This guide covers how to run an AI workflow on a project and review the results.
What is an AI Workflow?
An AI workflow is a collection of Tasks that can be run on a project. A task is a single function that your AI can run on one or multiple fields, such as extracting text from a photo or validating that a photo was taken at the correct location. Running a workflow will run each task at the same time and provide results under the designated fields.
How to Run a Workflow
Workflows can either be run manually, or based on setting a status associated with a workflow state determined in your site's Workflow configuration.
1. Open the project and select the workflow icon.
Open the project you wish to run the workflow on, then click the AI Workflow (Star) icon.
2. Select and configure the workflow.
Select the workflow you wish to run. If a workflow was previously run on the project, you can select Clear AI data before running to have the Workflow re-run fields that have already had a task run. Once you have configured your Workflow, click Run.
3. Check the workflow status.
After running the workflow, you can check the status by hovering your cursor over the AI Workflow icon.
4. View status from the project list.
The status of a Workflow can also be seen on the project list.
5. Reload to see completed results.
Once the workflow is complete, a green check-mark will display on the workflow icon. Hovering over the icon will prompt you to reload the page to see the results.
Viewing Results
There are multiple ways to view AI results; below is the most common way, using the Photo Manager.
1. Open the Photo Manager.
Click the Photo Manager icon at the upper right of the project to open the Photo Manager.
2. Identify failed validations.
Photos with failed validations will display a red triangle.
3. Identify inconclusive results.
Photos with inconclusive results will display an orange circle. This means that the AI could not confidently determine if a validation passed or failed.
4. Open a photo to view its results.
Click on the photo to open it. The right-hand panel will display the Tasks' results.
You can view a field's instructions and example photo at the top of the sidebar. This can be useful when comparing results for an Image Correctness validation.
Results by Task
Image Analysis
Analyzes images to detect and tag objects and conditions, extract all text in bulk, and generate context-specific descriptions and captions. Required for other tasks in the workflow to execute.
Media Tags can be used to filter results in the Photo Manager; click the filter icon at the top of the image list and select the Media Tags you wish to filter by, then click OK.
Video Analysis
Analyzes videos to transcribe audio to text and summarize key moments and items from the video.
Text Extraction
Extracts text from images in a structured format and writes the results into specified fields.
Unlike Image Analysis, Text Extraction pulls specific pieces of text based on your configuration and can automatically populate other template fields. It can also extract a list of attributes (such as serial numbers), combine them into a comma-separated list, and automatically remove duplicates.
1. View the extracted text in the Photo Manager.
2. The extracted text may be configured to automatically populate template fields.
Image Validation
Validates that the image and its contents meet validation rules defined by a prompt or meets the instructions and example photos provided on the field.
See Resolving Validations below for how failed validations are resolved.
Multi-Field Validation
Performs validations across multiple fields using images and field values. The results will only appear on the primary field specified in the workflow configuration.
Image Correctness Validation
The image correctness validation takes the available context from the field, including:
- Name
- Instructions
- Example Photo
and compares this data to the attached photo, flagging a failure if the AI perceives that the photo is inconsistent with the field requirements.
If the validation is run on the entire collection of photos in one field, it will pass if any one of the photos meets validation requirements, even if another photo in the collection fails.
Image Quality Validation
Assesses an image for technical quality issues such as blurriness, over/under-exposure, or other distortions. Must be used in a workflow that also includes an Image Analysis task. No additional tokens are consumed beyond the Image Analysis task.
Image Location Validation
Verifies the geotagging and metadata of an image to confirm it was taken at the expected site or location. The project must be geotagged for this task to run.
Image Timestamp Validation
Verifies the timestamp of all project images to confirm that project images were taken after the project's creation and not uploaded from previous projects. This task verifies the capture date, not the upload date.
Viewing Results on Mobile
Failed validations can be visible to your mobile users; when they re-submit the rejected photos, the workflow can be re-run. If the new photos meet the validation requirements, the field will no longer be flagged.
In your Admin Settings under Mobile Projects, check the Show AI Results on Mobile box.
Once enabled, your users will have the ability to view AI results on their mobile device. See the following articles for instructions on viewing AI results on mobile:
iOS App: Viewing AI Image Validations
Android App: Viewing AI Validations
Resolving Validations
Admins can also manually resolve a failed or inconclusive validation while reviewing. To resolve, click the Edit icon next to the validation.
Check Resolve this validation, and enter a comment for additional clarification if desired. Then click Save.
The validation will now show Resolved, and the user who resolved the validation will be displayed along with any comments.
To un-resolve the validation, click the edit icon again and uncheck Resolve this validation. You can also edit the comment to include additional information.
To mark an inconclusive validation as failed, edit the validation, leave Resolve this validation unchecked, add a comment, and click Save.
Giving Feedback
If a task in your workflow isn't giving the expected result, it's important to provide feedback so that SiteCapture can help adjust your workflow.
To give feedback, open the task results in the Photo Manager, then click Give Feedback on AI Results.
Select the Task category, and describe what the task got wrong or what you would like it to do differently. Then click Save.
Best Practices
- Re-run with "Clear AI data before running" checked if you want fresh results on a project that already has AI data — for example, if past results looked off or SiteCapture has released an AI improvement since the workflow was last run. Re-running without this option only fills in fields that haven't had a task run yet; it won't refresh results that already exist.
- Turn on "Show AI Results on Mobile" if field users need to fix failed validations on-site. Without it, a field worker won't see a flagged photo until someone reviews it in the Photo Manager, which can mean a return trip instead of a retake before they leave the site.
FAQ
What's the difference between Image Analysis and Text Extraction?
Image Analysis detects and tags objects/conditions and generates general descriptions and captions. Text Extraction is more targeted: it pulls specific pieces of text you've configured (like a serial number) and can write that value directly into a template field, including combining multiple values into a de-duplicated, comma-separated list.
What does an orange circle mean in the Photo Manager?
It means the AI returned an inconclusive result; it couldn't confidently determine whether the validation passed or failed. A red triangle, by contrast, means the validation failed outright.
Can I undo a resolved validation?
Yes. Click the edit icon on the validation again and uncheck Resolve this validation. You can also update the comment at the same time.
Why isn't my Image Location Validation task running?
The project needs to be geotagged for this task to run; it verifies the image's geotag and metadata against the expected site location.
Comments