Sharing the info wealth

Don Hatcher leads a team that shapes the strategic direction of SAS Institute's products and technologies. SAS's current initiatives focus in part on distributing analytics throughout its customers' organizations, rather than limiting access to the tools to a few highly trained individuals, said Hatcher in an interview with Computerworld's Tommy Peterson. He also said the biggest barriers to a business-intelligence (BI) implementation don't always involve technology.

Q: What is the thing your customers are asking for most often?

If I had to give you one thing, I think it is enabling a greater percentage of the enterprise to leverage our analytics. It's dissemination. There's this concept that we have, which is the information supply chain. It started by just getting access to data -- that's where all the vendors were focused. Then we all were focused on storing it and being able to analyze it. And then we were focused on having some tools so you could analyze it differently. The big focus nowadays for SAS is, How do we get this breadth of information out to the wider audience? Because we believe that 80 percent of a customer's enterprise needs are business intelligence today. We've had customers tell us that (they) can't find enough business analysts. We need to empower domain experts and information consumers to do some things themselves. That's what you (will) see us delivering in (Version) 9.1 when it comes out -- a breadth of interfaces so that 80 percent of an enterprise can use business intelligence, instead of a very small percentage of an enterprise.

Q: How tough was it to adjust your technology to do that?

Actually, we didn't have to adjust the technology; we just had to add delivery mechanisms on top of it. Our analytics still play a huge role within enterprises. We just needed to enable domain experts and information consumers to be able to leverage them. We've spent two or three years working on this project. We hired 32 new usability experts writing the front ends. We've got another 250 developers sitting there writing the BI back end. We spent a lot of time talking to our customers to understand how this dissemination needed to happen.

Q: Are you eliminating the hierarchy of people with doctoral degrees in statistics or some other math field so you can present the information to the people who need to know it?

You need the Ph.D.s to create the initial model. What we're not doing is dumbing down the analytics. We're just ensuring that when they get used, that people who know how to use them are creating a safe environment for the rest of the knowledge base. There's a lot of folks out there dumbing down analytics and black-boxing stuff. That's very dangerous, because models have to be retrained to notice the subtleties in the data.

Q: What does it mean to have to retrain models?

If you build the model, it's just a bunch of nodes hooked together that don't particularly know anything yet, and you need to run data through them, which creates, say, a decision tree, which is an example of one data mining model. It comes up with a decision tree that says, "When boys between 25 and 30 buy gym shoes, try to sell them gym socks also. But for a guy who's 45 to 50 who buys gym shorts, you don't necessarily want to sell him gym socks." You've got to train the model and then you leverage the model. But the model needs to be retrained from time to time because the information in the data changes. You make some adjustments in your model based on this new information to keep it fresh or even make it better at times.

Q: Does that mean that analytics is a technology that isn't going to be commoditized anytime soon?

Actually, that's exactly what I think we're doing. We're enabling an enterprise to take advantage of it but allowing them to do it in a controlled manner. But the models have to be kept fresh. To some banks, a half-percent increase in something represents millions of dollars. It's those kinds of things that you learn. You create a model that makes you better than you were, but then you gather new data, and you're able to tune the model even more.

The key thing we've been hearing is enabling the enterprise. People can be told they're empowered, but until you give them the information to truly let them be empowered, they won't believe you -- and they'd be right.

Q: What are the biggest problems companies face in trying to implement business intelligence and analytics?

The big pain I really think is organizational change. . . . I'm not trying to downplay the challenges with implementing technology, but I would wager that most of the challenges around technology have to do with culture, have to do with people, have to do with process. I know there are customers where I could go in and sell them systems that would make them more effective as a company, but their culture won't allow it to be successful. Each silo of the business is rewarded for maximizing their silo.

If you want a technology, I'd say data quality is the biggest pain. Data quality can delay an implementation of a warehouse or even a data mart upwards of six months or more. As soon as you pull data that's not accurate through and into a warehouse and report on it and give it to somebody, their trust of the new system instantly dies.

Join the newsletter!

Or

Sign up to gain exclusive access to email subscriptions, event invitations, competitions, giveaways, and much more.

Membership is free, and your security and privacy remain protected. View our privacy policy before signing up.

Error: Please check your email address.

More about SASSAS Institute Australia

Show Comments