CommercialPOV

Colin Shearer, vice president of data mining, SPSS

Does SPSS have a specific policy that advises companies that are collecting consumer data to do it responsibly?

We do not actually advise people on this. If we're asked about it, of course we'll advise them to check their local statutes. It's not an issue that we typically find we have to raise because most people who are dealing with customer data are well aware of it—and that doesn't just have to do with the data that we're analyzing it has to do with any data that they've held traditionally for marketing purposes or anything like that.

The normal concerns [with respect to analysis and data privacy] is that you're trying to dice people up as finely as possible, and at eventually at some level you're dealing with specific individuals. Really, you're abstracting and moving away from the individuals. We wouldn't formally call it metadata, but it is information about the customer base—as opposed to specific units of customer data.

Well look at a situation where I received not just one credit card offer—but four or five of them within the span of a couple of weeks. Is that based on some sort of personal data that might suggest I'm particularly receptive to that offer at that time? Is that not an individual level?

No. Here's the process that would have happened for that. Let's suppose it's a credit card company, and they have got information on all their customers. What they've built are descriptions of the groups of people—not individuals anymore—who are most likely to be receptive to a credit card. People in this class over here tended to take one, and people over there didn't. They don't care about the individuals in those classes—this is historical data, and you can build that model on anonymous data. You don't need to know who the individual is, you just need to know descriptions about them.

Where it becomes personal is how that data is applied. Like any type of marketing, what you're trying to do is target the right people with it. Where data mining comes in, is there's a model that's been built that can be a profile of people who are likely to go for this. So that can be applied to legally available data on all the people they might choose to sell to—and it selects the people most likely to respond to it.

Using the data mining model is no different than somebody in the marketing department saying we'll target all people who live in this zip code area or with this estimated income. Except marketing departments are now using data mining because it can give much better targeting than simple selection rules.

Is it possible that a consumer could be put in a group unfairly?

It's not due to any characteristic of data mining that that could happen. That tends to happen in most cases because you have credit agencies who have incorrect data. And they act on that and say that this person—because of some fictitious instances—isn't creditworthy. And indeed there have been many cases where people have appealed against that and gotten it change.

Can you kind of characterize the understanding that most companies have of their data?

I think it depends enormously on the type of data they've collected. In many organizations, the understanding of what we call 'structured data'—or the core database of the organization—is typically pretty good.

But you can get some wrinkles on that. For example, if you have companies who have been through many mergers and acquisitions, you often get totally inconsistent views of the customer that have to be resolved. You maybe have three insurance companies who have merged—all of whom have looked at the customer in three different ways—and you get three incompatible views of what makes a customer.

What are the specific challenges of merging customer data in a M&A?

They may not have the full set of data for each customer, so you may have to take the overlap of that and then try to fill in the rest through customer surveys and contact and an ongoing basis. You may have something as radical as one company that has everything organized by customer, and another company that has everything organized by products. In those cases, it can be hard to bring them together. Most people who put serious effort into collecting customer data and getting warehouses in place have a pretty good understanding of what's in the data, and they're looking for ways to leverage that data better and get more value out of it—be it for marketing or fraud detection or whatever it is.

Where is data mining technology heading?

One other very interesting area is the use of text mining technology. If you look at all the information in the world, something like 80% of it is in free text. That includes things like academic papers and web pages but also in terms of information of the customer, you might have a record of all the email they sent you or transcripts of calls. In many cases, organizations have no way of making use of unstructured data, like text. Text mining allows you to extract meaning from this unstructured data which previously wasn't used.

What would you say to a customer who had a kind of Big Brother apprehension about data mining technologies?

I would make the argument that what they're mining is anonymous data. I would also highlight the benefits of data mining. The risk of being deluged with irrelevant information—whether they're being sent junk mail or being shown nonsense ads on a web site—data mining can go in and eliminate those scenarios.