Comments on: ***** What Data Mining Can and Can’t Do

By: Jim Novo

Jim Novo — Fri, 06 Jul 2007 20:58:57 +0000

I’m not following you James. He’s a professor at the Wharton school, what benefit does he get from pushing an approach? On the other hand…you work for Fair Issac, right?

Do you dispute the basic point that data mining is good at classification but not good at probability? Just askin’.

By: James Taylor

James Taylor — Fri, 06 Jul 2007 20:34:14 +0000

I think the article is more an attempt to push his approach than anything else. The problem with data mining (or rules or probability models or anything really) is that the technology becomes the focus not the problem. Companies need to decide what DECISION they are trying to influence and then figure out what technology (business rules, optimization, analytics, data mining…) will help them do a better job of making that decision.
JT
Author, with Neil Raden, of Smart (Enough) Systems, a book about this.

By: Jim Novo

Jim Novo — Fri, 06 Jul 2007 00:31:25 +0000

Paul and Bhupendra, thanks for the comments. Based on your input, I’m thinking I did not express myself clearly. So I’ll try again, since there is a lot of confusion around data mining in the Marketing world. Rather than run off on a long comment, I will write a new post…

By: Bhupendra

Bhupendra — Wed, 04 Jul 2007 06:43:11 +0000

Jim, the article is interesting and I have taken your concern seriously. But I have some reservations against both the articles.

I have written a blog entry to take forward the discussion. We will need to address this issue quite seriously.

http://analyticsbhups.blogspot.com/2007/07/what-data-mining-can-do-and-cant-do.html

– Bhupendra

By: Paula Thornton

Paula Thornton — Tue, 03 Jul 2007 15:33:24 +0000

Both this and the other piece are very highly simplistic views of this space. AND, I’m agreeing what is fundamentally being said, but there are other considerations. My take is, keep it on the radar. Where it is absolutely the right mechanism for loss of money, it will raise itself (MCI used it extensively to detect fraudulent use of accounts, esp. calling cards).

And it has great marketing potential (actually, more about relationship potential, than marketing potential…but then, who owns relationships is the big fundamental disconnects for most companies). The problem is drawing conclusions from the findings. Data mining’s real potential is to call attention to things for further investigation. It helps to identify real possibilities (for significance of possibilities, see the 5P’s of Design & Development.

But it takes a good head to add more filters to the considerations. For example, in 1996 I was at a conference where a vendor was displaying a 3D rendering of the data and was pointing out the significance of the red area — people who had called the call center excessive times. They proposed that this be further investigated to cut down costs for the company…perhaps these were people you didn’t want to do business with. To the contrary, I would have been IN that data. At that time I had changed residences and had moved my phone number…all arranged weeks in advance. I spent hours getting hung up on and being given random answers to account for why my phone was not in service and/or when I might expect to have service.

As Avinash says, the tool should be 10% of the solution, the other 90% is raw brains. In the MCI scenario, that was the case…the ‘model’ was constantly tweaked by the guy who designed the algorithm — the tweaking was done based on how accurate the results were.

It is this and this alone that is CRITICAL to the use of any metric.