I was excited to be asked to preview the latest work of James Taylor. After all, it's not often that an ex-rock-and-roller turned humble business intelligent (BI) columnist gets the opportunity to hobnob with such an accomplished musician. Ah, I remember Sweet Baby James like it was released yesterday. James Taylor and Steve Miller - now you're talking serious 70s! Whoops, wrong guys ... Never mind.
It might not ensure their induction to the rock-and-roll hall of fame, but if the 25 testimonials promoting the book are any indication, Smart (Enough) Systems by James Taylor and Neil Raden should create a stir in the BI industry.1 And my sense after reading the book is that T&R have indeed made a noticeable contribution that can start to guide BI to a next generation of sophistication and utility.
Smart enough systems (SES) help automate decisions that drive day-to-day business processes. If, as articulated by Michael Porter of HBS, "The essence of strategy is choosing to perform activities differently than rivals do," then decisions become fundamental enablers of business strategy. Operational decisions - high-volume with low individual value - are under siege with demands for precision, agility, consistency, and cost-effectiveness. With the ever-increasing pressures of competition, according to T&R, operational decision-making must treat decisions as corporate assets - strategic, managed, visible, reusable and improving. SES focuses on high-performance, compliant, real-time execution of operational decisions, emphasizing agility and evidence-based learning through data and analytics.
If SES automates decisions, enterprise decision management (EDM) fundamentally enables such applications. The basic EDM process, developed and operationalized by Fair Isaac, uncouples decision logic from business processes, providing the means to manage decision-making deployment and change. Additional steps in EDM include prioritizing decisions through ROI, establishing automated decision services, driving results from analytics, and adapting/optimizing decisions over time. In short, EDM is a systematic approach of automating, optimizing, and managing operational business decisions that strives to maximize precision, consistency, agility and speed for minimal cost. Management in an EDM context means treating decision-making as a business issue and a business asset, focusing on accomplishing an end (taking an action) and on supervision to ensure that improvement and optimization of decisions are ongoing and proactive.
Smart (Enough) Systems is divided into two parts. The first, a shorter part devoted to both business and technical readers, discusses SES and the EDM concepts that drive them. The second provides a more technical discussion of the techniques and technologies that underpin EDM. I found a fair amount of redundancy to the reading that allowed me to skim several sections, effectively reducing book length by about 50 pages.
The meat of Smart (Enough) Systems - the justification for purchasing the book - lies in chapters five through nine that clearly outline the requisite steps to successfully implement EDM. Chapter five, Data and Analytics, differentiates from Competing on Analytics by its focus on BI engage - seeing intelligence and analytics as part of the larger EDM architecture that exploits the insights to make decisions automatically.2 "In EDM, simply knowing that something is true is not the focus; being able to do something in response to that knowledge is." The authors do a good job defining the logical characteristics of analytics models, as well as categorizing major analytic techniques. They then successfully translate those concepts to demonstrate the workings of models in practice, illustrating, for example, with lift charts and model scorecards that detail how predictions are derived. Supporting the analytic models is a data integration platform architecture that embellishes the Corporate Information Factory with components necessary to build, deploy and automate modeling in an EDM framework. Once in place, that data platform supports the model development life cycle - from requirements to deployment and control.
Chapter six, Business Rules, is an examination of the EDM component that ensures intelligence drives systems and processes - that decisions are linked reliably and automatically in services that fuel them. The exhaustive treatment presented here is distilled from both the experiences of the authors and the successful conceptualizations and deployments of Fair Isaac Corporation. Business rules derive from regulations, policies, expert knowledge, existing application code and analytic models. Driven from the premise that decision logic should be separate from application logic, Taylor and Raden discuss the workings of business rules management systems (BRMS), software for managing and executing business rules, in painstaking detail. From if-then logic, through rule syntax, decision flows, rules sets, decision trees and execution modes to semantics, metadata, model-driven designs and standards - Smart (Enough) Systems articulates the concepts. The authors also address business rules technologies, comprehensively covering repositories, design editors, maintenance applications, rule templates and deployment management. In the end, according to Taylor and Raden, a commitment to business rule management can assure the independence of business logic and process, allowing a better corporate understanding of the business while promoting control and agility in decision-making processes.
My favorite topic of Smart (Enough) Systems, the one that probably cemented a favorable impression, is chapter seven, Adaptive Control, which addresses the improvements and revisions to decision services over time. Adaptive control starts with the recognition that decisions optimized at one point might not be optimal at another, a consequence of changing business conditions, evolving strategies, new insights, accumulated evidence, etc. The organization benefits from adaptive control through learning facilitated by the mathematical sciences techniques of optimization, simulation, portfolio management and randomized experiments - all topics of interest to The OpenBI Forum.
One simulation from Fair Isaac noted by the authors projects profitability curves for different actions over time, tracing the different long-term profiles, each of which might be optimal in certain circumstances. The champion/challenger approach of testing incumbent approaches against viable alternatives through controlled experimentation has proven successful with many evidenced-based management companies, notably Internet juggernauts Amazon, Google, Yahoo! and eBay. Monte Carlo computer simulation techniques are now often supplanting pure mathematical optimizations for many statistical problems, while the efficient frontier from portfolio theory offers applications for business decisions. Packages such as SAS, Matlab and R that support statistical and mathematical adaptive control techniques are increasingly popular in business.