Guest Post: Connecting Decisions to Data: A Case Study, Part 7 (final): The Decision-Centric Dashboard
This is my final post in this series, where I pull everything together.
In Connecting Decisions to Data I introduced a Decision Intelligence case study where a manager needs to make three decisions about a new product’s first production run. In the following post I then identified three decision levers—sale price, manufacturing run size, and marketing spend—that the manager can use to achieve their outcome: maximize profit.
In CDD Basics I showed you a causal decision diagram with causal chains from the levers to the outcome.
Simulating Actions to Outcomes and Maximizing Opportunity while Limiting Risk then explored how simulation turns business intelligence data into Decision Intelligence information.
This post shows how to put the simulation information together in a decision-centric dashboard.
Differences between data-centric and decision-centric dashboards
We can use existing dashboard technology to support decisions, but there are some differences.
1. Decision-centric dashboards have a control panel where users can input decision parameters
Most notably, the dashboard above has a Control Panel (shown on the left) that allows the decision maker to enter their choices for decisions 1–3, along with the ability to set the Externals and adjust parameters on any dependency models for which they’re uncertain. Specifically, in this dashboard, the Demand Curve and Marketing Uplift functions are relationships that can be modified, if desired, by dragging the control points (small blue and red disks). You can also specify the acceptable risk, as I’ll explain below.
In addition to the Control Panel, a panel is shown called the “Outcome and Risk Envelope” that records the changes to the decision Levers made by the user on the Control panel (lower part of chart) and shows the corresponding changes to outcome and risk values in the upper part of the chart. This provides the user with the ability to explore and become familiar with how different decisions affect outcomes and risks, and, in so doing, develop intuitions that are consistent with the evidence presented in the Decision Model.
2. Decision-centric dashboards show outcomes, not data.
The Product Sales and Manufacturing Planning dashboard and Decision Dashboard shown above are both based on the same data and information artifacts. Specifically, they are based on the four data items that fulfilled the Unknowns identified earlier in the Decision Model. The key difference stems from the different ways that this information is used. In the former case, the dashboard simply displays visualizations of the data, in the form it was originally organized. In the above Decision Dashboard, in contrast, the data is used as an input to the Decision Model and subsequently, the simulator.
This process transforms the information available to the decision maker from being expressed as the relationships in the raw data, to an expression of the relationships that are aligned with the decisions to be made (that is, the relationships between the decision’s Levers and Externals, and its Outcomes). Specifically, the “Outcomes and Risks” panel in the Decision Dashboard provides a summary view of the Outcomes that relate to the current settings on the Control Panel; the “Outcomes and Risks Envelope” panel shows the how the settings of the values from the control panel have been changed (in the above example, the manufacturing order size and the marketing spend percentage are held constant and the sale price ranges between $10 and $20. The model then processes the complex interactions between the demand curve, profit, and available marketing spend and shows the effects on the various outcomes and risks in the top chart.
3. Decision-centric dashboards show risks, and limit outcomes to those within stated risk levels.
Using Risk Intervals, a decision dashboard can also provide a graphical visualization of both maximum and expected outcomes, along with the risk and volatility of each decision. Furthermore, the Control Panel section can be used to set specified risk tolerances, and the results displayed in the dashboard restricted to those that comply with these.
Putting you the decision maker in the driver’s seat
This data-centric dashboard puts you in the driver’s seat.
Let’s recap our exploration of data and decisions by viewing them within the context of many new inventions: When new technologies are invented, the early part of their lifecycle focuses on their technical capabilities, and making them more reliable. What follows is a transition from the lab workbench and a few experimental or “proof of concept” instances, to widespread adoption. A new branch of innovation then begins, focusing not on enhancing the technology itself, but on creating an ecosystem of material infrastructure and a body of best practice necessary for the safe and economically sustainable use of the technology.
Data-driven decision support is at just such a juncture. The discipline must turn “doing data science” into “making decisions that achieve better outcomes”. The innovation to do this will not occur within data science itself, but around it, specifically in disciplines such as Decision Intelligence.
- Two unsolved problems have significantly limited the effectiveness of advanced data science techniques, most notably machine learning, when applied to decision making in all but a small number of well-known contexts. The problems are:
- There is no systematic way to transform the statement of a decision into the specification of the data and other information artifacts needed to determine which actions lead to the outcomes desired by the decision maker.
- Even when such data is available, the way the information is organized does not align with the relationship between the actions available to the decision maker and the outcomes each of these will lead to.
- The solution to problems 1 and 2 lies in two techniques, supported in practice by software tools that implement them:
- A Causal Decision Diagrams (CDDs), which make explicit the dependency chains that link decision makers’ actions and external factors to the outcomes sought by the decision maker. Unknown values or relationships in these chains identify exactly the places where data science work is required, precisely define what is required, and provide a framework for using the results of data work when it is completed.
- Simulations, which use the Decision Model to create a data set that makes explicit the relationship between the actions available to the decision maker, and the outcomes in which these are likely to result. This gives the decision maker the ability to explore the consequences of the courses of action available to them, including:
- Which actions lead to favorable, or optimal outcomes?
- How do external factors over which the decision maker has no control affect the outcomes?
- For each course of action, what is the risk of various adverse outcomes occurring?
- How does uncertainty in various sources of information captured in the Decision Model propagate into uncertainty in the outcomes?
- How volatile are the outcomes, that is, how sensitive are they to small changes in the various inputs into the Decision Model.
- Placing the simulation results on a decision-centric dashboard put you, the decision maker in the driver’s seat, providing you a rich environment for exploring the decision and optimizing the actions you take to the risk/reward tradeoffs appropriate to your decision.
Want to learn more? We teach these concepts in a number of courses, and create solutions for clients to model important decisions. Please drop me a line for a conversation or demo.