Recommendation providers bill you mostly by the products bought from recommendations within 24 hours (these are counted and defined as sales and you have to pay commission only after these), or based on the number of recommendations provided in the current period.
Once the recommendation engine is running and the personalization is done, constant fine tuning follows.
The engine displays personalized and non personalized products: not all the Product Recommendations are personalized: well-trained engines are able to decide in milliseconds which of the 2 following possibilities to launch:
1. Non personalized recommendation calculated from aggregated customer behaviour (when we don’t know enough about the individual customer)
2. Personalized recommendation that display different items for any users based on their purchasing and browsing history.
Before you integrate with a recommendation system, don’t forget to nominate it in the list of data processors. In compliance with the GDPR you have to ensure your customers this way that their data and sensitive personal information collected is stored with the highest security level and cannot be accessed by unauthorized parties. Find more about securing your customers’ personal information in the unit detailing data security.
To take out the biggest ROI from your investments you or the specialists on the recommendation engine provider have to review the data flow to split those streams that doesn’t work and lift those that work to allocate your resources there to make the most of the personalization efforts. On the analytics dashboard you will see that sometimes the not personalized, sometimes personalized recommendations drive more conversions and the engine makes thousands of decisions in every minute:
1.If the customer has enough history he gets personalized offers ( that are predicted to imply higher conversions)
2. If the customer does not have enough history the algorithm will ignore his profile and applies general item-to-item recommendations,
This decision type is called fallback scenario, often used by recommendation engines. If the engine works behind a e-store with ten thousand of buyers and data-rich environment, plenty of non-personalized, personalized and mixed scenarios compete each other. Calculating fallback scenarios from user histories, similar users’ histories, product sales histories, AB tests results etc. might also need additional experiments from data scientists.
Recommendation engine analytics
Recommendation engines handle important metrics in a dedicated importance. These should be transparent for the client or the provider must enable you to access the reports anytime.
Choose a recommendation solution coming with user-friendly analytics dashboard
The revenue generated through recommendations is one of the most important metric when evaluating a recommender system.
GMV/1000 Recommendations – Take the total revenue driven by recommendations, and subdivide it to get the average revenue sent by 1000 recommendations.
% of Revenue Through Recommendations – Means the quotient of the revenue generated through recommendations / the total revenue. A cornerstone metrics.