Corvus AI

The Corvus AI is a service responsible for training the Contextual Bandit leveraged in Einstein Decisions. That service is made available through the Corvus gear, which allows the fetching of Contextual Bandit arms from within the context of a campaign template.

Import ContextualBanditConfig into any template and define a configuration, declaring the content zone, maximum number of promotions returned (up to 12), and explicit fallback promotions to use. Optionally, include a filter on which promotions are eligible for the bandit.

Import the decide function to be used to request contextual bandit promotions from within the template's run method in the server-side TypeScript. The decide function adheres to the following type signature.

The following is an example template that leverages ContextualBanditConfig.

To properly train the contextual bandit, it is essential to ensure click log generation. The Einstein Decisions global web template automatically tracks click logs and view logs in web campaigns. However, only view logs are tracked automatically in server-side Einstein Decisions campaigns, while clicks require manual tracking.

For instructions on ensuring clicks are correctly tracked, refer to the Einstein Decisions - Server-Side Template documentation.