Predictive Metadata automatically enriches your assets using AI, applying values based on your custom taxonomy. The result is structured, consistent metadata that powers fast discovery, seamless collaboration, and strong governance across your DAM.
Instead of time-consuming, inconsistent manual tagging, you get intelligent classification at scale, aligned to your brand and not generic labels.
Your teams benefit from:
Consistent, accurate classification
Faster search and retrieval
Stronger governance and workflows
Increased asset reuse
Clearer analytics and insights
More than auto-tagging, Predictive Metadata turns metadata into a scalable, strategic advantage.
Predictive Metadata is available to everyone using Automations. To learn how to set up and use automations please see the help article.
Please note: Automations is currently in Beta. It’s best suited for scoped, well-defined workflows rather than mission-critical, high-volume batch processing. We’re actively improving scalability and visibility.
How to install it
Navigate to the Apps section in the Marketplace.
Search for Predictive Metadata.
Click Install app.
Once installed, the functionality becomes available within Automations.
Step 1: Prepare your taxonomy
Predictive Metadata works as an Automation action.
Go to a Library.
Open Metadata Management.
Create or review the metadata properties you want to predict.
Make sure your taxonomy reflects the structure you want the AI to follow.
Step 2: Create a new Automation
Go to Automations.
Click Create Automation.
Add a Trigger, for example:
Asset created
Workflow status changed
Add the action Predict Metadata.
Step 3: Configure Predict Metadata
Inside the action setup:
Add Global Hints
These act as a system prompt that applies to all metadata predictions in this automation. Use this to provide context about your brand, naming conventions, or classification logic.
Select fields to predict
Choose the metadata properties (multi-select) you want the AI to populate.
Add follow-up actions (optional) For example:
Change workflow status to Manual review after prediction
Notify a responsible team (ensuring human validation where needed).
Fine-tuning predictions
You can improve prediction quality in two ways:
Global fine-tuning
Adjust your Global Hints in the Automation settings to provide clearer instructions.
Per-property fine-tuning
In Metadata Management, use the Help text field of each metadata property to guide the AI more precisely.
For example:
Define what qualifies as “Luxury” vs. “Premium”
Explain how verticals should be assigned
Clarify internal naming conventions
Provide as much relevant context as possible to improve prediction quality. Consider fine-tuning only if results are still insufficient after optimizing the input context.
How to run predictions
Predictions are triggered automatically based on your chosen trigger. For example:
Upload a new asset = metadata is predicted
Change workflow status = metadata is predicted
No manual action is required once the automation is active.
What works best
Predictions perform especially well when metadata can be derived from:
Image content
Campaign (e.g., title visible in image)
Theme (Luxury, Technology, Heritage, etc.)
Color variant
Mood
Subject (Cars, Products, People, etc.)
Car model and type (e.g., Range Rover Defender, SUV)
Asset context
Folder name
File name patterns
General model knowledge
Target audience (Families, Tech Enthusiasts, Luxury Buyers, etc.)
What may require more fine-tuning
Predictions are more challenging for highly internal or abstract taxonomies, such as:
Responsible internal team or group
Vertical / Sub-vertical
Internal project codes
These typically require strong global hints and detailed help text to improve accuracy.
Practical usage example
Predict car models and types
An editor uploads images of vehicles to the library.
An automation triggers when the asset is created.
The Predict Metadata action:
Identifies the car model (e.g., Range Rover Defender)
Classifies the vehicle type (e.g., SUV)
Assigns Theme, Mood, and Color Variant
The workflow then moves the asset to Manual review for validation.
Result: structured, searchable, and governance-ready assets - automatically!
Limitations
Rate limits may apply (Azure-based infrastructure).
Some internal taxonomies require significant fine-tuning for high accuracy.