ADVERTISING CANNIBALIZATION MANAGEMENT

1 . A system comprising:

a data module to access historical data that comprises advertisement revenue, advertisement parameters, and a cannibalization metric that is indicative of sales loss associated with an advertisement presentation;

an analysis module, implemented by at least one hardware processor of a machine, to determine a value for at least one of the advertisement parameters by an analysis of the historical data, use of the determined value causes a desired advertisement revenue with respect to a bounded cannibalization metric; and

a presentation module to cause presentation, in real time, of an advertisement on a user interface of a client device using the determined value.

2 . The system of claim 1 , wherein the analysis module is further to:

identify a cannibalization covariate from among candidate covariates that include the advertisement parameters;

generate a covariate model that models the cannibalization covariate with respect to the advertisement parameters;

generate a cannibalization model that models the cannibalization metric with respect to the cannibalization covariate;

generate a revenue model that models the advertisement revenue with respect to the advertisement parameters; and

determine the value for at least one of the advertisement parameters using the revenue model in conjunction with the cannibalization model and the covariate model.

3 . The system of claim 1 , wherein the cannibalization metric comprises at least one of a purchase per user per week metric (PPW) and a gross revenue per user per week metric (GPW).

4 . The system of claim 1 , wherein the advertisement parameters include at least one of an advertisement placement, impressions, clicks, an advertiser, and an advertisement type.

5 . A method comprising:

accessing historical data comprising advertisement revenue, advertisement parameters, and a cannibalization metric that is indicative of sales loss associated with an advertisement presentation;

determining a value for at least one of the advertisement parameters by analyzing the historical data, use of the determined value causes a desired advertisement revenue with respect to a bounded cannibalization metric; and

causing presentation, in real time, of an advertisement on a user interface of a client device using the determined value.

6 . The method of claim 5 , wherein the analyzing the historical data further comprises:

identifying a cannibalization covariate from among candidate covariates that include the advertisement parameters;

generating a covariate model that models the cannibalization covariate with respect to the advertisement parameters;

generating a cannibalization model that models the cannibalization metric with respect to the cannibalization covariate;

generating a revenue model that models the advertisement revenue with respect to the advertisement parameters; and

determining the value for at least one of the advertisement parameters using the revenue model in conjunction with the cannibalization model and the covariate model.

7 . The method of claim 6 , wherein the identifying the cannibalization covariate further comprises:

measuring a cannibalization value by comparing cannibalization of a control group of users shown advertisements and a treatment group of users not shown advertisements; and

identifying the cannibalization covariate from among the candidate covariates according to a correlation between respective candidate covariates and the measured cannibalization value, the cannibalization covariate being a highest correlated covariate among the candidate covariates.

8 . The method of claim 7 , wherein the candidate covariates include at least one of clicks, impressions, and page views.

9 . The method of claim 5 , wherein the cannibalization metric comprises at least one of a purchase per user per week metric (PPW) and a gross revenue per user per week metric (GPW).

10 . The method of claim 5 , wherein the advertisement parameters include at least one of an advertisement placement, impressions, clicks, an advertiser, and an advertisement type.

11 . The method of claim 5 , further comprising:

determining a lower limit for the bounded cannibalization metric according to a minimum advertisement revenue specified by an operator; and

determining an upper limit for the bounded cannibalization metric according to a maximum cannibalization cost specified by the operator.

12 . The method of claim 5 , further comprising:

accessing current data comprising the advertisement revenue, the advertisement parameters, and the cannibalization metric for a time period of a duration;

determining a change amount for at least one of the advertisement parameters by analyzing the historical data in conjunction with the current data; and

causing presentation, in real time, of the advertisement on the user interface of the client device using the determined change amount for another time period of the duration.

13 . A machine-readable medium having no transitory signals and storing instructions that, when executed by at least one processor of a machine, cause the machine to perform operations comprising:

accessing historical data comprising advertisement revenue, advertisement parameters, and a cannibalization metric that is indicative of sales loss associated with an advertisement presentation;

determining a value for at least one of the advertisement parameters by analyzing the historical data, use of the determined value causes a desired advertisement revenue with respect to a bounded cannibalization metric; and

causing presentation, in real time, of an advertisement on a user interface of a client device using the determined value.

14 . The machine-readable medium of claim 13 , wherein the analyzing the historical data further comprises:

identifying a cannibalization covariate from among the candidate covariates that include the advertisement parameters;

generating a covariate model that models the cannibalization covariate with respect to the advertisement parameters;

generating a cannibalization model that models the cannibalization metric with respect to the cannibalization covariate;

generating a revenue model that models the advertisement revenue with respect to the advertisement parameters; and

determining the value for at least one of the advertisement parameters using the revenue model in conjunction with the cannibalization model and the covariate model.

15 . The machine-readable medium of claim 14 , wherein the operations further comprise the identifying the cannibalization covariate by:

measuring a cannibalization value by comparing cannibalization of a control group of users shown advertisements and a treatment group of users not shown advertisements; and

identifying the cannibalization covariate from among the candidate covariates according to a correlation between respective candidate covariates and the measured cannibalization value, the cannibalization covariate being a highest correlated covariate among the candidate covariates.

16 . The machine-readable medium of claim 15 , wherein the candidate covariates include at least one of clicks, impressions, and page views.

17 . The machine-readable medium of claim 13 , wherein the cannibalization metric comprises at least one of a purchase per user per week metric (PPW) and a gross revenue per user per week metric (GPW).

18 . The machine-readable medium of claim 13 , wherein the advertisement parameters include at least one of an advertisement placement, impressions, clicks, an advertiser, and an advertisement type.

19 . The machine-readable medium of claim 13 , wherein the operations further comprise:

determining a lower limit for the bounded cannibalization metric according to a minimum advertisement revenue specified by an operator; and

determining an upper limit for the bounded cannibalization metric according to a maximum cannibalization cost specified by the operator.

20 . The machine-readable medium of claim 13 , further comprising:

accessing current data comprising the advertisement revenue, the advertisement parameters, and the cannibalization metric for a time period of a duration;

determining a change amount for at least one of the advertisement parameters by analyzing the historical data in conjunction with the current data; and

causing presentation, in real time, of the advertisement on the user interface of the client device using the determined change amount for another time period of the duration.