AI Extraction
WORKloom turns unstructured business documents into structured, reviewable records. Extraction is available for contracts, resumes, and expense receipts or invoices.
AI extraction is an input aid, not an approval or source of truth. Nothing should be treated as correct until a person has reviewed it.
Why document ingestion matters
Business information rarely starts in a database-ready format. The same kind of information can arrive as:
- A Word document created internally
- A PDF generated by another system
- A PDF whose useful information is embedded in its visual layout
- A scan
- A phone photo such as a JPEG receipt
- Plain text with little or inconsistent formatting
Traditional automation works well only after someone has normalized that input. WORKloom moves the normalization step into the product: it interprets a supported source and maps relevant information into the fields required by the workflow.
The result is useful beyond saving keystrokes. Structured dates can drive contract alerts. Reviewed compensation and approved hours can feed gross payroll. An extracted expense amount can move through approval and payment. The document becomes the beginning of a connected process rather than a file that someone must summarize elsewhere.
Supported inputs
Supported formats depend on the record being created:
| Workflow | Formats | Notes |
|---|---|---|
| Contract or offer | PDF, Word (.docx, .doc), plain text | Used to pre-fill a contract for review |
| Resume or CV | PDF, Word (.docx, .doc), plain text | Can also recover a suitable profile photo embedded in supported documents |
| Receipt or invoice | PDF, JPEG, PNG, WebP | Supports generated PDFs as well as photographed or scanned receipts |
A document type that is valid for one workflow may not be valid for another. The upload screen shows the formats and workspace file-size limit that apply in that context.
:::tip Different source, same destination A supplier-generated PDF receipt and a JPEG photo of a paper receipt both lead to the same reviewed expense-claim form. The source format changes; the structured business process does not. :::
Extraction workflow
The workflow is similar in each area:
- Upload a supported document.
- Wait while WORKloom extracts likely values.
- Review the proposed fields, confidence information, and warnings.
- Correct or complete the values.
- Confirm the form to create the record.
The uploaded file and the structured record serve different purposes. Extraction reads the file; saving the reviewed form creates the contract, person, or claim.
Contract extraction
Use contract extraction from Contracts, the contract creation flow, or the New Contract by Document dashboard widget.
WORKloom may propose:
- The person named in the document
- Position and contract type
- Start and end dates
- Compensation amount, rate, frequency, and currency
- Leave entitlement
- Notice and probation details
- Other supported contract fields
The review step shows extracted values and confidence information. After review, WORKloom pre-fills the contract form; the contract is not created until you check that form and submit it.
Pay particular attention to the selected person, dates, currency, pay frequency, and whether an amount is hourly or fixed. These values can affect reports and payroll.
Supported contract inputs include PDF, Word (.docx and .doc), and plain-text files.
Resume extraction
Use resume extraction when adding a person, or add the Add Person by Resume widget to your dashboard.
WORKloom may extract:
- Name and contact details
- Location
- Employment and education history
- Skills
- Candidate-related profile information
The result creates an inactive person or candidate profile that can be reviewed and completed before an invitation is sent. Check for duplicate people before saving, especially when a resume uses a different email address or name variation.
Supported resume inputs include PDF, Word (.docx and .doc), and plain-text files. WORKloom can also attempt to recover a suitable profile photo embedded in a supported resume.
Expense extraction
Open Expenses → Create with AI, or use the Create Claim by Document dashboard widget.
Upload a receipt or invoice in one of these formats:
- JPEG
- PNG
- WebP
WORKloom may propose:
- Claim type
- Expense date
- Merchant
- Amount and currency
- Description
You still choose or confirm the cost centre and any workspace-specific categorization. The review form also shows an extraction summary and warnings when the document is unclear.
Before saving, compare the proposed amount with the receipt total, confirm the currency, and check that the date represents the expense date rather than an upload, due, or payment date. The normal expense approval workflow begins only after the claim is created.
A receipt can contain several monetary values, such as subtotal, tax, tip, and total. Always confirm that WORKloom selected the reimbursable total.
Confidence and warnings
Contract and resume reviews can provide field-level confidence or comments. Expense extraction provides an overall confidence summary and warnings.
Treat these signals as review priorities:
| Signal | What to do |
|---|---|
| High confidence | Verify against the source, especially identity, dates, and money |
| Medium confidence | Check formatting and surrounding document text |
| Low confidence or warning | Compare carefully and expect to correct the value |
| Blank field | Enter it manually if the record requires it |
Confidence describes how clearly the model interpreted the document; it does not prove that the source document itself is correct.
From extraction to real workflow
Extraction becomes valuable when the reviewed data is reused:
- Resume → profile → candidate → member — identity and experience move forward without recreating the person.
- Contract → alerts, time, and payroll — reviewed dates and compensation become part of operational calculations.
- Receipt → claim → approval → payment or payroll — the extracted transaction follows a controlled reimbursement process.
The source file remains evidence and context. The structured record provides filtering, reporting, permissions, approvals, and automation.
Permissions and availability
Upload, AI extraction, and record creation can be separate permissions. A user may be able to create a record manually without seeing the AI option. Dashboard extraction widgets appear only when the necessary permissions and workspace feature are available.
File-size limits are set by the workspace. If an upload is rejected, check its type and size before trying again.
Safe review checklist
Before saving any extracted record:
- Confirm the person or claimant.
- Verify every date.
- Verify monetary amount, currency, and frequency.
- Review low-confidence values and warnings.
- Remove information that belongs to a different person or transaction.
- Fill required fields that were not present in the document.