Data, data, who’s got the data? Thanks to our luck of being product managers in the 21st Century we are privileged to have access to quantities of data about our product and our customers that product managers of old could only dream about. However, is this really a good thing?
Dr. Peter Fader is a professor of marketing at the University of Pennsylvania’s Wharton School. He spends his days studying how to use behavioral data to forecast product sales and manage customer relationships. What really made him famous was way back in 2000 when he testified at the Napster trial and said that Napster had actually boosted music sales!
Fader really knows his stuff and he’s got some stern words for us product mangers. When it comes to collecting data on our products and our customers, we’ve been doing a pretty good job. However, Fader says that how we’ve been using this data is not so good. He says that we’ve been attempting to make conclusions and future predictions that just don’t hold up. Oh, oh. Looks like it’s time for us to go back to school…
Fader says that what we are doing wrong is that we’ve just been doing simple data mining of our data. He says that what we need to be doing is to combine data mining with probability models. Can you say “math”?
Data mining is a great way to classify data. Specifically, if you are trying to figure out WHY one group of people is different from another group, then data mining tools and techniques are the way to go. However, data mining is not very good at saying WHEN things will happen.
That being said, data mining can be used to answer certain time-sensitive questions like what customers will order a given product during the holiday season. Where it will fall down is when you want to make a forecast about about what particular customers might do in the future – not just what product they are likely to buy next.
What product managers have been missing is that our customers are creatures with random behavior. No matter how much data you collect about them or your product, you’ll never collect enough information to accurately answer these sorts of questions.
At the end of the day, Fader reminds us that there’s only so much that we can ever hope to nail down just by capturing more data.
What Fader suggests that product managers start to do is to use probability models along with data mining. What he’s getting at is that for all of our talk about one-to-one marketing, we’ve really been missing the ball. Fader says because people are random, we can’t really say what any one customer is going to do. However, we can say what a group of customers will do (just not what any specific customer will do).
Fader says that by using probability models, product managers can answer three important questions about their products: timing – how long will it be until something happens, counting – how much of something (arrivals, purchases, etc.) will we see over some period of time, and choice – if we give our customers an opportunity to do something, how many of them will actually do it?
If you take these three questions, you can combine them in a number of interesting ways in order to get answers to more complex questions. For example, how long someone spends on your web site during a month is really two timing counts put together: number of visits and duration of each visit.
So, to wrap this discussion up, what Fader is suggesting that we do is to start with probability models to create forecasts for how long a customer will stay with our company or how many times they will buy from us this year. Once you have this basic behavior captured, then use data mining to get an understanding of why groups of customers who have different behavioral tendencies are so different from each other.
Do you use data mining to try to better understand your customers today? How has that been working out for you? Do you use probability models today? Do you think that they would work well with your data mining tools? Leave me a comment and let me know what you are thinking.