AI Form Fill is designed to help teams complete forms faster using voice recordings, photos, and contextual information. The quality of generated outputs depends on the quality of the information provided and the workflow being used.
Best Practices for Better Results
1. Speak Clearly and Naturally
Storm works best when recordings are clear, detailed, and easy to understand. For better results:
Speak at a normal pace.
Reduce background noise where possible.
Avoid multiple people speaking at the same time.
Include important details and context.
Useful details to include:
Company names, equipment names, locations, dates and times, defects or issues, and actions taken.
Use exact times where possible — '08:30' rather than 'around 8'. Storm flags approximate times for review rather than assuming
Relative dates resolve correctly — 'yesterday at 7am' or 'Monday morning' are understood. Storm has today's date.
2. Provide Detailed Context
Specific descriptions help Storm generate more accurate outputs, improving field mapping, overviews, dropdown selections, table entries, and photo matching.
Instead of: “There was an issue with the scaffold.”
Try: “The scaffold on the western elevation was missing mid-rails near the access ladder and was isolated until repairs were completed.”
Tip: Grouping in your dictation produces grouped output. For example 'Area A... Area B... north pit' yields area-headed bullets; 'concreters... steel fixers' yields trade-headed bullets. If the structure isn't what you wanted, restate the grouping explicitly — for example, 'group by trade today'.
3. Use Photos to Support the Recording
Photos can significantly improve AI Form Fill accuracy. Attach photos that clearly show defects, equipment, work areas, site conditions, signage, or permits and documents. Storm inspects photos individually and attempts to map them to the most relevant parts of the form.
Photos are processed in the exact order they are arranged on the canvas. Each photo in the canvas grid displays a numbered badge — this badge reflects the order Storm uses when processing. You can rearrange photos on the canvas before running AI Form Fill to control which photo is treated as first, second, third, and so on. This means positional voice instructions work reliably — for example, "put the first photo in the defect evidence field" or "the third photo shows the serial number".
For OCR-critical photos such as dockets, nameplates, and labels, shoot flat and square-on in good light. Framing determines how Storm routes a photo — a close-up reads as defect evidence, a wide shot reads as site context. Take one photo per subject; for example, a delivery docket and a product label should be separate photos. Multiple angles of the same defect attach to a single entry, not separate rows.
For field documents like plant dockets, permits, and snag lists, photograph them flat and square-on in good light. Storm has dedicated extraction rules for these document types and will pull the relevant data directly into the form.
4. Attach Documents as Context
You can attach documents directly to the AI Form Fill canvas using the paperclip button. This is useful when you have a delivery docket, SWMS, spreadsheet, or email chain you want Storm to read without re-uploading files already on the form. Supported file types: PDF, DOCX, XLSX, CSV, and EML.
📓 Note: Document attachment is currently in beta and available on Dashpivot Web only. DPM and SMM support is coming soon.
5. Give Clear Instructions
You can provide instructions during recordings or within typed notes. Storm attempts to distinguish instructions from actual form content during processing.
“Rewrite this as meeting minutes”
“Use paragraph format instead of bullet points”
“Add another row to the workers table”
“Provide an overview of the inspection findings”
6. Start With Lower-Risk Workflows
When introducing AI Form Fill, start with simpler workflows first to establish a smooth early adoption experience. Higher-risk workflows may require additional review processes and governance controls.
Recommended Starting Workflows: Site diaries, inspections, daily reports, pre-start briefings, and meeting minutes.
Higher-Risk Workflows: Permits, compliance-critical forms, and approval-heavy processes.
7. Use Multiple AI Form Fill Sessions
You can run AI Form Fill multiple times before saving a form, allowing documents to evolve progressively during a workflow. For example, you can record additional updates later in the day, attach more photos, ask Storm to rewrite sections, or continue filling missing details.
When re-running AI Form Fill on a form that already has content, Storm appends new information under the relevant headings. Existing content is preserved verbatim and never rewritten.
8. Reference Historical Forms from the Same Folder
Storm can use previously completed forms in the same folder as fill context. To get the best results:
Name the source form specifically in your directive — for example, "use yesterday's site diary" or "same crew as the last pre-start check"
Be precise — Storm may find it ambiguous which field you are referencing if your directive is too general
Confirm that both the form you are filling and the historical form you are referencing are in the same folder
Storm cites which historical form informed each fill decision, so you can verify the source
Works across template types within the same folder
If Storm cannot find the referenced form, it will tell you rather than guessing
Note: photo files from historical forms cannot be referenced
Current Limitations
Missing Information & Unfilled Fields
AI Form Fill may not fill every field. Storm only fills fields when confidence is high enough. If Storm is unsure about spoken information, dropdown matching, linked list selections, or photo interpretation, it may leave fields blank for manual review instead of guessing. This behavior is intentional to prevent inaccurate information.
Unsupported Form Inputs
Storm currently does not have the capability to work inside Dashpivot Actions
Processing Time
How quickly Storm fills your form depends on what you've submitted. Type a short, simple instruction — like "all pass" or "say yes to everything" — and you'll usually see results in under 10 seconds.
Short voice notes of 10 seconds or less (single recording, on Dashpivot Web) now take a fast lane and typically complete in approximately 10–20 seconds.
Voice recordings longer than 10 seconds, multi-modal inputs (such as recordings combined with photos or documents), and recordings on other platforms still use the full processing pipeline and will take a little longer. Either way, Storm figures out the best approach on its own — you just hit the button and wait.
Photo Mapping is Best-Effort
Storm attempts to associate photos with table rows, workflow steps, and relevant sections using image contents, surrounding context, upload order, and nearby form information. Photo mapping may occasionally require manual review or adjustment.
People Fields
Storm matches names from audio or photos to real workspace users and fills People fields automatically on Dashpivot Web.
If a name is unknown or ambiguous — no clear match to a workspace user — Storm leaves the field for review rather than guessing.
If a person is already present in a People field and Storm finds no new match from your input, the existing value is preserved and not overwritten.
When a match is found, the People field populates immediately — you do not need to save and reload the form.
People field support is available on Dashpivot Web only; mobile is not affected.
Group Discussions and Multi-Speaker Audio
Storm can process meetings, toolbox talks, and group recordings. Speaker labels are generated in transcripts when multiple speakers are detected. For better results, avoid people talking over each other, identify speakers verbally where possible, and provide a clear overview of key decisions. Speaker attribution may still require manual review.
Multilingual Behavior and Profanity
Feature | Storm Performance & Behavior |
Multilingual Input | Supports multiple languages. It attempts to detect the spoken language, detect the template language, and generate responses in the destination template language. (Results vary by recording quality and terminology). |
Profanity & Language | Storm does not automatically filter profanity or offensive language from transcripts or generated outputs. Recorded content may be transcribed exactly as spoken. |
Human Review Is Still Required
Storm is designed to assist with drafting and completing forms, but users remain responsible for reviewing submitted content. Always review generated overviews, mapped photos, dropdown selections, table rows, required fields, and logic-triggered sections.
What Storm Does NOT Do:
Approve workflows
Sign forms
Complete signature fields
Automatically submit forms