Building Buy-In for Decisions

To gain the best possible buy-in among stakeholders, it is best to include them in not only in evaluating choices but also in defining the decision criteria. The criteria drive the rest of the decision process so it is important to spend adequate time letting decision makers own the criteria and the decision by letting them create it. If decision makers are expressing a lot of concerns about a decision making process, spend 30 minutes to an hour letting them express their concerns and make sure they are documented in a parking lot with action items for when and how they will be addressed. When people offer their concerns, be prepared to ask them for their recommended solution to the situation. By making people the owners of their own decision process, they will gain confidence that their input not only is valued, but that it has been clearly articulated as part of the decision making process.

Decisions by Definition are Subjective

It is important to remember that there is no such thing as an objective decision. All decisions by definition imply some judgment and because judgment is subjective, all decisions are subjective. The best we can do in improving upon our decision processes is to make the subjectivity explicit so that we can evaluate, right or wrong, exactly what the thinking was that led to choices. This creates a structured baseline for improving decision processes in organizations. Without it, decisions are mostly comprised of ethereal table conversations.

Just because all decisions are subjective does not mean that you cannot create a consistent and repeatable process for making decisions. It also does not mean that you cannot use objective data in decision making. The issue is that objective data must still be interpreted for its value to the decision and that is a call based on subjective judgment.

How many people are needed to make a good, valid decision?

People often ask, “What is the right sample size of participants to make the decision credible?” Decision Lens does not use statistical processes, but instead a mathematical process for calculating priorities, so sampling rules of statistics do not apply. With that said, garbage in is garbage out so it is important to attempt to gather the right mix of people to represent the stakeholder positions and expertise required to make informed choices. From a group dynamics standpoint, it is hard for groups of larger than 15-20 voting members to have a meaningful discussion while using Decision Lens. Our recommendation is that participants should include no more than 15-20 voters, but as many other experts as are needed to cover the breadth of the problem. The more stakeholders who participate, the more important it is to have a skilled facilitator to foster a collaborative environment.

The Magic Number Seven Plus or Minus Two

Student question:

I am currently writing my diploma thesis at the University of Karlsruhe, Germany. For a certain subproblem, I am using the AHP to obtain the criteria weights. Now I want to calculate the consistency ratio. Unfortunately, in my case n=16. In all the publications I could find, there are only given values for R.I. up to n=10.Do you know of any publication or other source I can get that value from?

Thomas L. Saaty answer:

Thank you for writing. You may not know it but with the AHP one should not compare more than about 7 elements plus or minus 2 at a time. Thus you should not run into a situation requiring knowing the Random Inconsistency (R.I.) Index for 16 elements. Why? I think the attached paper will help you understand. No system that is made up of more than a few subsystems can function consistently because as more and more subsystems are introduced, a small malfunction in each subsystem would, due to the overall inconsistency, collapse the system.

There is also a grouping method you could use when you need to compare many elements with respect to a single parent. Divide the elements into smaller groups of no more than 7 elements, where the elements are homogenous in each group, that is, close enough on whatever property they are being compared with respect to that a judgment of more than 9 is never required. Put a common element, as a pivot, in the first group and include it in the second group as well. Pairwise compare the elements in the first group and the elements in the second. Divide every element in the second group by the pivot element’s weight in the second group and multiply all elements' weights in the second group by the pivot element's weight in the first group. Continue this process selecting a new common pivot element to place in groups that are contiguous. All the priorities are correctly linked in this way. Normalize them to get a final overall set of priorities.

Reproducing Formula Results with AHP

Critics of the Analytic Hierarchy Process (AHP) often mislead themselves and others by constructing examples that they claim show the AHP draws false conclusions. The AHP has clear requirements for both the hierarchical structure and the priorities in the structure. It must be shown that any model that finds the AHP does not give correct results has first met these requirements. Problems must be correctly structured and priorities correctly set in the AHP model to get results that match expectations. The most common error is the belief that the AHP should be able to reproduce results in some magical way from some specific formula without including enough data in the model to truly represent the situation. Scientific formulas usually involve combinations of mathematical operations such as adding, multiplying, and raising factors to powers and they have been devised through a process of experimentation and adjustment to make predicted results fit observed outcomes. Remember those mysterious constants that crop up in some physics formulas? Figuring out what the constant has to be is usually based on a large number of experiments and that is part of the adjustment process. AHP models can reproduce the results of many formulas if one understands the principles of AHP well enough to set up the model so it properly incorporates known data. We examine a few of these examples and show how they should be properly modeled in this rebuttal paper that appeared in the Proceedings of the 2004 Multi Criteria Decision Making Conference.