Performing interactive updates to a precalculated cross-channel predictive model
A computer-implemented method, simulation and prediction system, and computer program product for advertising portfolio management. Embodiments commence upon receiving data comprising a plurality of marketing stimulations and respective measured responses, both pertaining to a first time period. A computer is used to form a multi-channel simulation model, where the simulation model accepts the marketing stimulations then outputs simulated responses. The simulation model is used for determining cross-channel weights to apply to the respective measured responses pertaining to the first time period. The simulation model is updated to reflect updated marketing stimulations pertaining to a second time period. The updated marketing stimulations overwrite some of the plurality of marketing stimulations captured in the first time period. The updated simulation model is used in calculating an effectiveness value of a particular one of the updated marketing stimulations based at least in part on the cross-channel weights determined for the first time period.
1. A computer-implemented method comprising:
training, with machine-learning techniques, a learning model by correlating, for a plurality of stimulation vectors corresponding to a plurality of marketing channels of an advertising campaign, a first portion of time series stimulus data with a first portion of time series response data across the plurality of marketing channels of the advertising campaign, the first portion of time series response data measuring responses to the first portion of time series stimulus data by a plurality of users exposed to marketing media represented in the first portion of time series stimulus data, the correlating including exogeneous variables that attenuate effects independent from the first portion of time series stimulus data, the correlating to emphasize more recent entries over older entries in the first portion of time series stimulus data and the first portion of time series response data;
forming, by executing a plurality of simulations with the learning model, a weight matrix to be utilized by a first simulated model, the weight matrix including weights corresponding to effects in a first portion of the plurality of marketing channels of the advertising campaign caused by stimulus in a second portion of the plurality of marketing channels of the advertising campaign, the plurality of simulations including:
performing simulations at values through a sweep of a range of potential stimulus values for the first portion of time series stimulus data, the range of potential stimulus values associated with an increase or decrease in marketing media spend relative to a baseline value;
predicting, based on the simulations at the values through the sweep of the range of potential stimulus values, a plurality of responses across the plurality of marketing channels of the advertising campaign; and
generating, for the simulations at the values through the sweep of the range of potential stimulus values, a plurality of weight values representative of the effect of the potential stimulus values on the plurality of responses;
collecting, via an interactive user interface, a second portion of time series stimulus data and a second portion of time series response data for the plurality of marketing channels of the advertising campaign;
reducing computational resource consumption associated with generating an updated version of at least one of the learning model or the first simulated model by applying the second portion of time series stimulus data and the second portion of time series response data to the first simulated model to generate a second simulated model; and
executing, with the second simulated model, a simulation on a user provided scenario of the advertising campaign, the user provided scenario entered using the interactive user interface, the user provided scenario entered by configuring at least one change in a marketing media spend value of a first marketing channel of the plurality of marketing channels of the advertising campaign, the simulation to determine an effectiveness value to attribute to the at least one change in the marketing media spend value relative to other marketing media spend values represented in stimuli values for other marketing channels of the plurality of marketing channels of the advertising campaign, the effectiveness value determined based on the weight matrix and representative of an effect of the first marketing channel on the other marketing channels of the advertising campaign.
2. The computer-implemented method as set forth in claim 1 , wherein the effectiveness value is displayed as a percentage.
3. The computer-implemented method as set forth in claim 1 , wherein at least one of the first portion of time series stimulus data or the second portion of time series stimulus data includes at least one of, an advertising spend, a number of direct mail pieces, a number of TV spots, a number of radio spots, a number of web impressions and a number of coupons printed, and at least one of the first portion of time series response data or the second portion of time series response data includes at least one of a number of calls into a call center, a number of clicks on an impression, or a number of coupon redemptions.
4. The computer-implemented method as set forth in claim 1 , further including generating a report based on the effectiveness value that quantifies a return on investment for at least one of the plurality of marketing channels that accounts for cross-channel effects.
5. A computer program product embodied in a non-transitory computer readable medium, the non-transitory computer readable medium having stored thereon a sequence of instructions which, when executed by a processor causes the processor to execute a process, the process comprising:
training, with machine-learning techniques, a learning model by correlating, for a plurality of stimulation vectors corresponding to a plurality of marketing channels of an advertising campaign, a first portion of time series stimulus data with a first portion of time series response data across the plurality of marketing channels of the advertising campaign, the first portion of time series response data measuring responses to the first portion of time series stimulus data by a plurality of users exposed to marketing media represented in the first portion of time series stimulus data, the correlating including exogeneous variables that attenuate effects independent from the first portion of time series stimulus data, the correlating to emphasize more recent entries over older entries in the first portion of time series stimulus data and the first portion of time series response data;
forming, by executing a plurality of simulations with the learning model, a weight matrix to be utilized by a first simulated model, the weight matrix including weights corresponding to effects in a first portion of the plurality of marketing channels of the advertising campaign caused by stimulus in a second portion of the plurality of marketing channels of the advertising campaign, the plurality of simulations including:
performing simulations at values through a sweep of a range of potential stimulus values for the first portion of time series stimulus data, the range of potential stimulus values associated with an increase or decrease in marketing media spend relative to a baseline value;
predicting, based on the simulations at the values through the sweep of the range of potential stimulus values, a plurality of responses across the plurality of marketing channels of the advertising campaign; and
generating, for the simulations at the values through the sweep of the range of potential stimulus values, weight values representative of the effect of the potential stimulus values on the plurality of responses;
collecting, via an interactive interface, a second portion of time series stimulus data and a second portion of time series response data for the plurality of marketing channels of the advertising campaign;
reducing computational resource consumption associated with generating an updated version of at least one of the learning model or the first simulated model by applying the second portion of time series stimulus data and the second portion of time series response data to the first simulated model to generated a second simulated model; and
executing, with the second simulated model, a simulation on a user provided scenario of the advertising campaign, the user provided scenario entered using the interactive interface, the user provided scenario entered by configuring at least one change in a marketing media spend value of a first marketing channel of the plurality of marketing channels of the advertising campaign, the simulation to determine an effectiveness value to attribute to the at least one change in the marketing media spend value relative to other marketing spend values represented in stimuli values for other marketing channels of the plurality of marketing channels of the advertising campaign, the effectiveness value determined based on the weight matrix and representative of an effect of the first marketing channel on the other marketing channels of the advertising campaign.
6. The computer program product as set forth in claim 5 , wherein the effectiveness value is displayed as a percentage.
7. The computer program product as set forth in claim 5 , wherein at least one of the first portion of time series stimulus data or the second portion of time series stimulus data includes at least one of, an advertising spend, a number of direct mail pieces, a number of TV spots, a number of radio spots, a number of web impressions and a number of coupons printed, and at least one of the first portion of time series response data or the second portion of time series response data includes at least one of a number of calls into a call center, a number of clicks on an impression, or a number of coupon redemptions.
8. The computer program product as set forth in claim 5 , wherein the process further includes generating a report based on the effectiveness value that quantifies a return on investment for at least one of the plurality of marketing channels that accounts for cross-channel effects.
9. A computer system comprising:
a computer processor to execute a set of program code instructions; and
a memory to store the set of program code instructions, in which the set of program code instructions comprises program code to perform:
training, with machine-learning techniques, a learning model by correlating, for a plurality of stimulation vectors corresponding to a plurality of marketing channels of an advertising campaign, a first portion of time series stimulus data with a first portion of time series response data across the plurality of marketing channels of the advertising campaign, the first portion of time series response data measuring responses to the first portion of time series stimulus data by a plurality of users exposed to marketing media represented in the first portion of time series stimulus data, the correlating including exogeneous variables that attenuate effects independent from the first portion of time series stimulus data, the correlating to emphasize more recent entries over older entries in the first portion of time series stimulus data and the first portion of time series response data;
forming, by executing a plurality of simulation with the learning model, a weight matrix to be utilized by a first simulated model, the weight matrix including weights corresponding to effects in a first portion of the plurality of marketing channels of the advertising campaign caused by stimulus in a second portion of the plurality of marketing channels of the advertising campaign, the plurality of simulations including:
performing simulations at values through a sweep of a range of potential stimulus values for the first portion of time series stimulus data, the range of potential stimulus values associated with an increase or decrease in marketing media spend relative to a baseline value;
predicting, based on the simulations at the values through the sweep of the range of potential stimulus values, a plurality of responses across the plurality of marketing channels of the advertising campaign;
generating, for the simulations at the values through the sweep of the range of potential stimulus values, weight values representative of the effect of the potential stimulus values on the plurality of responses;
collecting, via an interactive user interface, a second portion of time series stimulus data and a second portion of time series response data for the plurality of marketing channels of the advertising campaign;
reducing computational resource consumption associated with generating an updated version of at least one of the learning model or the first simulated model by applying the second portion of time series stimulus data and the second portion of time series response data to the first simulated model to generate a second simulated model; and
executing, with the second simulated model, a simulation on a user provided scenario of the advertising campaign, the user provided scenario entered using the interactive user interface, the user provided scenario entered by configuring at least one change in a marketing media spend value of a first marketing channel of the plurality of marketing channels of the advertising campaign, the simulation to determine an effectiveness value to attribute to the at least one change in the marketing media spend value relative to other marketing media spend values represented in stimuli values for other marketing channels of the plurality of marketing channels of the advertising campaign, the effectiveness value determined based on the weight matrix and representative of an effect of the first marketing channel on the other marketing channels of the advertising campaign.
10. The computer system as set forth in claim 9 , wherein the effectiveness value is displayed as a percentage.
11. The computer system as set forth in claim 9 , wherein at least one of the first portion of time series stimulus data or the second portion of time series stimulus data in the stimulation vectors includes at least one of, an advertising spend, a number of direct mail pieces, a number of TV spots, a number of radio spots, a number of web impressions and a number of coupons printed, and at least one of the first portion of time series response data or the second portion of time series response data includes at least one of a number of calls into a call center, a number of clicks on an impression, or a number of coupon redemptions.
12. The computer-implemented method of claim 1 , further including validating the learning model by executing a validation simulation, the validation simulation including determining predicted response vectors for a portion of historical stimulus data and comparing the predicted response vectors to historical response vectors associated with the portion of historical stimulus data.
13. The computer-implemented method of claim 12 , further including determining whether the predicted response vectors satisfy a validity threshold with respect to the historical response vectors associated with the portion of historical stimulus data.
14. The computer-implemented method of claim 13 , wherein, in response to the predicted response vectors not satisfying the validity threshold, the learning model is subjected to further training.
15. The computer-implemented method of claim 1 , further including removing noisy weight values by:
comparing the plurality of weight values to a threshold; and
designating ones of the plurality of weight values which exceed the threshold as true score values.
16. The computer-implemented method of claim 1 , wherein the correlating is to emphasize the more recent entries over the older entries based on an exponentially decaying model.
17. The computer-implemented method of claim 1 , wherein the first simulated model includes the first portion of time series stimulus data, the first portion of time series response data, and the weight matrix, and applying the second portion of time series stimulus data and the second portion of time series response data to the first simulated model includes:
updating the first portion of time series stimulus data with the second portion of time series stimulus data; and
updating the first portion of time series response data with the second portion of time series response data with the second portion of time series response data.
18. The computer-implemented method of claim 1 , wherein:
the first portion of time series stimulus data and the first portion of time series response data correspond to first known stimuli and first known responses collected at a first time; and
the second portion of time series stimulus data and the second portion of time series response data correspond to second known stimuli and second known responses collected at a second time, the first time earlier than the second time.
19. The computer program product as set forth in claim 5 , wherein the first simulated model includes the first portion of time series stimulus data, the first portion of time series response data, and the weight matrix, and applying the second portion of time series stimulus data and the second portion of time series response data to the first simulated model includes:
updating the first portion of time series stimulus data with the second portion of time series stimulus data; and
updating the first portion of time series response data with the second portion of time series response data with the second portion of time series response data.
20. The computer system of claim 9 , wherein the first simulated model includes the first portion of time series stimulus data, the first portion of time series response data, and the weight matrix, and applying the second portion of time series stimulus data and the second portion of time series response data to the first simulated model includes:
updating the first portion of time series stimulus data with the second portion of time series stimulus data; and
updating the first portion of time series response data with the second portion of time series response data with the second portion of time series response data.