How can you make an optimization application be accepted by decision makers? The answer I gave that in my last post was to provide interactive applications. It so happens that colleagues of mine already discussed that in an IBM book Optimization and Decision Support Design Guide I can't resist quoting some of it given how it captures what I tried to express in my previous post.
Decision makers need decision support
Decision makers will not use any analytics tool unless they trust it. Trust arises from understanding the recommendations the decision support application makes and the reasons behind them. Trust is particularly important when the recommendations depart from familiar patterns, which can happen when the business enters uncharted waters. A decision maker will frequently want to test alternatives to the recommendation and to understand the trade-offs that are involved when making best use of limited resources. Also, conflicts often arise among the various targets and limits. These conflicts can result in failure of the tool to find a solution that satisfies all the criteria. It is seldom acceptable to simply announce such failures, especially in mission critical situations. Rather, the decision maker will usually want to relax some of the criteria in a controlled way until an acceptable solution is found. Finally, in a dynamic business environment, the decision makers must revisit and revise plans as conditions change.
Decision makers need good decisions
It might seem paradoxical to imply that optimal (that is, the best) decisions might not be good decisions. However, remember that optimality is measured with respect to a specific model (targets, limits, and choices). Any model represents only an approximation to the real world. You strive to make your models as accurate as possible. However, the limits of the modeling process will almost certainly lead to mismatches that can render the recommended decisions unimplementable. Thus, a good decision support tool needs to provide capabilities to ensure that optimal decisions are also good decisions. Often, an optimization model represents both objective criteria, such as physical laws or government regulations that must be enforced, and subjective criteria, such as customer service targets. If the optimization model proves to have no solution, the decision maker might need to relax some of the subjective criteria to attain feasibility Again, recalling that optimality is measured regarding a specific model, that model itself might become inaccurate because of the uncertainties that govern the evolution of the business. In that case, the optimal solution that is based on a nominal forecast might perform poorly under off-nominal conditions. This situation is called a fragile solution. In the face of such uncertainties, decision makers usually prefer a robust solution, which performs well under various future conditions. In other words, they might willingly give up some of the potential value in the nominal solution to gain better performance under uncertain conditions.
Decision makers need organizational decision processes
Only in the smallest organizations is there a single decision maker. In most organizations, multiple individuals with distinct roles participate in decision-making processes. A decision support tool needs to recognize these roles and provide capabilities that support the organization’s decision processes. As an organization grows, the amount of data and the number of people that support decision making also grow. Therefore, the decision support tool must scale comfortably to accommodate that growth. Decision support applications that use corporate information technology must conform to the organization’s IT governance. Decision support applications for collaborative decision-making processes must enable multiple users to interact with it and each other, sharing information and results.
Decision makers need to respect the people and processes in the organization
Adopting a decision support application changes the way that the organization works. Users need to learn to work in new ways, and they need to trust the new application. In addition to the software engineering tasks, you must accomplish organizational tasks in the course of developing and deploying the new application. Ultimately, success or failure often depends on the usability and user acceptance of the new tool.
This is well said. It can be summarized with this picture.
The book then goes on how to deliver applications that meet the above needs. Not surprisingly it does so using IBM Decision Optimization products, but lots of content is relevant if you use other tools and products.
Let me conclude with warm thanks to my colleagues, the authors of this book: Axel Buecker, Yana Ageeva, Veronique Blanchard, Dr. Jeremy Bloom, Dr. Mehmet F. Candas, Joao Chaves, Guang Feng, Abhishek Raman, Dr. Hans Schlenker