Channel Configuration — Document Processing

In the Document Processing area, you define what Workist does with an incoming document once it has arrived in the channel — which fields are extracted, the confidence level above which processing happens automatically, which master data is consulted, and how documents that do not belong in the channel are handled.

This page provides an overview of the seven subpages in the order in which they appear in the navigation. For each of the first five subpages, there is a detailed step-by-step guide:

1. Document — general processing settings per channel (localization, automation, date format, validation, …).
2. Header Fields — which header data is extracted.
3. Line Item Fields — which fields are extracted per line item.
4. Agentic Extraction — optional LLM-based extraction of additional fields. In the navigation it appears as two entries, Agentic Header Extraction and Agentic Line Item Extraction; both are described in a shared guide.
5. Master Data — which lookup definitions are used for articles, customers, addresses, … (see Step 5).
6. Discard — rules for documents that do not belong in the channel (see Step 6).

Prerequisites

• Access to the channel.
• For Master Data: lookup definitions set up in advance with master data imports (see Transformer configuration for master data imports).
• For Agentic Extraction: an understanding of which additional fields the AI should extract and which data sources are available to it for this purpose.

Step 1 — Document: central processing settings

In the navigation under Document Processing, select the Document entry. This page is organized by topic. Each topic block has its own Save button.

Topic block	Key settings
Localization	Time zone and holiday calendar of the channel. Overrides the customer-level settings for this channel.
Automation	Automation enabled as a global switch. Confidence threshold for header data / line items (%) defines the confidence level above which a document is handed over to the target system without assistance.
Data formatting	Date format, decimal separator, address format (street name first / street number first), options for decimal quantities and invalid article IDs.
Unit conversion	Conversion style (None / Round up / 30% rule), options for overriding the original fields and for partition-specific conversion.
Validation	Validation checks — multi-select from the business checks that Workist applies during processing (total amount, line item sum, line item number, unique order number, …).
Document-type-specific	Settings that only affect the respective document type (e.g. Support order forms, Remove freight cost line items, Enable Factur-X detection for invoices).
Backfilling	Backfill with email data and Document formats for backfilling — when a value is missing, Workist can fill it in from email content or a second document format.
Document filtering	Sort out document types and Filter sensitivity — Workist classifies incoming documents and can automatically sort out unsuitable ones (e.g. promotional emails).

Automation is set up together with Workist

Automation (the global switch and confidence thresholds) is normally set up and configured together with your contact at Workist. Making your own changes here is not recommended and is done at your own risk.

You can find the complete step-by-step guide to all topic blocks under Document.

Step 2 — Header Fields

In the navigation under Document Processing, select the Header Fields entry. Here you define which header data fields (e.g. order type, order number, customer contact) Workist extracts from the document and which properties each field has.

The layout has two columns: on the left, the list of configured fields; on the right, the configuration options for the currently selected field. In day-to-day use, you mainly define here which standard fields are extracted, whether a field is required, and in which order the fields appear. Most options are already sensibly preset for the prepared standard fields.

Complete guide: Header Fields.

Step 3 — Line Item Fields

In the navigation under Document Processing, select the Line Item Fields entry. The operating logic is identical to the Header Fields; compared to the Header Fields, a few line-item-specific options are added.

In addition to the header field options, you will find here, among others, Maximum field length, Allow invalid IDs, Keep external article number, Perform lookups, and Reset imperfect article matches.

Complete guide: Line Item Fields.

Step 4 — Agentic Extraction

In the navigation under Document Processing, you will find two entries: Agentic Header Extraction (per document) and Agentic Line Item Extraction (per line item). Here you set up LLM-based extractors that work in addition to the regular field extraction — typically for fields that are difficult to capture with classic extraction (e.g. free-text instructions, document-specific clauses).

When agentic extraction is worthwhile

When a field is not always in the same place, is embedded in free text, or requires combined logic (e.g. "derive order type from brand + shipping address"). For tightly structured fields, the regular extraction in Header Fields / Line Item Fields is sufficient.

Both variants are described in a shared guide: Agentic Extraction.

Step 5 — Master Data

In the navigation under Document Processing, select the Master Data entry. Here you link the channel with the lookup definitions used for matching.

A separate row is available for each lookup type — e.g. Client Lookup ID (customers), Article Lookup ID, Billing Address Lookup ID, Delivery Address Lookup ID, Contact Lookup ID, Conversion Factor Lookup ID, Framework Agreement Search ID, Order Matching Lookup ID, Quote Matching Lookup ID, as well as the corresponding Mapping Lookup IDs for stored mappings.

Each row contains:

• A text field for the lookup ID (UUID v4). Workist validates the format immediately after entry.
• An Upload file button to import master data as CSV without sending it via SFTP.
• A gear icon next to it that opens the transformer configuration of this lookup definition.

What happens if no lookup ID is set?

The corresponding master data matching simply does not take place. Example: Without an Article Lookup ID, the article number from the document is carried over to the output unchanged and is not checked against your product range.

After making changes, click Save. Using Reset, you can delete mappings for each lookup type before a selected date — useful when your master data has changed fundamentally and old learned data no longer fits.

Step 6 — Discard

In the navigation under Document Processing, select the Discard entry. Here you define what happens to documents that should not be processed in the channel (e.g. because document filtering classified them as unsuitable or an agent manually discarded them).

Field	Meaning
Recipient email addresses	Comma-separated list of addresses to which discarded documents are forwarded — typically an internal mailbox for manual review.
Send email back to the original sender	When enabled, a notification is additionally sent to the original sender. Caution: Only enable this once you have reviewed the reply template.

Recommendation

Always set at least one forwarding address. Otherwise, discarded documents end up in an internal audit log that is not part of the daily workflow.

Frequently Asked Questions

In which order do the configurations run? Per incoming document: document filtering → classic field extraction (header fields / line item fields) → agentic extraction (if configured) → master data matching → validation. If a stage is not sufficient, the document goes into assistance.

What is the difference between data label name and field label name? The data label name is the technical name in the output file and is therefore fixed. The field label name is only the label that appears in the Workbench (assistance, forms) and can be changed at any time.

Help and Support

If you encounter problems with processing, have the following information ready:

• the channel name and the area (Document, Header Fields, …),
• the ID or name of a specific example document that was not processed as expected,
• for master data topics: the affected lookup ID and ideally the most recently imported file,
• for agentic extraction: the configuration name, the prompt if applicable, and the most recent successful / failed examples.