Extract data from PDFs
In this recipe you turn a stack of supplier contracts into a structured sheet: one row per contract, with the counterparty, start date, contract value and a risk category extracted by AI — every value backed by a citation you can open in the source document.
What you need: a Quanty account with an active subscription, and a handful of PDF contracts on your machine.
Step 1 — Create a workbook
From Home, click New workbook. Give it a name such as Supplier contracts
(names must be at least 3 characters and unique in your organization), an optional
description, and pick or generate a Workbook Avatar. Open the workbook's sheet.
Step 2 — Add your contracts as rows
The fastest path is to drag the PDFs straight onto the grid — you'll see Drop files to add rows. Each file becomes a row in a File column.
Alternatively, click Add files in the sheet toolbar. In the dialog you can pick From Files (documents already uploaded to the Files page) or Upload New. Files are analyzed after upload; wait until the processing bar shows the rows as ready. If something fails, use Retry failed files.
Step 3 — Add AI columns
Click Add column for each field you want to extract:
| Column | Column type | Configuration |
|---|---|---|
| Counterparty | Text | Extraction type Entity Extraction, prompt: which company is the other party |
| Start date | Text | Extraction type Date — finds a date and normalizes its format |
| Contract value | Number | Data Format set to Currency with your symbol and decimals |
| Risk category | Single choice | Options such as low, medium, high; AI picks exactly one |
For the counterparty column, a Custom Prompt works too — it is the most flexible type. Prompts can be up to 1500 characters, and you can pick the Model per column (Economy, Standard or Premium tiers).
Step 4 — Turn on Cite sources
In each AI column's settings, enable the Cite sources toggle. It adds page references to every extracted value (at roughly 20% more tokens), which is what makes the review step below possible.
Step 5 — Run AI
Click Run AI in the toolbar. A confirmation explains that pending cells will be processed and that this uses tokens from your organization's AI budget — you can watch progress and Stop at any time (values already computed are kept). Cells move through Pending, Processing and Completed; failures show as Failed.
Step 6 — Verify values against the source
Click any completed cell to open the cell drawer. You'll see the Value, a Why this value explanation, a Confidence score and Citations as page chips. Click View in Document to jump to the highlighted region in the PDF.
- Approve or Reject each value you check.
- If a value is correct and must never change, use Lock as ground truth. Later reruns record drift instead of overwriting it.
- If a value is slightly off, use Refine with a follow-up instruction, for
example
exclude one-off setup fees.
Step 7 — Export
When the sheet looks right, export it as CSV or Excel from the sheet menu and share it with your team — or keep going and build a deck from it.
Where to go next
- Columns & AI extraction — all column types, @-mentions and branching
- Reviews & audit trail — approvals, ground truth and restore
- Files & Knowledge Base — how documents are processed and shared