AI-Powered Filling
When you use Auto-Detect Mappings, the AI matches extracted document data to your Excel columns. You can control how this happens with three fill modes.
Fill modes
Smart mode (default)
The AI intelligently fills only the columns where it has high confidence in the match. Columns where the data is ambiguous or no match is found are left empty.
Best for: Most use cases. Gives you a good starting point with fewer errors.
Replace mode
Overwrites all cells that have a match, even if they already contain data. Existing values in matched columns are replaced.
Best for: Re-processing a document when previous fills were incorrect.
Complete mode
Fills in all available data from the extraction, even for lower-confidence matches. More columns get filled, but some may need manual correction.
Best for: When you want maximum coverage and are willing to review/correct results.
How the AI matches fields
The AI uses semantic matching to pair extraction fields with column headers:
| Extracted field | Column header | Match reason |
|---|---|---|
invoice_number | "Invoice #" | Semantic similarity |
total_amount | "Total Due" | Meaning overlap |
vendor_name | "Supplier" | Synonym recognition |
invoice_date | "Date" | Context + type matching |
The AI considers:
- Field names — direct and semantic similarity
- Data types — numbers match number columns, dates match date columns
- Context — surrounding fields help disambiguate
Streaming fill
When you click Auto-Detect Mappings, columns fill in one at a time in real-time. This lets you:
- See which columns the AI is filling
- Catch obvious errors early
- Understand how the AI is mapping data
Cell-level retry
If a specific column was filled incorrectly:
- Click the cell
- Click Retry on that column
- The AI re-runs the mapping for just that column with more attention
This is faster and more targeted than re-running the entire auto-detect.
Retry All
Click Retry All to re-run the AI mapping for all columns simultaneously. Use this after you've made corrections to the template or extracted new data.