Extraction Models

Choose the right model for your document type, layout consistency, and volume.

Available models

Flash vs Deep tier

Instinct and Mastery support two processing tiers:

Set the tier in the workflow Extraction tab when using Instinct or Mastery.

Value verification (Instinct and Mastery)

Every value extracted by Instinct or Mastery is verified automatically before it reaches you. The confidence score on each field reflects that verification, so values that are likely wrong score lower and are flagged for review. No setup is needed.

Token usage per page

Usage depends on which workflow steps run:

Instinct and Mastery workflows with an OCR step consume 1 OCR page plus model tokens per page. Pretrained workflows consume 1 OCR page and 1 model token per page through extraction. OCR-only and Review workflows consume OCR pages only - there is no automatic extraction step to run.

Check Subscription for monthly limits on both pools.

Pretrained model examples

Common pretrained models include:

• Invoice
• Receipt
• Pay stub
• Tax forms
• ID document
• Bank statement
• Check
• Contract
• Mortgage forms
• Health insurance card
• Marriage certificate

Each pretrained model returns a predefined field set. Enable or disable individual fields when creating or editing the workflow.

Instinct setup

1. Create a workflow and select Instinct Model.
2. Add fields with a label and short description of what to extract.
3. Choose Flash or Deep based on document difficulty.
4. Upload test files and review results.
5. Adjust field descriptions if extraction misses values.

Instinct works well when you do not have a fixed template and need fast iteration.

Mastery setup

1. Create a workflow and select Mastery Model.
2. Define fields and provide an example document or mapping guidance.
3. Process a sample batch and correct fields in Files.
4. Use feedback to improve consistency on repeat layouts.

Mastery is best when you process the same document type at volume and want stable field output.

Bill of Lading and First Notice of Loss models

Bill of Lading and First Notice of Loss are Mastery models whose fields come prefilled for a document type, so you start from a working field list instead of an empty one. They behave and are billed exactly like Mastery.

1. Create a workflow and select Bill of Lading or First Notice of Loss.
2. On the Configure fields step, choose a type from the Document type list. The fields and the document description are filled in for that type.
3. Add, edit, or remove any field, table, or column, as you would with Mastery. Review the example values and replace them with ones from your own documents.
4. Choose Flash or Deep, then save.

Changing the document type replaces the fields. If you already edited the fields or the description, Foxello asks before replacing them. The Document type list appears only when you create a workflow; when you edit a workflow later, you edit its fields directly.

If a prefilled set is larger than your plan's field limit (see Field limits per plan), only the most important fields are added, in the order they appear on the document, and the least important ones are left out. You can add any of them back by hand as long as your plan has room.

Field options

When defining fields for Instinct or Mastery, each field has two optional toggles:

• Required field - if this field comes back empty on a document, that document is treated as failed extraction for this field (0% confidence). Under Conditional review, this fails the confidence threshold and sends the document to Review Hub even if every other field extracted cleanly. Use it for fields a document must always contain (e.g. an invoice number).
• True/false field - marks the field as a yes/no answer, such as a checkbox or tick mark on the document. The model returns true or false instead of free text.

You can also add an optional Document description above your fields - a short sentence describing what kind of document this is (e.g. "A quarterly board meeting minutes document"). This gives the model context before it looks for your fields, which can improve accuracy on ambiguous layouts.

Field limits per plan

Custom field labels (Instinct, Mastery, and Review) are capped per workflow by plan:

Each field counts as one label, and each column of a table counts as one label (the table itself does not count).

OCR use case

Choose OCR when you need searchable text from a document but do not need structured key-value fields. OCR output is not mapped to a custom schema.

Review use case

Choose Review when you want a defined field schema but not automatic extraction - for example, low-volume or highly sensitive documents where every value should be entered by a person rather than a model. Define fields the same way as Instinct or Mastery (name, description, and example), then fill in each value manually in Review Hub for every processed document.

Choosing a model

See Files docs for reviewing confidence scores and correcting extracted values.