Skip to main content

AI Design

The AI Design feature analyses an uploaded report document (PDF, image, Excel, or Word file) and suggests a blueprint structure — section names, field names, field types, and required flags — based on what it finds in the document.

Access: Blueprint Designer → ✨ AI Design button (header, purple, robot icon).

When to use AI Design​

AI Design is most useful when you are digitising an existing manual report:

  • A Word or PDF monthly progress report with named fields and tables.
  • An Excel spreadsheet template contributors currently fill in.
  • A printed form that has been photographed or scanned.

AI reads the document's structure — headings, table columns, field labels — and maps them to appropriate field types.

Access requirements​

The ✨ AI Design button is only visible when your account has AI access enabled. This requires:

  • Org-level AI access enabled by a Workspace Admin.
  • User-level AI access granted by an Admin.

If you do not see the ✨ AI Design button, contact your Workspace Admin.

Supported file types​

File typeHow AI reads it
JPEG, PNG, WebPSent directly as images to the vision model
PDFEach page is rendered as an image — all pages are processed
Excel (.xlsx)Sheet text and cell data extracted and sent to the AI as structured text
Word (.docx)Document text extracted and sent to the AI as structured text

Limit: Up to 20 files per session. For multi-page PDFs, all pages count toward this limit.

You can also paste an image directly from the clipboard (Ctrl+V / Cmd+V) — useful for pasting screenshots of existing reports.

Workflow: Upload → Analyze → Review​

Step 1 — Upload​

  1. Click ✨ AI Design in the Blueprint Designer header.
  2. The AI Blueprint Design panel opens on the right (560px drawer).
  3. Drag and drop files onto the upload area, or click to browse.
  4. Add multiple files if the report spans several documents or pages.
  5. Optionally, type additional instructions in the text field below the uploader. Example: "Focus on extracting cost and schedule fields. Ignore the cover page."
  6. Select the AI model:
    • GPT-4o ($$) — Best for general report analysis and table extraction.
    • Claude Sonnet 4.6 ($$$) — Excellent at structured data extraction and nuanced field typing.

Step 2 — Analyze​

Click Analyze. The AI processes your files:

  • Images and PDF pages are sent to the vision model.
  • Excel and Word files have their text extracted first, then sent to the model.
  • A loading indicator shows the analysis is in progress (typically 5–30 seconds depending on file count and size).

Step 3 — Review​

The AI returns a suggested blueprint structure. The review panel shows:

Suggested sections — collapsible panels, one per section the AI identified.

Each section shows:

  • Section name — the AI's proposed name (e.g. "Cost Report", "Risk Register").
  • Section type — single_row (one set of values) or multi_row (repeating rows / table).
  • Confidence — how confident the AI is in this section.

Each field within a section shows:

  • Field name — the proposed label.
  • Field type — the data type the AI chose (text, currency, date, select, etc.).
  • Required — whether the AI flagged this field as required.
  • Options — for Select fields, the AI proposes the option list.
  • Confidence — field-level confidence.

Confidence levels:

Tag colourConfidenceWhat it means
GreenHighAI is confident in this field's name and type
GoldMediumAI recognised the field but the type or name may need review
OrangeLowAI found a value but is uncertain about its classification
DefaultNoneAI extracted something but could not classify it reliably

Review medium and low confidence suggestions carefully before accepting.

Accepting suggestions​

After reviewing, click Accept & Apply to add the suggested sections and fields to the blueprint.

What happens:

  • Each suggested section is created in the Sections panel.
  • Multi-row sections are configured as repeating rows.
  • Each suggested field is added to its section with the proposed type, label, field key, required flag, and options.
  • A summary is shown: "Created 4 sections and 23 fields" (with any errors listed).

After applying, the blueprint design updates immediately. You can then:

  • Rename any field or section.
  • Change a field type that was misclassified.
  • Add additional fields the AI missed.
  • Reorder sections and fields.

The Undo stack does not cover AI Design application — review carefully before accepting.

Re-analyzing​

If the suggestions are poor quality:

  • Click Re-analyze to run the AI again with the same files (a different model may produce better results).
  • Remove files and add different/better-quality source documents.
  • Add specific instructions to the instructions field (e.g. "The 'Status' column contains the values: Not Started, In Progress, Complete, On Hold").

Troubleshooting​

AI suggested wrong field types:

  • Currency fields detected as Text: add an instruction "The Value column contains currency amounts in AUD".
  • Date fields detected as Text: ensure dates in the source document are in a recognizable format.
  • Select fields detected as Text: list the valid option values in the instructions field.

AI missed sections:

  • For multi-page documents, ensure all pages are uploaded.
  • If a section is handwritten or in an image-heavy layout, add a text description of it in the instructions.

No suggestions returned:

  • The document may contain only images with no readable text. Try a higher-resolution scan or a text-based version.
  • Very large PDF files (50+ pages) may time out — split into smaller files.

Button not visible:

  • Confirm AI access is enabled for your account. Contact your Workspace Admin.