Turning a Word doc into a SharePoint form (with AI)
We tried to migrate a 30-question intake form from a Word template to a SharePoint Forms365 form. AI did 80% of the work.
We've been quietly shipping AI form generation in SharePoint Forms365 for a few months now. This week we put it through a real customer scenario, migrating a 30-question intake form from a Word template into a designed SharePoint form, to see how much manual work it actually saved.
Short answer: ~80%. Here's what worked, what didn't, and what we're improving.
The starting point
The customer (a logistics firm we'll call Acme) had a paper intake form for new courier driver applications. It existed as a Microsoft Word document with:
- 4 sections (Personal details, License & qualifications, Vehicle access, Background check)
- 30 individual fields
- 8 conditional sub-questions ("If you ticked yes to X, please describe Y")
- 2 signature blocks
- 1 photo upload spec
Their existing process: print, fill in by hand, scan, file in OneDrive. They wanted: a SharePoint Forms365 form bound to a Drivers list.
The AI generation step
Total operator time: 3 minutes, mostly waiting.
- Created a new SharePoint list called Drivers with no columns.
- Opened SharePoint Forms365, clicked + New form → From a document.
- Dropped in the .docx.
- Chose Create list columns from AI suggestions.
~10 seconds later: a draft form with 30 fields, grouped into 4 sections that mirrored the Word doc's structure. Field types correctly identified for most: text for names, date for dates, choice for "Yes/No" questions, number for "Years of experience".
The list was simultaneously updated with 30 new columns matching the inferred schema.
What the AI got right
- Section grouping ✓, matched the Word doc's headings
- Field types ✓, date pickers for dates, choices for enumerated answers
- Required flags ✓, fields marked "(required)" in the source became required
- Choice options ✓, radio-button enumerations transcribed verbatim
- Field labels ✓, including punctuation and capitalisation
What needed manual fixes
- Cross-field rules, The "If you ticked yes" sub-questions weren't auto-wired with
visibleIf. We added them manually (8 rules, ~2 minutes). - Signature blocks, The AI emitted plain text fields. We replaced with two Signature components (~30 seconds).
- Photo upload, Emitted as an Image field with no constraints. We set max dimensions and allowed formats (~30 seconds).
- Sensitive field marking, Two fields (driver's licence number, date of birth) needed audience trimming so only HR sees them. AI didn't infer this; we added it (~1 minute).
Total manual fix time: ~5 minutes.
Net time saved
| Step | AI-assisted | Manual |
|---|---|---|
| Create list columns | 0 (auto) | 30 mins |
| Create 30 form fields | 0 (auto) | 45 mins |
| Group into sections | 0 (auto) | 5 mins |
| Conditional rules | 2 mins | 15 mins |
| Signature + photo + audience | 2 mins | 5 mins |
| Total | ~5 mins | ~100 mins |
That's a 20x speedup on a representative real form. Almost all of the saved time is in the "creating fields and setting basic properties" phase, the part that's pure mechanical work.
What we're improving
Based on this and 50-odd other test runs we've watched:
- Cross-field rule inference. We're training the prompt to spot
"If you answered yes" patterns and emit
visibleIfautomatically. - Field-type heuristics for signatures and photos. The text "sign here" is a strong hint we're not currently weighting heavily enough.
- Audience inference. If the doc says "HR only", we should propose audience trimming.
Try AI generation on your own form: open SharePoint Forms365 → + New form → From a document. We'd love to hear what worked and what didn't on your forms.