Marketing Optimization Using SAS

1
Marketing Optimization Using SAS
Randy Sherrod rasherro_at_cisco.com March 2008
2
Discussion Topics

• What is the impact of marketing investment on
  business metrics, e.g. sales?
• How can we determine the level of marketing
  investment that optimizes return?
• What data is required?
• What techniques are available?

3
Overview of Analytic Process
Source Cisco SMO
4
Econometric Model Quantifies Relationship Between
Bookings and Drivers First Step to Driving
Optimal Resource Allocation
Inputs

• Pricing
• Distribution Channel
• Sales People, Resellers, etc.
• Marketing Investment
• Macroeconomics
• Competitor Dynamics
• etc

Source Cisco SMO
5
Elasticity Measurements that Quantify the
Relationship Between Drivers and Bookings
x
x
x
x
x
x
x
x
x
x
x
Source Cisco SMO
6
Background Observations

• What range of elasticities can we expect?
• Sales Force ()
• Marketing ()
• TV ()
• Paid Search ()
• GDP ()
• What is the impact of GDP on marketing and sales?
  What might this mean for the optimal level of
  investment?

• How to determine the optimal level of sales force
  and marketing?
• Initial Values Sales Force400M,
  Marketing50M, Total Sales1B.
• Estimated Elasticities Sales Force0.40,
  Marketing0.20
• Suppose there is an additional 40M to allocate,
  how do you split between Sales Force and
  Marketing to maximize Sales?
• 40M10 of Sales Force?0.400.101B40M
  increase in Sales
• 40M80 of Marketing?0.200.801B160M
  increase in Sales

7
Modeling Possibilities

• Framework
• Log-linear models with SAS
• Proc GLM
• Proc Reg
• Proc Surveyreg
• Proc Genmod
• Proc Mixed
• etc.

Output from these procedures quantifies the
impact of marketing on sales
Source Cisco SMO
8
Modeling Details
Framework Log-linear model with customer-level
fixed effects
Log Salesitaiß1log Competitor Advertisingt-1
ß2log Sales Forcet-1 ß3log Marketingt-1 ß5log
Cust Satisfactionit-1 ß6log GDPt-1
Seasonality Where icustomer ttime (-1)lag 1
QTR

• 1. Imposes constant elasticity

2. Allows for many possible response curve shapes

• 3. Explicitly accounts for synergies between
  drivers

Source Cisco SMO
9
Modeling Details cont.
SAS Implementation proc surveyreg class
customer model log_salescustomer
log_comp_advertising_1 log_sales_1
log_marketing_1 log_cust_satisfaction_1
log_gdp_1 q4 /noint solution cluster
time quit
Creates cluster-consistent standard errors
Estimated model can then be solved for optimal
levels using proc optmodel.
Source Cisco SMO
10
Questions?