http://www.convergys.com/pdf/ES3-025N.pdf?TRID=1

Jeff Hazel
Director, Convergys Corporation

This is so true but equally so difficult to change on the ground. The marketer is constantly grappling with the contradictions of needing to act with “speed” and not having all the “data” to make the correct choices. He therefore relies on intuition and justifies it to managment later. An interesting thing is happening in the Indian Telecom market! All telecom companies put together are acquiring between 11 to 15 million new customers per month. But most telecom operators are just feeding a frenzy of downward spiralling ARPUs! With telecom companies capturing so much transactional data, I wonder if there is another, rational and more effective way to market-sell more services to existing customers rather than continue the rush to purely build acquisition scale. I have some thoughts on this at my blog:

http://blog.cequitysolutions.com/Customer-Management-blog/bid/9943/Making-the-data-talk-in-Telecom

I would love to have your comments!

By putting the right items in the upper model, you can go pretty far in explaining why you have observed the results that you have.

So instead of just being a model that predicts well in-sample, we can use the upper model to explain additional variation AND we can use the upper-lower relationship to predict how a new sample of people will respond.

Perhaps because I was trained in an experimental discipline, I tend to care about whether A caused B or both are caused by C. From the standpoint of action, this distinction may or may not be important. If I observe that some mold growing in my bacterial cultures is killing the bacteria, I may be able to devise an effective antibiotic treatment without understanding the mechanism of action. Understanding the mechanism of action, however, may make it easier to find other substances that are equally effective, or to even invent compounds from scratch that will have anti-bacterial properties.

Most, if not all, predictive models start with some idea, whether implicit or explicit, about the underlying mechanism relating the independent and dependent variables. Cinematch and other collaborative filtering systems are based on a “similarity” mechanism. The data that Netflix provided for the prize competition do not facilitate extensive exploration of other possible mechanisms. This becomes a potential limitation if I’m not satisfied with the films that Netflix recommends because my movie enjoyment is caused by factors that are not captured in similarity. On the other hand, if the similarity assumption captures the underlying mechanism (A and B both caused by C, which is unobserved but reflected in A and B), then it probably won’t be a practical problem for Netflix.

This question revolves about the nuanced meaning of “understanding”. I estimate that, of all the times that my analytical results have been challenged by disbelieving business people based on their experience in an industry, their rules of thumb were subsequently demonstrated to be preposterously wrong (most of the other half was a question of different definitions, etc.). Those business people certainly were comfortable thinking that they understood their business or market, but it is plain that they were mistaken. For may part, I would only claim to “understand” the situations in question to the extent that I could do a better job of anticipating their outcome.

“We might also lapse into an expectation that “prediction” based on past behavior is in fact destiny. We need to remind our selves that correlation or association is a necessary but not a sufficient condition to show a causal relationship.”

I submit that this is a separate issue entirely. I think the hand-wringing which stems from the correlation-vs-causation issue is misplaced. If I discover a relationship between measured events A and B, in which knowledge of A (which is available to me ahead of time) imparts some knowledge of B (which is of some practical interest), then, assuming that the relationship continues to hold for the population of interest, I do not care whether A “caused” B, or some third event C “caused” both A and B, or something else entirely is going on.

It’s a pity Mr. Lindstrom’s research didn’t provide more fodder for analysis. But your review of his book was truly useful, VERY even-handed, and a nice read all by itself, so thanks!

I agree that theory-building is worthwhile. If we understand the customer better we can make a number of useful generalizations. For example we already know that people are motivated by perceived scarcity and the desire to appear consistent, social proof and a number of there “tendencies.”

I have taken a different tact by using context-drive qualitative research. This assumes that each purchasing scenario is unique and that “context” is or particular importance. I develop a series of pictographs to get respondents to explain their purchasing behavior in a specific situation. I don’t ask questions, rather, I allow the respondent to tell their story.

I admit that this is not directed at theory building. The only theory involved is that this can be considered an extension of Thematic Apperception Testing and that comes from psycho-analytics not marketing theory. Nevertheless, I find adding context useful. See: http;//beyondfocusgroups.blogspot.com.

Dale Paulson, Ph.D.

Very “insightful” observation, and it teaches a very simple and powerful lesson. We must constantly evaluate what goal we are trying to accomplish, and how we are trying to accomplish that goal. Assessing what we know, and what we do not know, are critical first steps that will help us decide what further data is necessary in the context of a business problem (very important – gets back to the goal). From there, we can apply analysis to help transform our findings into actionable insights. Keep up the great posts.

jim