HI.

We’d love to learn about your predictive questions and show you how Endor can help.

Thank you!

Your submission has been received

Oops! Something went wrong while submitting the form.
< USER GUIDE

(protocol) How to ask a predictive question?

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.

In simple terms, asking predictive questions means learning patterns from past behaviors, and applying them to new data. As this is the case, our process requires two types of data sources:

  1. The list you want to rank – Who is most likely to take a loan? Buy you product? Require your service?. This list is called the “LIST TO RANK” and it includes all the entities you would like to rank according to your predictive question.
  2. A list of entities that have shown the desired behavior in the past – People who have taken loans, purchased your products or hired your services. This is the “EXAMPLE” list.

The Endor Protocol identifies common qualities among the entities in the example list, and using behavioral clusters based on social physics, it applies this logic to the LIST TO RANK entities. In doing so, the Protocol finds behavioral resemblance (look-alikes) in the target list and ranks the entities according to the probability of them acting in the same manner as the entities in the example list.

By controlling the characteristics of the example list, you can fine-tune the prediction results. While a smaller example list creates a more specific resemblance group and a very clear entity type (the minimal example group is 100 entities); a larger example list increases your potential reach but reduces the level of similarity between the entities you rank and the example list. We generally recommend an example list of between 1,000 to 10,000 entities.


PLEASE NOTE: EXAMPLE list and LIST TO RANK should be approximately the same size.

Let’s dive in a bit deeper to better understand and create the lists.


  1. List to rank - The list should contain the target population group which you are interested to rank.
  2. Example - The list should contain the people with the known desired behavior, it will be used for the look alike ranking.

Defining a LIST TO RANK:

In order to explain the structure of the LIST TO RANK file in the simplest way possible, we’ll share an example. If a bank, for example, wanted to predict which customers are most likely to take a loan, it would create a list of all its customers with an active credit account, who have an available credit of at least $50, and are left with at least 25% of their available credit, who haven’t taken a loan in the last 90 days.


‍‍This is what your query definition will look like:


Important

Following is an example of the correct date format:

Defining an EXAMPLE list:

The example list would then include all of the bank’s customers who have taken a loan in the last 30 days but did not take one in the 90 days prior.

‍‍‍This is what your query definition will look like:


Notice that the same query definitions apply in both union sections of the query. You run the same logic on both points in time and gather the information into a single list for the protocol to use when running your predictions.

Important Guidelines:
  • CSV files should include two columns of ‘ID’ and ‘Date’.
  • Example file should contain at least 100 records and no more than 10,000 records.
  • LIST TO RANK file should contain at least 300 records and no more than 500,000 records.
  • The LIST TO RANK file should be up to two orders of magnitude larger than the example file.
  • Example and LIST TO RANK files should not be larger than 10 MB.

More related content

Fast, Accurate & Scalable
business predictions

Contact US

Social

© Endor Software Ltd All Right Reserved
Terms & conditions | Privacy policy