This inaugural interview of Inside Analysis features the CTO and Senior VP of SAS Institute, Keith Collins and host Eric Kavanagh discussing what’s happening in the industry, where it’s all heading, and what’s new at SAS.
Eric Kavanagh: As many of you may know, SAS is a multi-billion dollar privately owned software company, focused on high-powered analytics (and various other things as well). Over the years, I’ve heard lots of good things about working at SAS which is reflected in the extremely high employee retention rates. What are the key success factors that you credit for creating a culture that has yielded such success?
Keith Collins: I guess, Eric, I’ll start with the fact this year we got first place in Fortune’s Best Places to Work for the second year in row. I’m even more proud about the fact that we’re now the best place to work (or at least in the top ten) in the world in 14 different countries. That says a lot about what we’re trying to achieve with our culture. There are two main factors.
One is, building a framework where all the employees feel like there’s trust. That trust figures into everything you do: how efficient, effective you are at collaboration and how you deal with your customers. Secondly, It allows us to be very, very customer-focused; this low turnover rate puts us in a position where our relationships with our customers have longevity. We’re able to build a sense of trust and commitment with our customers, too, and it’s a very good dividend.
Eric: Let’s talk about what’s happening in the industry now in terms of computing. Some folks are talking about super-computing or high-powered computing. How do you define that, and what are you doing with it at SAS?
Keith: A lot of the patterns that we’re pursuing have been known for a long time. People talked about grid, now they talk about cloud. Let’s talk about high-performance computing, specifically for SAS. By high-performance analytics we mean all the ways you leverage all these different architectures that are converging so quickly. You can see the price point for storage drop significantly. The opportunity to learn from the data is quite amazing, if you have the ability to scale your analytics to get through all of that data.
SAS is doing a tremendous amount, for example, in understanding and calculating value at risk in the financial sector, which with the meltdown of the financial sector is unusually important. We’ve changed that computation so that it changes the business process. Can you determine what the risk is before you take that risk? Or in retail, how do you optimize your markdowns? If you can do that during the planning cycle and not just when you’re actually marking down the material, you can save millions or billions of dollars, in some cases.
What’s important, I think, for people to understand is that it’s not just one type of pattern or approach. In the case of loan deficit defaults, the simple first part of the problem is to divide and conquer. Put it on a grid, split it naturally, and send it to a large grid instead of an SMP box. You go from 96 hours down to 4 hours. And you cut your costs, so you can get in on cheaper computing hardware. We cut the price for one customer on their hardware costs by 80%.
The industry is changing so I think you’ll see a lot of appliance-based approaches to this problem set in the near term.
Eric: Are you focusing heavily on multi-core and on parallelizing your applications?
Keith: We are. And what we’re trying to make sure we do is that we understand each pattern: when do you use a large grid, and when do you leverage the new high-performance of having lots of cores on the same blade that all share the same memory?
Eric: What are those lines of demarcation? How do you know whether to go for this solution or that solution? Can you offer some sort of overview of the key components to keep in mind when deciding which approach to take?
Keith: The simple demarcation is those things that have natural partition: time, geography, some dimension where there’s no dependency across those dimensions. Then you can easily divide and conquer.
Eric: There are a few other big factors out there. One is the Service-Oriented Architecture (SOA). And the other is the value of open-source, like Linux, for example. What is your take on the rise of Linux as an operating system and the value of open source, at least at the operating system level?
Keith: Let’s start with the value of Linux itself. What’s really happening is that open source has taken what’s already been proven in some commercial instances, where a pattern has recurred and turns it more into a commodity. So it’s just part of a normal business cycle.
Now, what that enables is the next level of opportunity and innovation, because if you reduce the cost basis of an operating system, now how can you extend that operating system to do even new or more valuable things? Or how to build things on top of that? If we take all that back to Linux, you see that it’s good from a technology sense to see a reduction in complexities, fewer operating systems to deal with. Now more things can happen in the industry to build on top of that value.
Eric: Open source provides a bigger foundation upon which to do even more interesting things when you get right down to it. And that’s the same thing with standards. This brings us to the concept of software as a service, (SaaS). There are obviously a lot of different standards out there, and some of the SaaS vendors have designed their system apparently to be backwards compatible or at least, to be retro-fitted to existing SAS applications, such that SAS customers can still use their models, just in a new environment. So, and of course the price point is strategically set at a fairly low level. What are companies like SAS, which have made a tremendous amount of money from fairly expensive license fees, going to do about this sort of growing threat posed from software as a service?
Keith: I guess there are a few things to talk about here. One is, there are multiple players out there who have worked toward providing simple frameworks for implementing simple models. And yes, are we concerned and aware of those? Absolutely. You may also be aware of the work we’ve been doing in-database analytics with Teradata, Netezza and DB2, which is the traditional implementation of models. We’ve been on a fairly active development cycle to enable very complex models to be hoisted into any of these environments, whether it is inside a database, inside a call center application, or inside a web-service application. We’re really trying to change the game that makes it much easier to do the straight transformation from all the heavy analytics and modeling to how you actually deploy that in operational systems and use the same techniques regardless of location.
Eric: Basically, you’re talking about embedding your power of analytics into all manner of operational systems, thereby spreading your footprint within the organization and letting people in operational roles leverage the kind of predictive capabilities that you offer, is that right?
Keith: Very well said. Most people talk about software as a service as if they’re going to rent it, just use it once, and it’s gone. The reality of any user agreements, for the most part, is that they are usually a minimum of a one-year deal. We’ve been, if you will, leasing SAS software for 35 years. So this is not a new model for us. The software as a service model has helped validate the value proposition of the way SAS has been doing business, and I’m quite excited about that.
Two, is we have to be very, very careful and understand that software as a service for transactional operations makes lots of sense. But when you’re talking about business analytics software, and you’re talking about analyzing a few terabytes (maybe a petabyte) of data, hoisting that into the cloud isn’t something you do in an automatic period. It is important to talk about different paradigms when we talk about software as a service and business analytics. Now what’s interesting is, our business of being a host and provider is growing at well over 20% a year. You’re right, the demand for people to have the ability to host these types of applications is definitely growing.
Eric: It seems that the excitement has moved beyond stand-alone BI into the realm of real time operational connection. It’s literally closing the loop; where the power of analytical software is being embedded right into operational systems to help people either make decisions or change processes. Is that your impression, too?
Keith: To answer that, yes the broader market waking up to this opportunity, moving beyond just understanding data and transactions. That’s the big market shift. And so to say, software-as-a-service for just for business analytics, that’s not where our customers have been pushing this. They want fraud management solutions. They want claims fraud. They want network analysis. They want things that are closely tied to the business problem, which comes back to why it’s important for us to tie into the operational process, to be able to turn what’s learned and understood into real action in the business.
Eric: What do you see in terms of the impact of renting analytics instead of building systems and generating them yourself?
Keith: I think it probably depends on your definition of renting analytics. If you mean renting a markdown optimization solution, or a loan fraud analysis solution, or any money laundering solution, then the answer to that is, yes. If you want to rent a logistic regression, not so much. Because once again, it goes back to the data quality, the data provisioning, understanding what the business process is. Once again going back to how you close your loop and turn this into an action. There are a small number of people in the marketplace who can say, “Oh, I know exactly how I’m going to apply this logistic regression or this non-linear program to get the appropriate outcome. “ It sort of goes back to the statement about how we’re moving beyond our initial buzz. “Oh, we’re going to have BI!” Well, beyond the service, yeah, I can produce a report, but if I can’t get data to it, so what? So I think if your definition is, are we going to rent analytics? Yes, I think people are going to rent business analytics solutions, not renting tools, I don’t see that as a real viable solution. It’s sort of a blend between, is it really Software as a Service or is it a package solution? A lot of people want to debate that. All I really care about is that we’re able to help make our customers more efficient, reduce their cost basis, and deliver their results. They want help in skills, for example, data management, data quality, which is a big part of feeding any real business analytics solution. And they want help in getting enough skills to get into this business analytics space — who can help me understand what I might model? Who understands my business?
Eric: I’m glad that you brought up integration, because I have heard that SAS, which used to have its own integration stack, has now moved that under the direction of Data Flux, which, of course, is a SAS company. Can you talk at all about what went into that decision, and what you think about it?
Keith: I’m sure I can. We have this great value proposition of understanding data, data integration, and data quality. Part of that vision was the original acquisition of Data Flux. Then recognizing that we had first-in-class data quality that we could build out in a broader vision to establish a brand and a sales channel into IT, where IT traditionally hadn’t considered SAS. One of the challenges is to make sure that we continue to help our large SAS customer base and understand that it isn’t something different, that we haven’t actually dropped them in the dust. We have to be careful when we take that kind of branding strategy.
Eric: Speaking of data integration, there’s a lot of activity in the federation space these days and the virtualization space. How big a factor do you think that these technologies are along with resurging technologies like changed data capture?
Keith: I think we’re just now beginning to see that resurgence; and quite frankly, it’s going to be driven by the previous conversation around software as a service. As the market adopts more of these specific solutions, both operationally and strategically, to run their businesses, once again this idea of one version of the truth being one physical location can’t survive. You’re going to have lots of copies of your data, both in your organization and outside of your organization. How are you going to apply the concepts of master data management to synchronize that? How are you going to use data federation to help you have a seamless view across those? These will be hugely important issues.
Eric: The Bloor Group is working on research on the component technologies of an Information Oriented Architecture, which is to data what the Service Oriented Architecture is to services. The created of the IOA, Robin Bloor says that MDM is kind of like the registry to an SOA. It’s like the data component, or the data-focused registry in the IOA type environment. Does that make sense to you, or what do you think about that?
Keith: I think I’d have to study a bit more to be honest with you. It is very, very true that people are forgetting that when you start getting into more complex solutions that are hosted, whether as software as a service, or platform as a service, or infrastructure as a service, that provisioning becomes a crucial part of the activity, which means that the things that Data Flux has been doing around the data governance piece, the business data glossary, the master data management become critical to enabling you to continue to run your business as business within your budget. How do you govern that process? A prime example: how do you bulk load or bulk extract from Salesforce.com versus just having a single transaction? They’re getting there, but it’s just like it was in the early days of SAP, even an on-site application. It had lousy bulk load, bulk movement, if you were really going to do some serious analysis of the data that was in the system.
Eric: Right. No question. You can use things like CDC and other technologies, federation, virtualization, but they’re tricky things to do. There’s just no question that it really requires a lot of patience, good developers, good strategy, and you need to be very careful about things if you’re going to use those technologies effectively.
Keith: I’ll finish up with this saying that the whole change data capture will have a significant research aspect that sort of blends in and probably will be just part of what we talked about in MDM in the synchronization process.
Eric: Here’s the last question, and this is sort of near and dear to my heart. I’m a big fan of Apple, but I’m also a big fan of the PC world. They do different things well, and of course, Steve Jobs has rolled out this thing called the iPad, which is just stunning. At the same time, it was a tad frustrating when I realized that they intentionally did not support Flash by Adobe. What’s your take on the form factor effect of the iPad and the standards-related issues that it brings up yet again?
Keith: Well, I think if you’ve seen the simple shift from Netbook to tablet is huge: the instant on, the delivery of the information, the interaction. I’d have to agree to some extent with Steve Jobs that what he was wanting people to understand is that the interaction of gestures and everything that changed the game wasn’t initially really part of the Flash paradigm. At the same time, there is a lot you could do for vision with Flash that would work appropriately. I hope that there will be some change there because I do think it’s an annoyance that I can’t appropriately view about a third of the websites I go to.
But your main point: is the tablet going to change the game for information delivery and the way people consume information, and where they consume their information? Absolutely. Our customers are already talking to us about integrated applications. Once again, how do we put these analytics into action so that those on the sales floor know the customer and know the best way to help that customer find what they’re most likely to care about. That moves decisioning out of the back room directly onto the floor. We are just going to see that pattern recur over and over, in health care, in the retail store, or on a plant floor. There’s going to be this new expectation driven by, quite frankly, the user and consumer environment. Just you and I using it at home changes our expectation of what we want the business to do for us.
Eric: I’ll just throw one last question. What does the future hold at SAS?
Keith: Well, one of the reasons that I like being at SAS is the opportunity to apply analytics across every industry. This shows that problems can transfer across those industries, so what we do in understanding breakout fraud in LA County for childcare can become an influencer for what people are doing on the web or telephone companies. There is never a dull moment at SAS because the opportunity to apply analytics to business problems is changing the business game. It has never been like this before with the ability to drop huge volumes of data through high-performance computing and to get it to devices like the tablet.