When you create a predictive question, you choose a “List to rank” audience that you would like to rank and an Example audience, who showed the known desired behavior in the past.
When you create a predictive question, you choose a “List to rank” audience that you would like to rank and an Example audience,who showed the known desired behavior in the past. We identify the common qualities of the people in the example and with the use of behavioral clusters (which are based on the social physics, see how behavioral cluster are created) we apply it to rank the “List to rank” audience according to their behavior resemblance (or "look like") to the provided example. Ranking the “List to rank” audience according to the probability of them to act in the same manner in the future like the provided example audience.
By controlling the characteristics of the example audience you can tune the final outcome. Smaller audience group more closely match your example audience, and specific as to what you would like to find more of. The minimal group of examples should be 100.
Creating a larger example audience increases your potential reach, but reduces the level of similarity between the List to rank audience and example audience. We generally recommend an example audience with between 1,000 to 10,000 people. Example group quality matters.
Calibration and prediction sets should be approximately of the same size.
The system requires 2 sets of files:
All the customers with a credit account in active status, who have at least $50 of available credit and at least 25% of available credit, who did not purchase product A in the last 90 days.
All the customers who purchased product A in the last 30 days, but did not purchase product A in the 90 days before that.
Notice that the same query definitions apply in the 2 union sections of the query. You run the same logic on 2 points in time and gather it to a single list that will be utilized in your interaction with Endor.