Automated specification, estimation, discovery of causal drivers and market response elasticities or lift factors
View Patent ↗In some embodiments, a software facility performs a method of automated specification of models, estimation of elasticities, and discovery of drivers using the framework(s) discussed elsewhere herein is provided. The facility first obtains the client, business, and/or brand goals in terms of profit optimization, volume or revenue goals, acquisition of new customers, retention of customers, share of wallet and upsell. In conjunction with these goals, the facility obtains cross-section meta-data related to the planning time horizon, markets, geographies, channels of trade and customer segments. In combination, the goals and meta-data define the structure of the data stack and the number of demand generation equations that are needed.
1. A method in a computing system for constructing a demand generation equation to predict the level of demand for an offering that will be generated by allocating particular levels of resources to different marketing activities, comprising:
receiving a plurality of observations from a user;
for each of at least two of the plurality of observations:
for each of a plurality of candidate driver variables, retrieving a value for the candidate driver variable as it relates to the observation;
for each of at least one outcome variable, retrieving a value for the outcome variable as it relates to the observation;
using the variable values retrieved for the observations, estimating an equation that relates the candidate driver variables to the outcome variable to produce coefficients for each of the candidate driver variables, wherein the estimating is performed using a generalized least-squares technique; and
conducting statistical diagnostics on the estimated equation to validate the estimated equation, wherein conducting statistical diagnostics on the estimated equation to validate the estimated equation comprises requiring that the estimated equation fall within the range of elasticity norms from an elasticity library.
2. The method of claim 1 wherein each of the candidate driver variables falls into a candidate driver variable class, and wherein estimating is performed under the constraint that at least one candidate driver variable in each candidate driver variable class must have a non-zero coefficient.
3. The method of claim 1 , wherein the step of conducting statistical diagnostics on the estimated equation to validate the estimated equation comprises verifying compliance with at least two pre-determined business or statistical confidence rules.
4. The method of claim 3 , further comprising applying the estimated equation to candidate driver variable values and outcome variable values not among those retrieved to validate the estimated equation.
5. A computer-readable storage medium whose contents cause a computing system to perform a method for constructing a demand generation equation to predict the level of demand for an offering that will be generated by allocating particular levels of resources to different marketing activities, the method comprising:
receiving a plurality of observations from a user;
for each of at least two of the plurality of observations:
for each of a plurality of candidate driver variables, retrieving a value for the candidate driver variable as it relates to the observation;
for each of at least one outcome variable, retrieving a value for the outcome variable as it relates to the observation;
using the variable values retrieved for the observations, estimating an equation that relates the candidate driver variables to the outcome variable to produce coefficients for each of the candidate driver variables, wherein the estimating is performed using a generalized least-squares technique; and
conducting statistical diagnostics on the estimated equation to validate the estimated equation, wherein conducting statistical diagnostics on the estimated equation to validate the estimated equation comprises requiring that the estimated equation fall within the range of elasticity norms from an elasticity library.
6. The medium of claim 5 wherein the estimating is performed by stacking multiple equations under the Seemingly Unrelated Regression technique or by solving simultaneous equations using 2SLS, 3SLS, or maximum likelihood methods.
7. The medium of claim 5 wherein each of the candidate driver variables falls into a candidate driver variable class, and wherein estimating is performed under the constraint that at least one candidate driver variable in each candidate driver variable class must have a non-zero coefficient.
8. The medium of claim 5 , wherein the step of conducting statistical diagnostics on the estimated equation to validate the estimated equation comprises verifying compliance with at least two pre-determined business or statistical confidence rules.
9. The medium of claim 5 , further comprising applying the estimated equation to candidate driver variable values and outcome variable values not among those retrieved to validate the estimated equation.
10. One or more computer memories collectively storing a demand generation equation data structure to predict the level of demand for an offering that will be generated by allocating particular levels of resources to different marketing activities, the demand generation equation data structure comprising:
a plurality of observations received from a user;
for each of at least two of the plurality of observations:
a plurality of candidate driver variables, and a value for the candidate driver variable as it relates to the observation;
at least one outcome variable, and a value for the outcome variable as it relates to the observation;
a demand generation equation that relates the candidate driver variables to the outcome variable to produce coefficients for each of the candidate driver variables, wherein the demand generation equation is estimated using the variable values retrieved for the observations and a generalized least-squares technique;
an elasticity library;
coefficients for each of the candidate driver variables; and
at least two statistical diagnostic results generated from conducting statistical diagnostics on the demand generation equation to validate the demand generation equation.
11. The one or more computer memories of claim 10 , wherein each of the candidate driver variables falls into a candidate driver variable class, and wherein the demand generation equation is further estimated under the constraint that at least one candidate driver variable in each candidate driver variable class must have a non-zero coefficient.
12. The one or more computer memories of claim 10 , wherein the at least two statistical diagnostic results are further generated by verifying compliance with at least two pre-determined business or statistical confidence rules.
13. The one or more computer memories of claim 10 , the demand equation data structure further comprising:
coefficients for candidate driver variables other than the plurality of candidate driver variables.
14. The one or more computer memories of claim 13 , wherein the coefficients for candidate driver variables other than the plurality of candidate driver variables are generated by applying the estimated equation to the candidate driver variables other than the plurality of candidate driver variables.
15. A method in a computing system for constructing a demand generation equation, comprising:
receiving a plurality of observations from a user;
for each of at least two of the plurality of observations creating a data stack comprising data observations and explanatory variables by:
retrieving a value for a candidate driver variable as it relates to the observation; and
retrieving a value for an outcome variable as it relates to the observation,
estimating a demand generation equation that relates the candidate driver variable to the outcome variable to produce coefficients for the candidate driver variable using the variable value retrieved for the observations,
wherein the estimating is performed using a generalized least-squares technique; and
conducting statistical diagnostics on the demand generation equation to validate the demand generation equation, wherein conducting statistical diagnostics on the demand generation equation to validate the demand generation equation comprises requiring that the demand generation equation fall within the range of elasticity norms from an elasticity library.
16. The method of claim 15 wherein each of the candidate driver variables falls into a candidate driver variable class, and wherein estimating is performed under the constraint that at least one candidate driver variable in each candidate driver variable class must have a non-zero coefficient.
17. The method of claim 15 , further comprising applying the estimated equation to candidate driver variable values and outcome variable values not among those retrieved to validate the demand generation equation.