This article explains how AI Decisioning uses customer data during model training and inference, including where data is stored, how it is processed, and how customer privacy is maintained.
Data storage
The customer data used for AI Decisioning model training and inference remains in your data warehouse.
Hightouch also stores a temporary cache of customer data in the storage bucket configured in your Hightouch workspace for processing purposes. This bucket is owned by your workspace and follows the same access controls and encryption settings as other Hightouch-managed storage.
The data used by AI Decisioning may include:
- User features
- Outcome data (for example, purchases or clicks)
- Recommendation data
For more information on how Hightouch secures customer data, see Hightouch’s security documentation →.
User features
User features are the user attributes used during the model training process.
Eligible user features are explicitly defined and whitelisted in your Hightouch schema. These may include:
- Demographic attributes
- Marketing engagement data (email clicks, SMS opens, etc.)
- Product interaction data
- User behavior data
User features do not include personally identifiable information (PII) such as names, email addresses, or phone numbers.
How AI Decisioning uses your data
Data usage during model training and inference
User features and outcome datasets are used in-flight during the process of generating recommendations.
At a high level, the process works as follows:
- User features and outcomes are pulled temporarily into our execution environment
- The data is used to train a model specific to your workspace data
- The model is used to generate recommendations for eligible users
- The model and user data are not retained after inference completes in the Hightouch environment
Data sharing and privacy guarantees
All machine learning models produced by Hightouch for AI Decisioning are single-tenant, scoped to your workspace, and are never shared with other customers.
They are used for the sole purpose of generating recommendations and are not retained after the inference process.
Customer data is also never shared with other customers for training or analytics purposes.