Subscribe to DSC Newsletter

How To Measure Performance of TV, magazine, email campaigns in Google Analytics with Campaign Tracking / Tagging

This question was initially asked on the WAA Linked group.

Here's my answer:

You don't need to do any tagging. At the end of the month, check out how much money came in, how much came out, and use a parametric advertising mix optimization model based on contrast analysis. Not only you can tell (from using such a model) if print advertising works better than TV, but you can also tell: 

  • how long it takes for TV or print ads to reach its peak 
  • how a TV ad viewed today, compared to a TV ad seen 3 months ago, contributes to a sale (in short: square root function of time, 6-month time lag between first view of TV ad and sales peak for one of my clients) 
  • how all the channels (radio, TV, print, direct mail, online ads) interact 
And you can do this without gathering any personal information about your prospects - not even web log stats. 

Note: when you send out monthly newsletter, your newsletter vendor should provide you with a bunch of metrics, including open rate, click rate, attrition, etc. at the member level. If not, you need to talk to me.

Additional note:

I did this type of analysis for NBCi (NBC Internet), and we were able to tell: 

  • which TV show worked best in terms of user retention 
  • what application was best in terms of user retention / customer cost of acquisition / lifetime value

 

Views: 152

Replies to This Discussion

Interesting answer from  Mattheos Protopapas on our LinkedIn group: 

I agree that tagging is not necessary. There are several types of marketing mix models (regression-time series models that explain sales on advertising and in-store promotional activities such as price offers and gift packs, and other significant variables such as prices and product availability) that have been proven to be successfull in providing important insights, such as Little's ADBUG model, multiplicative models, Distributed Lags Models, VAR's, state space models, Bayesian approaches etc. Some preliminary exploratory data analysis should hint the proper model to use in a specific case. Just note that in some of those classes of models lagged values of the advertising investment are not used, so appropriate transformations that combine the past and current values of the number of ads viewed/read should be used to form the corresponding explanatory variables to capture all the media effects (such as wear-in/wear-out and carry over effects, and permanent shifts to the base level of sales due to an advertising campaign). Also, the effect of marketing activities in generally non-linear; so transformations should also be used to link the current media investment/ad views and their current effects (a logistic curve might be used for example). All these non-linearities might call for the use of non-linear optimisation methods for MLE / NLS. Numerical methods might be unsuccessful - my experience is that methods such as BFGS/SQP sometimes diverge. A meta-heuristic that I have found to be useful on that is Differential Evolution.
On the other hand, the enhanced capabilities of the Internet when it comes to data that is available to the modeller, allows for better customer profiling, route tracing inside the stores' website and so on, and even more complex models that might address issues such as the impact of the design of the website on sales, or the impact of the ranking on google searches of the site on the sales of a specific product in conjunction with the keywords used in the search etc. So tagging, and any data available in general, might be useful (that's always true, of course).

RSS

On Data Science Central

© 2020   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service