Intelligent Enterprise

Better Insight for Business Decisions

RSS
Webcasts
Whitepapers
Subscribe
Home

June 30, 2003

A New Analytic Perspective

Decision makers throughout your enterprise need an IT architecture that serves their needs — rather than the other way around. Here is a view of emerging enterprise analytic systems that uses BI and analytic requirements as the point of origin

by Kemal A. Delic and Umeshwar Dayal

Continued from Page 2

The Architecture

The typical artifacts for an enterprise analytic architecture are estimates, forecasts, scenarios, views, and rules. They are the usual result of human observations, reasoning, and intelligent analysis. We will call computing systems specifically designed to render analytic artifacts "enterprise analytic systems." Today, these are typically data warehouses, knowledge bases, and simulations. Their objective is to make sense out of large data and information sets and pack final results into the appropriate form-factor for the decision maker.

An enterprise analytic artifact might be delivered in the form of a graph, view, document, program, interactive graphic, simulation, or model. Some of these artifacts are passive, while others are interactive. Some may be delivered from the single analytic server, while others may require grid-level computing power. Typically, the computing power and resources engaged are inversely proportional to the volume of rendered artifacts. Some business intelligence analytics software vendors are already able to deliver tens of thousands of analytic artifacts to corporate employees' desktops.

The purpose of the enterprise analytic systems ranges widely, from providing holistic views of the enterprise state and subsystems to the tasks of optimization and forecasting. The earliest and most typical deployment was in identification and reduction of various inefficiencies leading to optimization of overall enterprise performance. Highly abstracted, enterprise architecture can be dissected into event, transaction, and analytic layers, each having different purposes, objectives, and design constraints (see Figure 2).

Looking at Figure 2, interactions with customers, clients, partners, and suppliers create a constant flow of business and IT events that should be accurately and timely forwarded to appropriate event-management applications. Within IT, the architecture must address IT events that will enable the monitoring and management of IT fabrics: network, routers, switches, servers, operating systems, applications, desktops, notebooks, and personal mobile platforms.

Transactional applications aggregate, transform, and capture a fraction of this endless event stream as transactions. Business processes are also a critical part of enterprise operations. They create the host of interrelated transaction records typically guided by the established policy or automatic execution of the business processes. Business processes represent the key business chores executed daily in each enterprise. Transaction persistence and recovery are necessary features of this middle layer; these features play a key role in billing and auditing processes, for example.

Financial Services: The Leading Edge

The financial services industry is shaping up to be the first to profit from new technology and architectural developments in enterprise analytic systems, serving both clients and customers. Let's consider an example of a large North American bank that's trying to grasp the incredible volume of events and transactions, and estimate the number of potential users and consumers of analytic artifacts. This bank processes more than 5 billion transactions per year (about 150 transactions per second), which are initiated through 5,000 point-of-sale devices and automated teller machines. In its data center, more than 30 high-end servers and 1,000 multiprocessor servers support not only customers and clients but also bank personnel, who use 25,000 desktops and laptops. The bank has more than 8 million retail customers using Web banking services.

Giving advice to a large number of banking customers about how they could better manage their money in current accounts might be an example of a noninteractive, high-volume analytic artifact. Supplying business clients with different investment scenarios would be an example of an interactive analytic artifact with low-volume but high practical potential.

It will take an intricate computing structure to create, transfer, and render enterprise analytics. Algorithms will need to capture precious expertise representing the most dense and valuable intellectual property. Our guess is that data center capacities and enterprise grids will be the fabrics necessary to cope with complexity, volume, and dynamics of enterprise analytic systems. In financial services, the number of events might be in the tens of billions; the number of transactions might reach 5 billion per year; and the number of static analytic artifacts might reach into the tens of millions per year. We could see intricate interactive analytics reaching hundreds of thousands of instances per year for business customers, representing the next generation of financial services that would gradually replace simple and increasingly outmoded reporting systems.

Analytic applications operate on amassed data and information sets; these implement sophisticated (parallel) algorithms on specialized data repositories, such as operational data stores, data marts, data warehouses, and knowledge bases with the purpose of delivering accurate analytic artifacts in the appropriate form-factor. Enterprise portals are rendering engines for different user populations. Enterprise analytic systems should support better decision-making across the range from daily operational decisions to strategic and disruptive changes.

Enterprise event, transaction, and analytic subsystems serve as the business communication, coordination, and cooperation conduit with markets and suppliers. They're created on heterogeneous platforms for which interoperability standards are now sorely absent. Web services are addressing the standards and interoperability needs in an innovative way, although the proliferation of informal standards and proprietary solutions continues. Analytic system functioning is based on advances in various computing domains, especially marked by developments in data mining and business intelligence.

Technology Developments

The headwaters of most analytic technologies today are in the research of information theory and artificial intelligence, which has helped to create data mining, data warehousing, online analytic processing, and business intelligence. These technologies have combined with advances in hardware and improved software creation methods, leading to the creation of the impressive superstructures. The results have been huge, dynamic programs typically encountered in large-scale simulations and real-world modeling for weather prognosis, traffic simulations, and financial analysis.

Rate This Article

Comments:

Optional e-mail address:

As enterprise analytic systems move out of research, industry-specific requirements will begin to dictate specialization as well as new methods and approaches. (See the sidebar, "Financial Services: The Leading Edge") Advances will likely stimulate the creation of technologies that better fit the needs of enterprise analytics. We believe we'll see a distinctive analytic architecture develop, as we saw when the industry responded to the need for management of enterprise events and transactions. The new technology will support the set of analytic tools and methods to analyze the past, assess the present, and predict the future according to key enterprise business parameters.

Kemal A. Delic [kemal_delic@hp.com] is a lab scientist with Hewlett-Packard's operations research and development and a senior enterprise architect with relevant experience in knowledge management, Bayesian nets modeling, and real-time intelligent systems.

Umeshwar Dayal [umeshwar_dayal@hp.com] is principal laboratory scientist with HP Laboratories, where he leads a team performing research in data mining, knowledge management, and business process management.

RESOURCES

Delic, Kemal A. and Umeshwar Dayal. "The Rise of The Intelligent Enterprise," ACM Ubiquity, vol. 3, no. 45.

Delic, Kemal A. "Enterprise Models, Strategic Transformations and Possible Solutions," ACM Ubiquity, vol. 3, no. 20.

Grigori, D., F. Casati, U. Dayal, and M.C. Shan. "Improving Business Process Quality through Exception Understanding, Prediction, and Prevention," Proceedings of VLDB 2001, Rome, 2001.

Han, J. and M. Kamber. Data Mining: Concepts and Techniques, Morgan Kaufmann, San Francisco, 2000.

Inmon, W.H. Building the Data Warehouse, Second Edition, John Wiley, 1996.

Chaudhuri, S. and U. Dayal. "An Overview of Data Warehousing and OLAP Technology," ACM SIGMOD Record, March 1997.

Quinn, James Brian. Intelligent Enterprise, New York, The Free Press, 1992.