CROSS-CHANNEL PREDICTIVE MODEL
A method, system, and computer program product for advertising portfolio management. The method form processes steps for determining effectiveness of marketing stimulations in a plurality of marketing channels included in a marketing campaign. The method commences upon receiving data comprising a plurality of marketing stimulations and respective measured responses, then determining from the marketing stimulations and the respective measured responses, a set of cross-channel weights to apply to the respective measured responses, where the cross-channel weights are indicative of the influence that a particular stimulation applied to a first channel has on the measure responses of other channels. The cross-channel weights are used in calculating the effectiveness of a particular marketing stimulation over an entire marketing campaign. The marketing campaign can comprise stimulations quantified as a number of direct mail pieces, a number or frequency of TV spots, a number of web impressions, a number of coupons printed, etc.
1 . A computer-implemented method comprising:
receiving, using a computer, historical data that comprises stimuli data that represent a plurality of stimuli deployed through a plurality of channels to a plurality of users, the stimuli data comprising the stimuli and a time for deployment of the stimuli, the historical data further comprising response data that measures responses to the stimuli by the users and times the responses occurred;
generating, in a computer, from the stimuli data, a plurality of “n” time series stimuli vectors, SV n , that characterizes the stimuli with a plurality of values taken over a historical time period for each of the “n” stimuli vectors;
generating, in a computer, from the response data, a plurality of “n” time series response vectors, RV n , one for each of the corresponding “n” time series stimuli vectors, SV n , that characterizes the responses to the stimuli with a plurality of values over the historical time period;
generating a cross-channel predictive model, using machine-learning techniques in a computer, by identifying, for each of the “n” time series stimuli vectors, SV i , time and value based correlations between SV i and each of the time series response vectors, RV j , wherein “i” and “j” comprise integers from 1 to “n” and “i” is not equal to “j”, and by generating a plurality of cross-channel contributions that specify, for the time series stimuli vectors, SV n , a weight that quantifies the contribution of the time series stimuli vectors, SV i , to the time series response vectors, RV j , and
executing, in a computer, using the cross-channel predictive model, a simulation to change at least one of the values of the stimuli, SV i , associated with a first channel, and to predict, using at least one of the weights for the cross-channel contributions for one or more time series response vectors, RV j , associated with a second channel, at least one cross-channel response, wherein the first and second channels comprise different channels.
2 . The computer-implemented method as set forth in claim 1 , wherein generating a cross-channel predictive model, using machine-learning techniques in a computer further comprises training a set of exogenous variables into the cross-channel predictive model to attenuate effects independent from the stimuli.
3 . The computer-implemented method as set forth in claim 1 , further comprising:
receiving, in a computer, a range of values to vary the stimuli in at least one of the channels; and
executing, in the computer, a simulation, using the cross-channel predictive model, that sweeps across the range of values as the stimuli and produces additional weights for the cross-channel contributions.
4 . The computer-implemented method as set forth in claim 1 , further comprises generating a report that quantifies the return on investment for at least one of the channels that accounts for the cross-channel contributions.
5 . The computer-implemented method as set forth in claim 1 , wherein generating a cross-channel predictive model, using machine-learning techniques in a computer, further comprises:
filtering the weights so as to eliminate small cross-channel weights, statistically insignificant cross-channel weights and statistically outlying cross-channel weights.
6 . The computer-implemented method as set forth in claim 1 , wherein generating a plurality of “n” link series stimuli vectors and a plurality of “n” time series response vectors comprises introducing at least one delay between the “n” time series stimuli vectors and the “n” time series response vectors.
7 . The computer-implemented method as set forth in claim 1 , wherein generating a cross-channel predictive model, using machine-learning techniques in a computer, comprises generating the cross-channel predictive model in accordance with the expression:
r
=
n
∑
xy
-
(
∑
x
)
(
∑
y
)
n
(
∑
x
3
)
-
(
∑
x
)
3
n
(
∑
y
2
)
-
(
∑
y
)
3
where:
x represents components of the time series stimuli data,
y represents components of the time series response data, and
n is the number of {x, y} pairs.
8 . The computer-implemented method as set forth in claim 1 , wherein generating a cross-channel predictive model comprises generating a cross-channel predictive model with only a portion of the stimuli data.
9 . A computer program product embodied in a non-transitory computer readable medium, the 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:
receiving, using a computer, historical data that comprises stimuli data that represent a plurality of stimuli deployed through a plurality of channels to a plurality of users, the stimuli data comprising the stimuli and a time for deployment of the stimuli, the historical data further comprising response data that measures responses to the stimuli by the users and times the responses occurred;
generating, in a computer, from the stimuli data, a plurality of “n” time series stimuli vectors, SV n , that characterizes the stimuli with a plurality of values taken over a historical time period for each of the “n” stimuli vectors;
generating, in a computer, from the response data, a plurality of “n” time series response vectors, RV n , one for each of the corresponding “n” time series stimuli vectors, SV n , that characterizes the responses to the stimuli with a plurality of values over the historical time period;
generating a cross-channel predictive model, using machine-learning techniques in a computer, by identifying, for each of the “n” time series stimuli vectors, SV i , time and value based correlations between SV i and each of the time series response vectors, RV j , wherein “i” and “j” comprise integers from 1 to “n” and “i” is not equal to “j”, and by generating a plurality of cross-channel contributions that specify, for the time series stimuli vectors, SV n , a weight that quantifies the contribution of the time series stimuli vectors, SV i , to the time series response vectors, RV j ; and
executing, in a computer, using the cross-channel predictive model, a simulation to change at least one of the values of the stimuli, SV i , associated with a first channel, and to predict, using at least one of the weights for the cross-channel contributions for one or more time series response vectors, RV j , associated with a second channel, at least one cross-channel response, wherein the first and second channels comprise different channels.
10 . The computer program product as set forth in claim 9 , wherein generating a cross-channel predictive model, using machine-learning techniques in a computer further comprises training a set of exogenous variables into the cross-channel predictive model to attenuate effects independent from the stimuli.
11 . The computer program product as set forth in claim 9 , further comprising:
receiving, in a computer, a range of values to vary the stimuli in at least one of the channels; and
executing, in the computer, a simulation, using the cross-channel predictive model, that sweeps across the range of values as the stimuli and produces additional weights for the cross-channel contributions.
12 . The computer program product as set forth in claim 9 , further comprises generating a report that quantifies the return on investment for at least one of the channels that accounts for the cross-channel contributions.
13 . The computer program product as set forth in claim 9 , wherein generating a cross-channel predictive model, using machine-learning techniques in a computer, further comprises:
filtering the weights so as to eliminate small cross-channel weights, statistically insignificant cross-channel weights and statistically outlying cross-channel weights.
14 . The computer program product as set forth in claim 9 , wherein generating a plurality of “n” time series stimuli vectors and a plurality of “n” time series response vectors comprises introducing at least one delay between the “n” time series stimuli vectors and the “n” time series response vectors.
15 . The computer program product as set forth in claim 9 , wherein generating a cross-channel predictive model using machine-learning techniques in a computer, comprises generating the cross-channel predictive model in accordance with the expression:
r
=
n
∑
xy
-
(
∑
x
)
(
∑
y
)
n
(
∑
x
3
)
-
(
∑
x
)
3
n
(
∑
y
2
)
-
(
∑
y
)
3
where:
x represents components of the time series stimuli data,
y represents components of the time series response data, and
n is the number of {x, y} pairs.
16 . The computer program product as set forth in claim 9 , wherein generating a cross-channel predictive model comprises generating a cross-channel predictive model with only a portion of the stimuli data.
17 . A computer system comprising:
a computer processor to execute a set of program code instructions; and
a memory to hold the program code instructions, in which the program code instructions comprises program code to perform:
receiving, using a computer, historical data that comprises stimuli data that represent a plurality of stimuli deployed through a plurality of channels to a plurality of users, the stimuli data comprising the stimuli and a time for deployment of the stimuli, the historical data further comprising response data that measures responses to the stimuli by the users and times the responses occurred;
generating, in a computer, from the stimuli data, a plurality of “n” time series stimuli vectors, SV n , that characterizes the stimuli with a plurality of values taken over a historical time period for each of the “n” stimuli vectors;
generating, in a computer, from the response data, a plurality of “n” time series response vectors, RV n , one for each of the corresponding “n” time series stimuli vectors, SV n , that characterizes the responses to the stimuli with a plurality of values over the historical time period;
generating a cross-channel predictive model, using machine-learning techniques in a computer, by identifying, for each of the “n” time series stimuli vectors, SV i , time and value based correlations between SV i and each of the time series response vectors, RV j , wherein “i” and “j” comprise integers from 1 to “n” and “i” is not equal to “j”, and by generating a plurality of cross-channel contributions that specify, for the time series stimuli vectors, SV n , a weight that quantifies the contribution of the time series stimuli vectors, SV i , to the time series response vectors, RV j ; and
executing, in a computer, using the cross-channel predictive model, a simulation to change at least one of the values of the stimuli, SV i , associated with a first channel, and to predict, using at least one of the weights for the cross-channel contributions for one or more time series response vectors, RV j , associated with a second channel at least one cross-channel response, wherein the first and second channels comprise different channels.
18 . The computer system as set forth in claim 17 , wherein generating a cross-channel predictive model, using machine-learning techniques in a computer further comprises training a set of exogenous variables into the cross-channel predictive model to attenuate effects independent from the stimuli.
19 . The computer system as set forth in claim 17 , further comprising:
receiving, in a computer, a range of values to vary the stimuli in at least one of the channels; and
executing, in the computer, a simulation, using the cross-channel predictive model, that sweeps across the range of values as the stimuli and produces additional weights for the cross-channel contributions.
20 . The computer system as set forth in claim 17 , further comprises generating a report that quantifies the return on investment for at least one of the channels that accounts for the cross-channel contributions.