Expanding the Data Miner's Toolbox

Mary Grace Crissey, analytics marketing manager at SAS, and Mike Gilliland, SAS's global marketing manager, speak to Business Finance about how the developer's recently enhanced data mining offerings are helping finance executives to drill ever deeper into their company's data pools.

BF: Your SAS Enterprise Miner offering was introduced probably 10 years ago, so why are we hearing so much about it today? What's new or different here?

Crissey: What we released was probably the biggest upgrade for this product in, say, almost 4 or 5 years. Now, different companies may brag about every single new release, but this was much more than just some little enhancement. This was a substantial addition to its power, and by that I'm referring to the algorithms, the math underneath the solutions. I counted 15 different analytical additions to the toolbox that SAS Enterprise Miner has, so you can now throw whatever kind of data at it and hopefully it will just find the pattern for you. Obviously, it still needs to be driven by somebody who is aware of the power of data.

What we've added now is things that customers have told us that they've either wished they had or they had done themselves. ...We have some firms, especially in the financial world, that have their own programs that they've been writing. Now we have a way of easily transforming their in-house scoring code and their in-house rules so that they will fit and just click kind of into place in our work box, our toolbox that we call Enterprise Miner.

BF: So what would the data mining finance executives out there say? Are they saying, "It was about time, we've been waiting"?

Crissey: It wasn't as if we were limping along until we got these things fixed. It's more that they're so eager to apply their data mining contributions to more parts of the company, to more challenges, without having to hire more people, because those salaries are quite expensive. They're just eager to get more done faster, and mostly to find out where the decisions are that they can make that will make the biggest impact on the bottom line.

A lot of this is investigation work. We called it data mining similar to the old-fashioned digging for diamonds in a bunch of dirt. You don't always know where to look. With our toolbox, now you can apply it to more data, more kinds of data, faster. It'll be more scalable. We can reach across different departments that before we didn't really have the time to explore. With this toolbox, yes, they'll be finding patterns better because the algorithms are going to pop up with some interesting results more quickly and more visually, so that finance guys can take action. They can say, "Well, I know what can improve this little department here, but the biggest improvement we would make would be if we spent some time and devised a better plan over here in this department." It's the bigger reach, I think, that'll make the difference.

BF: As far as the development of the new Enterprise Miner goes, what part of SAS undertakes these types of upgrades and what talent does it require to take offerings to the next level?

Crissey: It is part of our research team. We are always proud about our R&D. They did spend a lot of time actually going to data mining conferences, participating -- making presentations as well as listening. But this is just not the R&D team with the latest technique. In fact, some of the math involved here was published as recently as 2002, 2003, by a data mining person at Stanford. This is fairly recent math that's really just been out and tested.

In addition to the R&D guys -- whom we do brag about at SAS, and we do hire, and incentivize, and try to keep working for us as a good place to work -- in addition to R&D, I'd say that it's the customer advisory boards. We've actually started a pretty significant effort to enhance the feedback from customers. When we had our Model Manager product -- that was a new release that we had out -- that actually was a product designed to help deploy some of the data mining results, and that whole product was a result of talking to some of our customers.

We have large banks, such as U.S. Bank and Wachovia. We have credit card agencies. We made a significant effort in getting some of these focus groups and customer feedback -- a lot of on-site visits -- to understand what the customers' challenges are. Meanwhile, in the government sector, they're trying to find a terrorist at the airport (types of scenarios) or perhaps it's activity involving credit cards where there are stolen identities. There are lots of different types of customer examples. We did a substantial focus on what the customers were looking for so that we could better create a product this time.

BF: Are your largest customers likely to be the first on board?

Crissey: Actually, if they are a current customer of our data mining offerings, they get this for free. This is part of our licensing procedure. It's a perpetual license, and this means that you get the next version free if you already have the previous version's license. When you ask, Will they be the first on board?, well, it does sometimes take an installation, a download. It's not always a quick little add-this-color-to-your-inventory-that-you're-hanging-in-your-wardrobe-closet. It sometimes does mean that they need to install it on their weekend or whatever they have. It's the bigger the place. Sometimes there are restrictions on when they can do their upgrades, but we are all eager to have them do that first. There are some customers that actually are about three releases behind. They have chosen not to upgrade because there were some interactive features that they liked to keep or that they were just comfortable with. Now, with this current switch, we have made sure that all of the reasons that people haven't upgraded in the past are called upon and made easy. I think that this is the first easy upgrade that we've had in the last, I'd say, three releases of the data mining products, typically. Yes, we do hope that this will be one that small companies, as well as the big ones, will jump on board with.

BF: You would think that it would be easier for the larger players with a lot of dedicated people, who perhaps specialize in data mining, to adopt offerings such as these more quickly than, say, midsize accounts. And some of the midsize accounts might say, "Well next year we'll have more people to be doing something more there. We'll talk to you then." What can you tell us about the types of resources required to adopt these types of offerings?

Crissey: Yes, at first I thought you were asking about titles and scoring officers -- a title you will find in a large bank. Scoring is when you apply your model, your predictive model, to new data. So you've already figured out what patterns you want to find, and then you want to put it in the new stuff and find out what actions to take. That's where you find out, okay, this is the group that we care about and these are the people whom we want to send our mail out to, or something similar.

Some departments have a scoring officer already assigned, and that's a very unique title. It comes only if you have a data mining group. But yes, like you said, the small to medium-size companies, they probably have an analytical group, but not dedicated to data mining. The reason they will be eager to upgrade is that they would like to have one place to go for their analytics. And SAS is able to offer that. There are several different niche vendors out there with different kinds of data mining solutions. You could buy a decision-tree product, or you could buy a product that does regression really well. Or you might buy a different kind of software that's into some of this artificial intelligence--neural network analysis. But you don't have to make that decision with Enterprise Miner and SAS, because we offer you the whole suite.
Especially a small or middle-size company, which doesn't want to have lots of different software packages on their desks, they want it all together. They don't always know what kind of data they're going to get to work on. The data might be varied a lot depending on whether they are a consulting base, or they've got a new company that's been in business only, say, five years or less and they don't have a lot of historical trends and a bunch of data. They're just looking at the current stuff. And they would be most needy of a wide range of tools.

BF: How are SAS offerings today helping to supply more BI to more people across organizations?

Gilliland: As far as expanding the tools goes, and really making them available to a broader audience within a company, a good example might be in forecasting, where SAS has provided forecasting tools for 30 years. We realized probably 5 to 7 years ago that not everybody wants to have to hire a statistician to go in there and build these models and write custom code. Two years ago, we released what is our flagship product, SAS Forecast Server, to provide large-scale automation of the forecasting process.
What we added was a user interface to our forecasting engine. This is high-performance forecasting, essentially giving the user the choice of running completely on autopilot -- letting the system read the data, diagnose the history, create the models, and generate the forecasts -- or letting them step in and tweak, make adjustments, set options, and so on, and have some level -- actually quite a high level -- of control over the modeling, without having to write code.

Crissey: Yes, I think that this is a strength of SAS -- the BI not only gives you the data, but also lets the decision makers touch it and play with it a little.

Gilliland: Particularly in forecasting, where some companies will go to the effort of actually hiring a Ph.D. in statistics or something like that -- a very high-powered analyst -- to do the forecasting for them. From my experience, which is predominantly in a consumer-products manufacturing areas, I think that most companies just have a business analyst who may have a little bit of statistics from their college education, but not a whole lot of specific stuff related to forecasting. They're the ones responsible for forecasting. With Forecast Server and the user interface, they can now go in and get access to all the power of SAS forecasting, without having to learn how to write the code and without necessarily becoming experts in statistics. They can, as a result, can get better forecasts from a system -- again, without having to rely on some Ph.D. in statistics to build the models for them.

Crissey: It's better, because they know what to do. They can develop an action plan. They already have some ideas from having identified trends that they see coming, and they can prepare. And that's what these enhancements truly offer.