Step 1 - Examine the Recipe
The first step in troubleshooting your recipe is to take a closer look at how it was constructed. If the results don’t match your expectations, verify that the ingredients, exclusions, and boosters were added as intended. If the recipe isn’t returning any results, there are several things to check:
What ingredients were used to build the recipe? If the recipe was built with an ingredient requiring an “on page” anchor item, or point of reference, which includes Co-Buy, Co-Browse, SmartBundle, and Similar Items, you also need to enter a sample item name or ID as the anchor item for the recipe to be able to return results. As you enter different item names or IDs into the Viewing field, you’ll see that the recipe returns different results based what the visitor is viewing on the page at that moment.
Is the anchor item is returning results? If the recipe was built with an ingredient that requires any anchor item - in cart, on page, or last purchased - it’s possible that the recipe training didn’t generate the requisite data for that anchor item. For example, if you trained a Co-Buy ingredient, there may not be enough other items bought by the visitors that bought the current anchor item. If the ingredient in use requires an "in-cart" or "last purchased" anchor item, find a visitor that has an open cart or recently purchased an item. If the recipe doesn’t already include a backup ingredient, try adding either Trending or Collaborative Filtering as a fallback ingredient. Then, retrain the recipe, and preview it again to see if the recipe returns results.
Is there an Exclusion in the recipe? Exclusions always override inclusions, so If there is a conflict, the exclusion will always win. Check for conflicting inclusions and exclusions that could result in the recipe not returning any results.
Step 2 - Examine Unified Customer Profile
Once you’ve confirmed that the recipe was built correctly and the returned recommendations still don’t make sense, investigate the unified customer profile of the visitor you are using to preview recommendations returned by the recipe. It can be insightful to look at his or her affinities and data at a more granular level. You can review this information to help determine if the results actually make sense based on the individual visitor’s data.
Step 3 - Methodically Test the Recipe
If you’ve researched the Unified Customer Profile for different visitors and are still not satisfied with the returned results, begin methodically testing the recipe:
Remove all ingredients, exclusions, and boosters
Add one recipe component at a time
- Retrain the recipe, if prompted (required after removing or adding ingredients)
Preview results to uncover the component most responsible for the results you are seeing
This page has no comments.