System of determining advertising incremental lift

A non-transitory computer-readable medium having stored thereon instructions to cause a computer to execute a method, the method comprising:

• receiving a bidstream, the bidstream including bids on available advertising inventory and a plurality of consumer characteristics associated with the bids;
• identifying, during a web session, personally identifiable information (PII) that matches a consumer characteristic from the plurality of consumer characteristics in the bidstream;
• linking a cookie associated with the web session to a universal identifier for a consumer that corresponds to the consumer characteristic in the bidstream;
• recording, during the web session, impression data for a plurality of webpages having at least one executable tag, each of the plurality of webpages requested by a browser application associated with the cookie;
• accessing, via a data cloud, user profiles of a plurality of consumers, the accessing including using the cookie to limit a respective user profile for each one of the plurality of consumers to a particular device;
• determining a target group in the bidstream, by matching, via the data cloud, a plurality of consumers to at least one successful bid in a campaign, each consumer in the plurality of consumers in the target group having an impression for a piece of content displayed associated with the at least one successful bid;
• retroactively creating a synthetic group in the bidstream by matching, via the data cloud, a plurality of consumers to at least one unsuccessful bid in the campaign, each consumer in the synthetic group being not identified to be included in a holdout group at a time of running the campaign and not being exposed with a piece of content associated with the at least one unsuccessful bid;
• determining, by a machine learning module implemented on a hardware processor, the plurality of consumer characteristics in the target group and the retroactively created synthetic group by:
• training a neural network machine learning model on past impression data, past consumer actions, and past consumer characteristics;
• determining, using the trained neural network machine learning model, statistically significant positive and negative correlations between actions performed by consumers in the target and the retroactively created synthetic groups to multiple random behavioral patterns;
• for each consumer in the target and retroactively created synthetic groups, evaluating the statistically significant positive and negative correlations to each random behavioral pattern using a machine learning decision tree model that determines behavioral patterns that are positive and negative for a desired event, the evaluating of the positive and negative correlations to each random behavioral pattern including determining a first behavioral pattern indicating consumers who would not perform the desired event regardless of ad exposure, a second behavioral pattern indicating consumers who would perform the desired event regardless of ad exposure, and a third behavioral pattern indicating consumers whose behavior would be influenced by ad exposure, the third behavioral pattern representing an advertising influence on the consumers for performing the desired event;
• and
• in response to determining the third behavioral pattern using the machine learning decision tree model, testing, using a regression model, the third behavioral pattern to determine an explanatory value of an independent variable that is representative of the third behavior pattern for predicting advertising influence;
• determining that a consumer characteristic corresponding to the universal identifier for the consumer in the bidstream is likely to lead to the desired event based on the explanatory value of the independent variable representative of the third behavior pattern;
• measuring an incremental impact of the at least one successful bid on the plurality of consumers in the retroactively created synthetic group based on the determining of the consumer characteristic;
• and
• dynamically adjusting transmission of content from available inventory based on the measured incremental impact.

A system, comprising:

• a processor, and a computer-readable medium having stored thereon instructions to cause the processor to execute a method, the method comprising:
• receiving a bidstream, the bidstream including bids on available advertising inventory and a plurality of consumer characteristics associated with the bids;
• identifying, during a web session, personally identifiable information (PII) that matches a consumer characteristic from the plurality of consumer characteristics in the bidstream;
• linking a cookie associated with the web session to a universal identifier for a consumer that corresponds to the consumer characteristic in the bidstream;
• recording, during the web session, impression data for a plurality of webpages having at least one executable tag, each of the plurality of webpages requested by a browser application associated with the cookie;
• accessing, via a data cloud, user profiles of a plurality of consumers, the accessing including using the cookie to limit a respective user profile for each one of the plurality of consumers to a particular device;
• determining a target group in the bidstream, by matching, via the data cloud, a plurality of consumers to at least one successful bid in a campaign, each consumer in the plurality of consumers in the target group having an impression for a piece of content displayed associated with the at least one successful bid;
• retroactively creating a synthetic group in the bidstream by matching, via the data cloud, a plurality of consumers to at least one unsuccessful bid in the campaign, each consumer in the synthetic group being not identified to be included in a holdout group at a time of running the campaign and not being exposed with a piece of content associated with the at least one unsuccessful bid;
• determining, by a machine learning module implemented on a hardware processor, the plurality of consumer characteristics in the target group and the retroactively created synthetic group by:
• training a neural network machine learning model on past impression data, past consumer actions, and past consumer characteristics;
• determining, using the trained neural network machine learning model, statistically significant positive and negative correlations between actions performed by consumers in the target and the retroactively created synthetic groups to multiple random behavioral patterns;
• for each consumer in the target and retroactively created synthetic groups, evaluating the statistically significant positive and negative correlations to each random behavioral pattern using a machine learning decision tree model that determines behavioral patterns that are positive and negative for a desired event, the evaluating of the positive and negative correlations to each random behavioral pattern including determining a first behavioral pattern indicating consumers who would not perform the desired event regardless of ad exposure, a second behavioral pattern indicating consumers who would perform the desired event regardless of ad exposure, and a third behavioral pattern indicating consumers whose behavior would be influenced by ad exposure, the third behavioral pattern representing an advertising influence on the consumers for performing the desired event;
• and
• in response to determining the third behavioral pattern using the machine learning decision tree model, testing, using a regression model, the third behavioral pattern to determine an explanatory value of an independent variable that is representative of the third behavior pattern for predicting advertising influence;
• determining that a consumer characteristic corresponding to the universal identifier for the consumer in the bidstream is likely to lead to the desired event based on the explanatory value of the independent variable representative of the third behavior pattern;
• measuring an incremental impact of the at least one successful bid on the plurality of consumers in the retroactively created synthetic group based on the determining of the consumer characteristic;
• and
• dynamically adjusting transmission of content from available inventory based on the measured incremental impact.