Eric Thomsen
October 1998, Volume 1 - Number 1
Decision Alchemy
Turning rules, data, statistical analyses,
assumptions, and goals into valuable decisions
I remember the closing scene of a “Star Trek: The Next Generation” episode in which the captain was lecturing a relatively supreme being who was about to insist on the execution of one of the crew for having unknowingly broken an exceedingly trivial rule on a planet where no rules were allowed to be broken. The captain said something like, “A true God would know when it is just to make an exception to a rule.” The crewman was subsequently released.
Whether passed along by word of mouth, referenced through a rules manual, embedded in application code, or isolated within a separate structure, rules are an important part of an enterprise’s decision-making environment. Hiring and firing procedures, product return policies, sales markdown strategies, manufacturing methods, and corporate giving are all forms of rules. But what are rules? Can rules be data driven? How do they relate to decisions?
One of the interesting things that happens at the point of decision making is that we attempt to reconcile data-driven understanding and model-driven beliefs. For example, we might not believe a report, so we instead trust our model-based assumptions more than the data; we update a data mining model in the face of new facts and trust the data more than the model.
What are decisions? Can there be decisions without rules? What implicit goals or agendas may be affecting a decision? How do we ensure consistent decisions? Can we analyze the reasons we made particular decisions? How do we know whether we made the right decisions?
The Roots of Decisions
When you think of decisions, you normally think of “actionable” decisions—that is, decisions about some course of action. But decisions can be about anything that we can describe. We can make decisions about the goals of a corporation, as is the case when an executive committee decides to change the corporate mission from software products to services. We can make decisions about the rules of a corporation, as human resources does when it decides to change the dress code to business casual. We can make decisions about predictive models, as business analysts do when they change the attributes used for predicting credit risk. In short, any aspect of an organization could be the object of a decision. Of course, we can only implement decisions on things we can change. In other words, although you could easily decide to increase sales by opening up channels in neighboring solar systems, you would have a hard time implementing such a decision.
The act of making a decision is the process or function of combining goals and predictive models. To decide that you need to lower prices of certain products is the result of a goal to maximize sales and a predictive model that relates sales to product price. A bank’s decision to deny credit to a certain loan applicant is the result of a goal to minimize loan write-offs and a predictive model that relates certain applicant attributes with the likelihood of loan default.
If there were no goals, it would be impossible to decide what course of action to take; one action would be as acceptable as any other. Without the goal of maximizing sales, for example, there is no correct decision concerning product pricing, and without a predictive model equating product prices with product sales there is no way to know which decision will be most likely to maximize sales.
Decision-Making Challenges
Decision-making challenges may arise from any number of things: the need to automate certain decision-making functions, the need to ensure consistent decisions, difficulties analyzing how a decision was made, complexities in the predictive models, difficulties interpreting stated goals, instability in the goals themselves, interpersonal dynamics, fluctuations in the predictive models, or conflicts between data-driven and model-driven beliefs. Business-rule automation tools focus on the first two challenges. Decision analysis tools focus on the third through sixth challenges. Group decision-support tools focus on the seventh challenge. And the solutions to the last two challenges lie a little further down the road.
Business rule automation tools provide a rule base within which rules can be expressed. Most business rules have the form “If condition, then action.” For example, “If this customer has ordered books from our Web site before, then make the customer aware of books similar to earlier purchases.”
The use of the term rule to describe the combination of condition and action goes back to AI lingo and expert systems. This is different from how the term is used in common parlance where it is used to refer to the condition. For example, “Thou shalt not steal” is a condition. In contrast, the term rule as used by business-rule automation software vendors subsumes the notion of consequence, reaction, or action. The combination of a triggering condition and a consequent action represents the structural component of a decision with an implicit goal.
Business rules may have multiple conditions and each condition may be represented by a variable such as “strong sales” or “poor credit risk,” which itself may be in need of evaluation based on data provided at run time. The evaluation of condition variables is often called backward chaining.
The business rules market includes such companies as Neuron Data, Usoft, and Ilog. They differ in terms of whether the rules base is an isolated data structure that, in theory, can be connected to anything, or whether the rules base is a specialized database structure made to interact solely with database data.
Typically, the rules base connects to transaction systems and helps to automate decision-making processes that were once human operated. In other cases, a rules base may tailor Web interfaces based on customer attributes. In general, the goals are fixed because they are implicit, and although the condition statements may be phrased in terms of functions that need to be evaluated at run time, the parameters of the functions tend to be fixed as well. In other words, whether a particular person is or is not a bad credit risk is a function of the predictive model in place, and although the model is dynamically evaluated, the coefficients of the model and the variables included in the model are fixed.
Looking ahead, I think we will start to see self-modifying rule systems that continuously monitor the world to see if it behaves as predicted, and when it doesn’t, the predictive models it used to make rules will change. In the process, systems could try out different scenarios or predictive models and analyze how well the system would have fared under each scenario. I would also like to see rules bases connect to OLAP tools wherein the rules base is the source of cost allocation rules used in the OLAP system. Although OLAP tools provide a sophisticated calculation environment, they would benefit from an organized method of defining and managing rules.
You should also be able to deduce rules given goals and predictive models, which brings me to the next major category of decision-making software: decision analysis. The need for decision-analysis software kicks in where decisions are based on multiple predictive models with complex measures of uncertainty and where the goals themselves are variable. This functionality appeals to executives in an organization. Decision analysis is closely related to operations research where several mutually exclusive goals and shared scarce resources exist. The trick is to maximize global properties such as profit, stability, or happiness.
There are quite a few companies in this area, including Lumina Decision Systems, Logical Decisions, Strategic Decisions Group, Applied Decision Analysis, and Decisioneering, as well as societies such as INFORMS Decision Analysis Society. (You can find links to these and others at www.dimsys.com ) Some of these companies focus more on the challenge of working through a system of multiple predictive models using such techniques as decision trees and sequential diagrams. Frequently used features of these tools are probability functions or distributions, which take the place of single value estimates for predictions that are used to drive decisions. Other companies focus more on resolving multiple conflicting goals. With these products, you might assign weights to the various goals as a means of determining an optimal decision. All these products focus on improving the auditability, consistency and correctness of decisions.
Finally, there are challenges that are more interpersonal or political. For example, a bunch of managers may be trying to arrive at a common decision on whether to fire 300 people or increase sales enough to justify keeping them on, and each manager has his or her own agenda making it difficult to know what people really think during open brainstorming, discussions, and voting.
One way to overcome these challenges is to provide an anonymous electronic meeting environment where people can present, discuss, and vote on ideas based on their merit rather than on the identities (and associated interpersonal dynamics) of the persons involved. Companies such as Boeing that have tested these sorts of environments found them to provide a tremendous savings in time and energy as well as a significant improvement in decision quality. Terms normally associated with software tools that provide this sort of functionality include groupware, group decision-support systems, and corporate memories.
By fusing the combination of group-decision systems, decision-analysis frameworks, and business rule managers with data-driven tools for understanding, such as OLAP, data mining, and visualization, you can create transform your analyses into timely, consistent, and accurate decisions. That last step—the decision-making step—which lies at the juncture of a data-driven and a typically human model-driven information loop, presents far more challenges than can be met by traditional data access, analysis, and reporting tools.
Erik Thomsen is an author, lecturer, researcher, and consultant
focusing on OLAP and decision-support applications. He is cofounder of the
Cambridge, Mass.-based consultancy Dimensional Systems and author
of the book OLAP Solutions (John Wiley & Sons, 1997). He wrote the
Decision Support column for Database Programming & Design. You can
reach him via email at erik@dimsys.com.
