Deploying recommendations

This is for:

Developer

Product Recommendations (PR) models require a substantial amount of data for training to deliver high-quality recommendations. If a PR model goes live without sufficient training, recommendations may either be missing or of poor quality until enough data has been ingested by the model. Therefore, we recommend gathering data as soon as possible so that PR models can provide relevant recommendations as soon as they’re implemented in your storefront.

The following events are essential for training PR models effectively:

Sending click events isn’t necessary for PR model training. Additionally, clicks require the use of a responseID, which is returned by querying the Commerce API, making them less practical for pre-go-live scenarios.

We strongly recommend implementing event tracking before your storefront goes live with product recommendations. This approach allows PR models to train on a robust dataset, ensuring high-quality recommendations from day one. For the best results:

  1. Start tracking product views, cart events, and purchase events a few weeks before your Coveo implementation is deployed. The way to track events depends on your chosen approach for building product discovery interfaces.

  2. Deploy all of your product discovery solutions: search, listing pages, and recommendations interfaces simultaneously after event tracking has collected sufficient data.

Alternative approach

If pre-go-live event tracking isn’t feasible:

  1. Deploy search interfaces and listing pages. Additionally, ensure event tracking is implemented.

  2. Deploy recommendations after a few days or weeks of data collection, depending on the volume of events, to allow ML models to train on the necessary events.