Systems and methods for segmenting customer targets and predicting conversion
Methods, systems and apparatuses, including computer programs encoded on computer storage media, are provided for generating prediction models related to targeting and acquiring customers. Thousands of variables of historical data, including data for prospects and external data, are used to train the prediction models. The variables are pre-processed, then sensitivity analysis is performed on the input variables with respect to the target. The variables with the most influence on the target are selected and added to the feature set used for training a prediction model.
1 . A method for creating a set of look-a-like prospects, the method comprising:
receiving historical data regarding prospects, wherein the historical data comprises one or more input variables and a target, wherein the target comprises a conversion status of each prospect;
selecting features from the historical data to create a feature set, wherein the feature selection comprises:
performing correlation shape analysis on each input variable with respect to the target to generate a set of highly correlated input variables;
training at least one machine learning model based on the feature set;
identifying segments in a predefined range of conversion percentage based on the trained at least one machine learning model;
determining a most suitable model from the at least one machine learning model; and
generating a set of look-a-like prospects based on the determined model.
2 . The method of claim 1 , wherein the prospects comprise quote-generating prospects.
3 . The method of claim 1 , wherein the correlation shape analysis comprises a Pearson chi-square or phi_k algorithm.
4 . The method of claim 1 , wherein the determined machine learning model comprises a decision tree algorithm.
5 . The method of claim 1 , wherein the determined machine learning model comprises an association rule mining algorithm.
6 . The method of claim 1 , wherein the most suitable machine learning model is determined based on one or more of the following: the number of segments identified, the percentage of the historical conversion prospects included in the identified segments, and the number of features from the feature set that were used to generate the identified segments.
7 . The method of claim 1 , wherein the prospect data comprises geo-demographic data or geographical risk assessment data.
8 . A method for creating a conversion prediction model for predicting an aspect of conversion of a prospect, the method comprising:
receiving historical data regarding prospects, wherein the historical data comprises one or more input variables and a target;
feature engineering the historical data to create a feature set, wherein the feature engineering comprises:
determine the relative importance of the input variables with respect to the target;
selecting the variables of most importance and adding them to the feature set; and
performing dimension reduction on the non-selected variables to create extracted features and adding the extracted features to the feature set; and
creating the conversion prediction model based on the feature set.
9 . The method of claim 8 , wherein the conversion prediction model comprises a tree-based algorithm.
10 . The method of claim 8 , wherein the conversion prediction model comprises a neural network algorithm.
11 . The method of claim 8 , wherein the step of determining the importance of the input variables comprises generating at least one preliminary model and identifying the variables with the most influence on the target.
12 . The method of claim 11 , wherein the at least one preliminary model is a gradient boosted decision tree.
13 . The method of claim 8 , wherein the step of determining the importance of the input variables comprises Shapley values for each of the input variables.
14 . The method of claim 8 , wherein the dimension reduction is performed using an autoencoder.
15 . The method of claim 8 , wherein the conversion prediction model comprises a conversion iterations model, and wherein the historical data further comprises quote or inquiry data.
16 . The method of claim 8 , wherein the conversion prediction model comprises a conversion cost model, and wherein the historical data further comprises quote or inquiry data.
17 . The method of claim 8 , wherein the conversion prediction model comprises a conversion time model, and wherein the historical data further comprises quote or inquiry data.
18 . A method for creating a conversion prediction model for predicting an aspect of conversion for a set of look-a-like prospects, the method comprising:
receiving historical data regarding prospects, wherein the historical data comprises one or more input variables and a first target, wherein the first target comprises a conversion status of each prospect;
selecting features from the historical data to create a first feature set, wherein the feature selection comprises:
performing correlation shape analysis on each input variable with respect to the target to generate a set of highly correlated input variables;
training at least one machine learning model based on the first feature set;
identifying segments in a range of conversion percentage based on the trained at least one machine learning model;
determining a most suitable model from the at least one machine learning model;
generating a look-a-like segment based on the determined model;
feature engineering the historical data for the look-a-like segment to create a second feature set, wherein the feature engineering comprises:
determining the relative importance of the input variables in the historical data for the look-a-like segment with respect a second target;
selecting the variables of most importance and adding them to the second feature set; and
performing dimension reduction on the non-selected variables to create extracted features and adding the extracted features to the second feature set; and
creating the conversion prediction model based on the second feature set.
19 . The method of claim 18 , wherein the conversion prediction model comprises a conversion cost model, and wherein the historical data further comprises quote or inquiry data.
20 . The method of claim 18 , wherein the conversion prediction model comprises a conversion time model, and wherein the historical data further comprises quote or inquiry data.