Guest Post: Connecting Decisions to Data: A Case Study, Part 3: Uncertainty and Assumptions

In the prior post I used a circled red question mark to label some of the decision model elements to indicate that they were missing one of the pieces of information needed to define them, either directly or because of an upstream dependency. 

Unknowns play a crucial role in the decision model: they provide a precise specification of the information resources that need to be obtained to complete the dependency chains. Unknowns are usually either values that need to be determined (externals) or relationships whose form is not known (dependencies). By way of example, the decision model for our product launch scenario has the following unknowns: 

• The value of the market size. 
• The demand curve (the relationship between the sale price vs. the percent of the available market that will purchase the product). 
• The relationship between the increase in demand vs. marketing spend.
• The relationship between unit price of production, vs. the size of the production order.  

If you’ve worked with a team to create the Causal Decision Diagram (CDD), then it can now be used by the analytics and data science team to ensure that they understand the data and related resources needed to support the decision. (We recommend setting up a CDD store as a shared resource that all participants in the decision have access to through a URL in their browser). The “business owners” (formally called Decision Customers) of the decision are responsible for the structure of the model, while the data team is responsible for providing the information needed to resolve the unknowns. The Decision Diagram is a powerful collaboration resource and facilitates transparent communication between business-oriented and technical teams. 

In summary, you may recall that Problem 1 from my first post in this series is to ask:

How can decision makers identify the data they need to support their decisions?

The answer is as described above:

The solution to Problem 1 is to develop a Decision Model and use it to identify and communicate the unknowns. 

This is the systematic method for using the statement of a decision to derive requirements and specifications for the data science and analytic work required to support that decision. 

What about assumptions?

You might have noticed that there is no CDD element type called “assumption”, even though many decisions require assumptions. The reason is that “assumption” is a characteristic of an external or other model. It captures our degree of confidence in its correctness, not in the role it plays in the decision, nor how the decision model element is defined. Rather than labelling an element of a decision model as an “assumption”, it is more precise to specify that the element in question has a high level of uncertainty. Another advantage of treating “assumptions” in terms of uncertainty is that the uncertainty introduced by each assumption can be propagated through the model to estimate the uncertainty of the affected outcomes. 

More benefits of decision models and decision diagrams

Above, I described how a decision model provides a rigorous approach to generating data requirements from the decision statement. Decision models also provide the following benefits: 

• A common understanding, or mental model that can be shared, including with new collaborators as they join the decision-making team. 
• A repository for institutional memory of the elements that come into play in complex decisions, and how these are related to each other, and to external factors. 
• A basis for optimizing the decision against multiple criteria, and determining the uncertainty and risk associated with each decision choice. 

At this point you may be wondering how all this decision modeling relates to the decision support provided by Business Intelligence. That’s the topic of my next post.