Skip to main content

AI Fill: Extracting Edition Data from a Source Document

This tutorial walks through using the AI Fill feature in the Edition editor to extract data from a contractor PDF progress report and populate the edition fields — replacing manual transcription.

Scenario: Your contractor has emailed a monthly progress report as a PDF. The report contains cost data, schedule status, RAG indicators, and a risk summary. You need to enter this data into the edition before submitting for review.

Assumed blueprint: A "Monthly Progress Report" blueprint with sections:

  • Project Status — single entry (Status, RAG, Narrative)
  • Cost Report — repeating rows (Period, PeriodCost, BudgetCost, Variance)
  • Risk Register — repeating rows (Risk ID, Description, Severity, RAG, Mitigation)

Before you start​

  • Confirm the edition is in Draft status.
  • Confirm the ✨ AI Fill button is visible in the edition toolbar. If not, contact your Workspace Admin to enable AI access.
  • Have the contractor's PDF report ready (or another source document).

Step 1 — Open the AI Fill panel​

  1. Open the edition in Draft status.
  2. Click ✨ AI Fill in the header toolbar (purple button with robot icon).
  3. The AI Fill panel opens on the right side of the editor.

Step 2 — Upload the source document​

  1. Drag the contractor PDF onto the upload area in the panel, or click to browse to the file.
  2. The file appears in the upload list with a preview thumbnail.
  3. If the report spans multiple documents (e.g. a separate cost schedule Excel file), add them too — click to upload more.

Supported formats: JPEG, PNG, WebP images; PDF (all pages rendered); Excel (.xlsx); Word (.docx).

You can also paste a screenshot (Ctrl+V / Cmd+V) if you have a screengrab of a dashboard view.

Step 3 — Add instructions (optional but recommended)​

The Instructions field below the uploader accepts free-text guidance. This is particularly useful when:

  • Field names in the document differ from blueprint field names.
  • The document has multiple sections and you only want part of it.
  • Date or number formats need clarification.

Example instructions for this scenario:

Extract cost data from the "Monthly Cost Summary" table.
The "This Period" column = PeriodCost, "Budget" column = BudgetCost.
Dates are in DD/MM/YYYY format.
Risk severity values in the document are: Critical, High, Medium, Low.

Step 4 — Select the AI model​

Choose the model from the Model dropdown:

  • GPT-4o ($$) — good for general table and text extraction from PDFs and images.
  • Claude Sonnet 4.6 ($$$) — excellent at structured extraction from complex, dense documents.

For a standard contractor PDF, GPT-4o is usually sufficient. Use Claude Sonnet 4.6 for documents with complex formatting, mixed languages, or very dense tables.

Step 5 — Analyze the document​

Click Analyze. The AI processes your documents.

Progress messages rotate in the panel:

  • "Reading report values..."
  • "Matching data to blueprint fields..."
  • "Extracting table rows and cell values..."
  • "Normalizing dates and numeric values..."
  • "Mapping values to your report schema..."

Analysis typically takes 5–20 seconds for a standard PDF report.

Step 6 — Review extracted values​

The review panel shows suggested values for each field in your edition, organised by section.

Example output for the Project Status section:

FieldExtracted ValueConfidenceNotes
StatusOn TrackHigh—
RAGGreenHigh—
NarrativeProgress this period...MediumTruncated — verify full text

Example output for the Cost Report section (3 rows extracted from a table):

PeriodPeriodCostBudgetCostConfidence
January$45,200$48,000High
February$51,300$48,000High
March$39,800$48,000Medium

Confidence tag colours:

  • Green (High) — accept with confidence.
  • Gold (Medium) — review before accepting.
  • Orange (Low) — verify carefully, likely needs correction.

Step 7 — Correct a misread value​

In the review panel, you notice the RAG for one risk was read as "Amber" but the PDF shows "Red". Before accepting:

  1. Click the extracted value cell in the review panel.
  2. Edit it to the correct value: "Red".
  3. The confidence level is not affected by your edit.

Step 8 — Check unmapped data​

Below the section tables, the Unmapped data panel shows values the AI found but could not match to any blueprint field.

For example:

  • "Contractor Reference: CR-2026-45" — no corresponding field in the blueprint.
  • "Sub-contractor: Acme Civil" — if there is a "Sub-contractor" field, add an instruction for future runs.

Unmapped values are not applied — they are shown for information only.

Step 9 — Accept and apply​

Click Accept & Apply.

The AI applies all suggestions to the edition:

  • Single-entry fields are populated with the extracted values.
  • Repeating section rows are created with the extracted data.
  • A progress bar shows the apply operation.
  • A summary shows: "Populated 3 sections, 27 fields. 0 errors."

The edition data updates immediately. The AI Fill panel closes.

Step 10 — Verify and correct​

Review each section in the edition:

  1. Click the Project Status tab — verify the Status, RAG, and Narrative fields.
  2. Click the Cost Report tab — check each row's cost values against the PDF.
  3. Click the Risk Register tab — verify risk descriptions, severity, and RAG values.

Correct any values the AI misread directly in form or grid mode. Use Undo (Ctrl+Z) to revert the AI fill entirely if needed.

Step 11 — Save and submit​

  1. Press Ctrl+S to save all changes.
  2. Verify no validation errors (red indicators in grid cells or the validation panel).
  3. Click Submit to send the edition for review.

What AI Fill does not do​

  • Does not auto-submit the edition — you always review and save manually.
  • Does not replace a reviewer's judgement — the review step is still required.
  • Does not process password-protected PDFs — remove the password before uploading.
  • Does not extract reliably from handwritten text — typed or digital documents give better results.