February 01, 2007

An Answer: Measuring Advertising Effectiveness

Back in November, I issued a challenge to the analytical multichannel community. Given a dataset with four independent marketing activities, mostly targeted to the same set of customers, measure the incremental sales that each marketing activity drove to the catalog and online channels.

This challenge is important, because executives at many companies have asked me if I have a solution to this problem. Many vendors provide solutions to these problems. However, the problem should be able to be solved without somebody having to spend $75,000 for a solution. The good folks at KDNuggets posted the challenge, which I am appreciative of.

I did forward the challenge to all of the major vendors in the multichannel analytics industry. The folks at Abacus were kind enough to have a brief dialogue, and promised to attempt to solve the problem if there were enough internal resources available.

Free advertising and thought leadership was at stake for these businesses. Sadly, I did not receive one response from folks at SAS, SPSS, Abacus, Experian, Acxiom, Merkle, KnowledgeBase, Unica, or others.

More than five hundred individuals downloaded the dataset. I received several solutions. One solution passed the test.

The solution was provided by an individual named Grigorios Tsoumakas. His solution is elegant, simple, and directionally correct. His solution is outlined below. Kudos to Grigorios for finding an exceptionally simple solution!!


For the organic sales we consider the sum of sales for which no advertising tactic was used. Then for each advertising tactic we calculate the sum of sales when no other advertising tactic is used. The exception to this rule is the email2 tactic, which has a temporal relation with the email1 tactic, i.e. it follows email1. For the email2 tactic we calculate the sum of sales when both email1 and email2 tactics are used. This will give us an initial estimate of the sales related to a single tactic.

Then we consider pairs of tactics (apart from the email1, email2 pair). We divide the sum of sales for every pair of tactics to the two tactics that comprise this pair, based on the initial estimates of sales. For example if a tactic A had initial estimate sumA1 and tactic B had initial estimate sumB1, then we divide the sum of sales for the pair of tactics A,B sumAB as follows:

sumA2 = (sumA1/(sumA1 + sumB1)) *sumAB and

sumB2 = (sumB1/(sumA1 + sumB1))*sumAB.

We continue by considering triplets of tactics and again we divide the sum for a triple according to the initial estimates as before.

Finally we distributed the sales for all tactics again according to the initial estimates.

The resulting table is as follows:

Final Results Of The Advertising Challenge

Telephone Online Total

Sales Sales Sales
Driven By The Catalog $57,843 $44,524 $102,366
Driven By The Postcard $1,221 $18,229 $19,451
Driven By The 1st E-Mail $0 $5,350 $5,350
Driven By The 2nd E-Mail $0 $2,400 $2,400
Sales That Happened Organically $244 $1,778 $2,023
Grand Totals $59,308 $72,282 $131,590