8812. No credits are used, because the PDF has a text layer.

  4. On Wednesday a scanned copy of the same invoice arrives. The number chip is amber: already on document "Invoice NW-004417". The clerk discards it instead of creating a duplicate payable.

  5. On Friday a paper invoice is scanned. Its zones take the vision path, using a small credit charge per zone, and fill correctly.


7. The admin contract​

What must be configuredOtherwise
An active type with a published version and fieldsZone values have nowhere to land
The classifier suggesting this typeZones are read only for the suggested type (use Read zones for this type otherwise)
Document OCR enabled, with AI creditsScans fall back to ordinary classification; digital PDFs still read
A capture source feeding the inboxNothing to read — see capture inbox

Zones can also be managed through the API (GET and PUT on /api/v1/ext/papers/types/<id>/zones) and the assistant — see API and automation. Start with your highest-volume form.


8. Downstream​

  • Convert carries zone values into the new document's fields.
  • Duplicate warnings never block. The decision stays with the person converting.
  • The file uploaded as a sample is never stored. Only the captured files are.

9. Don't confuse this with…​

Capture inboxWhere files arrive and are classified. Zones enrich that step
Smart fieldsFill values from Orbit data, not from paper
Extract key termsAI reading of a whole agreement's meaning, not fixed positions

10. Troubleshooting​

SymptomCause
No chips after ClassifyThe zone is paused, the classifier suggested another type, or the layout differs from the sample
The chip holds extra wordsTighten the rectangle, or add a cleanup pattern with a group
The zone reports an error instead of a valueThe pattern did not match what was read
Scans never produce chipsDocument OCR is disabled or out of credits
Every document is flagged as a duplicateThe check is on a non-unique field, such as a date
Save refusedSee the zone messages in §4 — an empty field or a rectangle off the page