Sourcing goods based on pre-feature analytics

1. A computer-implemented method for determining a quantity of goods to source for a promotion offering the goods for purchase, the method comprising:

adaptively training, by a processor of a pre-feature analysis module, a machine learning odds model based on historical sales velocity data until tuning criteria for the machine learning odds model is satisfied, wherein the historical sales velocity data corresponds to a historical number of sales per day over one or more promotion durations;

receiving, by the processor, pre-feature data describing a pre-feature impression of the promotion that was distributed to a selected subset of consumer devices associated with a selected subset of consumers based on at least one of attributes of the promotion and attributes of the consumers during a pre-feature duration in a limited distribution of the promotion to only the selected subset of consumers, pre-launch and prior to a full distribution of the promotion, wherein the promotion is indicative of a promotional value that upon purchase or acceptance results in issuance of an instrument configured to be used toward at least a portion of the purchase of the goods defined by the promotion;

wherein the promotion is optimized based on the attributes of the consumers and the attributes of the promotion by being selected based on at least one of type of goods being offered, promotion price, offered discount, gender of the consumers, device location of the consumer devices, click stream activity and past transaction data of the consumers;

wherein the pre-feature duration corresponds to a first period of time prior to a first day that the promotion is launched in the full distribution,

wherein the pre-feature impression of the promotion comprises a set of email impressions of the promotion that were distributed by a promotion and marketing service to the selected subset of consumer devices during the pre-feature duration in the limited distribution of the promotion prior to the full distribution of the promotion, in a custom configuration based on a ranking of the promotion, wherein the ranking includes varied graphical interface positioning specific to a user interface of a consumer device and for display on a display of the user interface, and the email impressions include a layout of a set of graphics with the graphic that is for display at position 1 in the layout being for display larger than the other graphics, at a top of the display and representing a featured deal;

wherein the pre-feature data comprises data collected from the set of email impressions,

wherein the pre-feature data further comprises responses from the selected subset of consumers to the pre-feature impression of the promotion, and

wherein the pre-feature data further comprises a number of pre-feature sales resulting from the pre-feature impression of the promotion, wherein the pre-feature sales comprise sales during both the limited distribution of the promotion and during the full distribution of the promotion which both result from the pre-feature impression emailed to the selected subset of consumers during the pre-feature duration in the limited distribution of the promotion;

generating, by the processor, an estimated sales velocity for the promotion by applying a portion of the pre-feature sales data associated with the selected subset of consumers to the machine learning odds model,

wherein the estimated sales velocity is a predicted number of sales per day over a promotion duration,

wherein the promotion duration corresponds to a second period of time that begins on the first day that the promotion is launched in the full distribution,

wherein the estimated sales velocity is based at least on the number of pre-feature sales and a velocity multiplier that is a function of a promotion price and the promotion duration, and

wherein the machine learning odds model comprises a scaling parameter that corrects for quantity of the pre-feature sales based on an email-only-attributed quantity of the pre-feature sales data and a total quantity of the goods being offered during the first period of time; and

determining, by the processor, the quantity of goods to source by the promotion and marketing service inventory based on the estimated sales velocity.

2. The method of claim 1 , comprising a demographic analysis of the selected subset of consumers to whom the pre-feature impression was distributed.

3. The method of claim 1 , wherein the pre-feature impression of the promotion is included in a set of pre-feature impressions that are distributed as content describing available promotions.

4. The method of claim 3 , wherein the content is email content.

5. The method of claim 1 , wherein the selected subset of consumers to which the pre-feature impression of the promotion is distributed during the pre-feature duration is a subset of a set of consumers to which the promotion is distributed during the promotion duration.

6. A computer program product, stored on a computer readable medium, comprising instructions that when executed on one or more computers cause a processor of a pre-feature analysis module to perform operations implementing determining a quantity of goods to source for a promotion offering the goods for purchase, the operations comprising:

adaptively training a machine learning odds model based on historical sales velocity data until tuning criteria for the machine learning odds model is satisfied, wherein the historical sales velocity data corresponds to a historical number of sales per day over one or more promotion durations;

receiving, from an impressions data repository, pre-feature data describing a pre-feature impression of the promotion that was distributed to a selected subset of consumer devices associated with a selected subset of consumers based on at least one of attributes of the promotion and attributes of the consumers during a pre-feature duration in a limited distribution of the promotion to only the selected subset of consumers, pre-launch and prior to a full distribution of the promotion, wherein the promotion is indicative of a promotional value that upon purchase or acceptance results in issuance of an instrument configured to be used toward at least a portion of the purchase of the goods defined by the promotion;

wherein the promotion is optimized based on the attributes of the consumers and the attributes of the promotion by being selected based on at least one of type of goods being offered, promotion price, offered discount, gender of the consumers, device location of the consumer devices, click stream activity and past transaction data of the consumers;

wherein the pre-feature duration corresponds to a first period of time prior to a first day that the promotion is launched in the full distribution,

wherein the pre-feature impression of the promotion comprises a set of email impressions of the promotion that were distributed by a promotion and marketing service to the selected subset of consumer devices during the pre-feature duration in the limited distribution of the promotion prior to the full distribution of the promotion, in a custom configuration based on a ranking of the promotion, wherein the ranking includes varied graphical interface positioning specific to a user interface of a consumer device and for display on a display of the user interface, and the email impressions include a layout of a set of graphics with the graphic that is for display at position 1 in the layout being for display larger than the other graphics, at a top of the display and representing a featured deal;

wherein the pre-feature data comprises data collected from the set of email impressions,

wherein the pre-feature data further comprises responses from the selected subset of consumers to the pre-feature impression of the promotion and

wherein the pre-feature data further comprises a number of pre-feature sales resulting from the pre-feature impression of the promotion, wherein the pre-feature sales comprise sales during both the limited distribution of the promotion and during the full distribution of the promotion which both result from the pre-feature impression emailed to the selected subset of consumers during the pre-feature duration in the limited distribution of the promotion;

generating an estimated sales velocity for the promotion by applying a portion of the pre-feature sales data associated with the selected subset of consumers to the machine learning odds model,

wherein the estimated sales velocity is a predicted number of sales per day over a promotion duration,

wherein the promotion duration corresponds to a second period of time that begins on the first day that the promotion is launched in the full distribution,

wherein the estimated sales velocity is based at least on the number of pre-feature sales and a velocity multiplier that is a function of a promotion price and the promotion duration, and

wherein the machine learning odds model comprises a scaling parameter that corrects for quantity of the pre-feature sales based on an email-only-attributed quantity of the pre-feature sales data and a total quantity of the goods being offered during the first period of time; and

determining the quantity of goods to source by the promotion and marketing service inventory based on the estimated sales velocity.

7. The computer program product of claim 6 , wherein the operations further comprise a demographic analysis of the selected subset of consumers to whom the pre-feature impression was distributed.

8. The computer program product of claim 6 , wherein the pre-feature impression of the promotion is included in a set of pre-feature impressions that are distributed as content describing available promotions.

9. The computer program product of claim 8 , wherein the content is email content.

10. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause a processor of a pre-feature analysis module to perform operations implementing determining a quantity of goods to source for a promotion offering the goods for purchase, the operations comprising:

adaptively training a machine learning odds model based on historical sales velocity data until tuning criteria for the machine learning odds model is satisfied, wherein the historical sales velocity data corresponds to a historical number of sales per day over one or more promotion durations;

receiving, from an impressions data repository, pre-feature data describing a pre-feature impression of the promotion that was distributed to a selected subset of consumer devices associated with a selected subset of consumers based on at least one of attributes of the promotion and attributes of the consumers during a pre-feature duration in a limited distribution of the promotion to only the selected subset of consumers, pre-launch and prior to a full distribution of the promotion, wherein the promotion is indicative of a promotional value that upon purchase or acceptance results in issuance of an instrument configured to be used toward at least a portion of the purchase of the goods defined by the promotion;

wherein the promotion is optimized based on the attributes of the consumers and the attributes of the promotion by being selected based on at least one of type of goods being offered, promotion price, offered discount, gender of the consumers, device location of the consumer devices, click stream activity and past transaction data of the consumers;

wherein the pre-feature duration corresponds to a first period of time prior to a first day that the promotion is launched in the full distribution,

wherein the pre-feature impression of the promotion comprises a set of email impressions of the promotion that were distributed by a promotion and marketing service to the selected subset of consumer devices during the pre-feature duration in the limited distribution of the promotion prior to the full distribution of the promotion, in a custom configuration based on a ranking of the promotion, wherein the ranking includes varied graphical interface positioning specific to a user interface of a consumer device and for display on a display of the user interface, and the email impressions include a layout of a set of graphics with the graphic that is for display at position 1 in the layout being for display larger than the other graphics, at a top of the display and representing a featured deal;

wherein the pre-feature data comprises data collected from the set of email impressions,

wherein the pre-feature data further comprises responses from the selected subset of consumers to the pre-feature impression of the promotion and

wherein the pre-feature data further comprises a number of pre-feature sales resulting from the pre-feature impression of the promotion, wherein the pre-feature sales comprise sales during both the limited distribution of the promotion and during the full distribution of the promotion which both result from the pre-feature impression emailed to the selected subset of the consumers during the pre-feature duration in the limited distribution of the promotion;

generating an estimated sales velocity for the promotion by applying a portion of the pre-feature sales data associated with the selected subset of consumers to the machine learning odds model,

wherein the estimated sales velocity is a predicted number of sales per day over a promotion duration,

wherein the promotion duration corresponds to a second period of time that begins on the first day that the promotion is launched in the full distribution,

wherein the estimated sales velocity is based at least on the number of pre-feature sales and a velocity multiplier that is a function of a promotion price and the promotion duration, and

wherein the machine learning odds model comprises a scaling parameter that corrects for quantity of the pre-feature sales based on an email-only-attributed quantity of the pre-feature sales data and a total quantity of the goods being offered during the first period of time; and

determining the quantity of goods to source by the promotion and marketing service inventory based on the estimated sales velocity.

11. The system of claim 10 , wherein the operations further comprise a demographic analysis of the selected subset of consumers to whom the pre-feature impression was distributed.

12. The system of claim 10 , wherein the pre-feature impression of the promotion is included in a set of pre-feature impressions that are distributed as content describing available promotions.

13. The system of claim 12 , wherein the content is email content.