What's Next for Business Intelligence?An Exclusive Roundtable Discussion
by Justin Kestelyn and David Stodder In 2002, business intelligence (BI) and analytic technology has reached the apex of importance. Whether the goal is inventory control, financial modeling, customer retention, or even national security, the creation of "knowledge" from the whole cloth of raw information is no longer of debatable value; often, it is the source of value. Even so, customers of BI and analytic software face the traditional quandaries: Do I build or buy? How do I convince my end users to share information? Is it time to invest in real-time technology? Will my software work properly in a heterogeneous environment? In this Intelligent Enterprise Roundtable, we've convened four experts to discuss directions in the marketplace: Seth Grimes, principal of decision-support consultancy Alta Plana Corp. and an IE contributing editor; Henry Morris, Group VP for applications and information access at IDC; Nigel Pendse, BI consultant and lead author of The OLAP Report and The OLAP Survey, and Philip Russom, research director for BI, data warehousing, and data quality at Giga Information Group, as well as an IE contributing editor. IE: How successfully have OLAP and reporting vendors delivered on their promises to make BI mainstream and close the loop between decision-making and operations in the last five years? MORRIS: I think there's been some progress in terms of making BI mainstream, or at least increasing the base of users for BI, particularly in the area of query and reporting. Distributing reports over the Web certainly increases the base of users who are accessing this information, and that's important. But in terms of closing the loop between decision-making and operations, a lot more progress needs to be made. In fact, I could argue that BI, which is often labeled decision support, doesn't really support decision-making at all; more can be done to provide collaborative support to improve decision-making processes themselves. Linking decisions to operations via rules engines or rules transport, necessary for a closed loop system, is an integration issue that often goes beyond BI's current boundaries. GRIMES: I agree with Henry; BI has become mainstream. Its functions and capabilities are well understood, and you can find commercial tools for just about any common analytic function. But closing the loop is another matter — many, many integration and process issues are unresolved, not to mention what constitutes best practices in linking analysis to operational decisions. For example, OLAP and reporting tools are great at telling you when something happens, but they can't necessarily tell you why it happened. If sales are down 12 percent from last year, you can't tell whether your salespeople are slackers, your products are outmoded, or you have price competition. Those what-if and goal-seeking capabilities are available only in limited sets of tools, not in the mainstream ones at all. PENDSE: I'm more pessimistic about the mainstream question. Large sales have been made, but most of them have ended up as shelfware; true production deployment levels are much lower than the vendors claim. The products are still too expensive, too hard to deploy, and not that popular with end users. BI tools may be more mainstream now than five years ago, but are they actually mainstream today? Absolutely not; they're nowhere nearly as ubiquitous as genuinely mainstream products, such as Microsoft Office. IE: To what extent has customer corporate culture been a factor in that failure? PENDSE: Many adoption challenges are corporate culture- and people-related. But the technology itself has also been part of the problem — these products are a lot more expensive than the vendors would have you believe. Furthermore, because most big deals are negotiated with the IT department, businesspeople rarely buy into them. And, often, the data involved is the stuff that's easy to deliver based on what IT already has. It doesn't reflect what the end users want. MORRIS: I agree; cultural issues are very important where decision-making is involved. If there is no organizational consensus about how decisions are made, transmitted, and applied, adoption can be a problem. For example, in the CRM area, clashes about decision-making can occur between the people on the front lines — that is, the call center people — and the people who are doing the analysis and establishing guidelines for recommendations. RUSSOM: I want to be sure that we all understand: The question is worded as if moving BI into the mainstream is the responsibility of software vendors. But I think some of you have touched on the fact that the responsibility is actually much broader; that IT and the business end users are involved as well. In the early to mid-1990s, IT simply could not keep up with the workload of creating new reports. Imagine a business manager telling an IT staffer during that time, "Okay, we've got a few dozen high-level users today. We want to go to a few hundred in the next couple of years, and eventually, a few thousand." That's a huge change. Clearly, a certain amount of mindset readjustment has been required of IT, and there has been some push-back. One of the grander, more noble aspects of BI for the masses is that it enables decision making not only for executives, but for people much deeper in the org chart. I recently met a guy who gets his reports on a CRT screen, which is located on a workbench next to the door of his shipping dock. That's how relevant the reports are to what he does every day. He's making decisions about which loads to put into which trucks at which times, and what kind of pallets he needs to order. Are these deep, strategic decisions that are going to affect the entire corporation? No, of course not, but they really do help him run his little piece of the business better. IE: Vendors are trying to increase the robustness of their analytic engines. In your opinion, what are the most critical technology features that define analytic robustness? RUSSOM: Well, robustness can mean a variety of things, but clearly scalability and availability have been hidden "gotchas" in server-based products. I'm old enough to remember when if the data warehouse was down for a few days, nobody cared. Today, that's no longer the case. Historically, some products have been available only on Windows NT or Windows 2000. I don't want to disparage Windows — I know users who have very large, highly robust Windows-based applications — but there is a perception out there that it's a limited platform. Thus, we've seen some BI vendors, such as Sagent Inc. and MicroStrategy Inc., slowly port their products over to Unix. Furthermore, recent releases from Crystal Decisions and Actuate Corp. support server clustering. So I'd say the port to Unix and oncoming support for clustering are going to improve robustness in that particular dimension. MORRIS: I agree with Philip in that availability and scalability are equally important. At IDC we regularly ask IT professionals about their major challenges in data warehousing, and availability is steadily rising to top of the list. The scalability issue is interesting because many people think about the concept only in terms of how much many rows of data can be accessed, not columns. But consider customer information: Not only are there typically many records of customer transactions, but they are usually very wide records with many different attributes — particularly when enhanced by demographic information. The ability to examine many attributes from a variety of sources and to determine which attributes are the best predictors of customer retention is a measure of analytic robustness. GRIMES: I would define robustness more broadly: A robust tool does what it's advertised to do, for the users for whom it's made. These days, a robust tool also has to integrate well into the user's computing environment, and those environments usually include heterogeneous platforms, databases, operational systems, and so on. In short, a robust system is going to play nice and not spring surprises on you. Regarding analytic robustness, many engines fall short in four respects. First, it is still too hard to bring together data from across departments or business functions, not to mention organizational boundaries. Second, most tools are insufficiently robust in terms of letting you do real forecasting vs. linear projections, and they don't let you take a snapshot of point-in-time view of data. Third, on the modeling front, there's an assumption that you're going to adapt your data to the mostly multidimensional models that the vendors support; there's very little flexibility in the kind of data models you can use. And last, from the ease of analytical use point of view, we need more tools that will show us how to get the most out of our data, sometimes in spite of ourselves — in other words, something that will really tell us what to do and not just how to do it. The tools aren't sufficiently robust there, either. PENDSE: As these responses indicate, robustness has many definitions. In British English, robustness tends to mean resilience or strength. On that score, the products have clearly improved. But if we use the broader, American English definition that implies functionality, I'd argue that the average tool is dumbed-down, considering what you could do with Express in the 1970s. There were a lot of fancy features that few people used, features the vendors have deliberately removed in the interest of making their products more mainstream. The fancy statistics or calculation capabilities that we used 20 or 30 years ago, such as Monte Carlo analysis, have largely vanished from today's products. But maybe that's not a bad thing; perhaps the vendors got too carried away with high functionality. The key idea about making a technology mainstream is that you have to make it simple. And too much robustness, if you care to define that term as functionality, may get in the way of achieving that goal. IE: Speaking of working as advertised, real time has got to be the term of the year in BI. Is real-time information truly important to customers in the real world, or have marketers simply created the appearance of that need? What are the highest value BI questions that require real-time derived information to be answered? GRIMES: There are certain very high-value BI applications, such as a credit card application scoring and real-time fraud detection, where real time is largely delivering on its promises. But on the whole, I think that real time is overblown. Unless there's some crisis going on, no strategic decision should be made in real time with real-time analytics. Rather, there is a strong need for reflective analysis, which allows the decision maker to evaluate factors coming in from other places than the analytic reports showing up on screen. PENDSE: I agree. In operational decisions such as the ones Seth described, real time is often valued. But that's not really BI. Those apps have fairly rigid rules; there's no analytic question involved. BI involves managerial decisions, where you're examining a wider range of data and looking at history; thinking about it, reflecting on it. And for that, real time has no relevance whatever. MORRIS: I think some of the confusion arises between the building of an analytic model and the application of the model. In cases such as fraud detection, you've integrated a particular model that has been built offline and is now applied to deliver real-time scoring, not actual real-time analysis. In most cases, you'll see the incorporation of analytic models into operational systems so you can apply them in real time, but in few cases will you see the need to actually construct and revise the model in real time. There may soon be ways to adjust the model in real time, but we're really not at that point yet. PENDSE: When it comes down to it, real time is a good example of vendor push, typically from certain ERP vendors. They're trying to push a capability by making it sound like a key requirement. I don't believe they're reflecting the market accurately. RUSSOM: That's a good point. Part of the vendors' marketing message is that they're trying to enable real-time alerts for operational situations. That kind of alert requires real-time data movement behind it, so we've seen major ETL vendors, such as Informatica Corp., Ascential Software Inc., and Acta Technology Inc. [now Business Objects], integrate their products with message queues or other message-oriented middleware to deliver data in real time. But that really isn't BI — it's more like information delivery that supports operational activities. IE: Let's talk about what you may also consider a case of vendor push: Mainstream BI-analytics vendors are scrambling to re-brand themselves as business performance management (BPM) companies. What new technology features will these vendors need to add to their solutions before they can truly make that claim? MORRIS: I don't think it's just a question of technology. I started getting interested in BPM about five years ago. I made the claim then, which I still subscribe to, that for any functional application area such as human resources, finance, or marketing, there is, or will soon be, a market for analytic applications that apply BI to that subject area. Soon I began to see that certain types of applications — such as balanced scorecards — are inherently cross-functional. The premise of these apps is to show the relationship among different types of metrics across business functions. Beyond the balanced scorecard idea, there are many other methodologies that look at leading indicators of financial performance as well as nonfinancial indicators, such as customer or employee retention.
|
 |
Most Popular This Week
IE Weekly Newsletter
Subscribe to the newsletter
| |
| |||||||||||||||||||||||||||||||
