Information Impact:
|
Information doesn't have to be precise to be valuable. A rough prediction early in the quarter that earnings may tumble between 20 and 60 percent below expectations is far more useful than precise information at the end of the quarter that earnings fell 35.65 percent. Figuring out the value of your information by linking it to decisions made is an ongoing, qualitative, and, to a large degree, imprecise process. But the insights it will give you into your organization and how to improve it are invaluable.
|
Interview-based decision reconstruction is the first precision level at which you can attempt to capture the incremental value associated with incremental improvements or degradations to decisions. Decision reconstruction needs to involve the people who actually make the decisions, whoever they may be. This process creates two types of analyses:
- Decision value graphs that put actual decisions and their associated values in the context of potential decisions and their associated values
- Strategy tables representing the impact of possible changes in information attributes, the costs of those changes, and their associated impact on decision quality and decision value.
DECISION VALUE GRAPHS
Before you create a decision value graph, you need to take several steps after you've identified the appropriate decision maker(s):
1. Connect the decision made, D, with a measurable value outcome, V (recognizing that the outcome may not appear for a considerable time). For any decision that can be made, you can always ask the question, "How do you measure the relative success or failure of the decision?" For example, promotional decisions may be judged against resulting sales; reconstruction could determine that the decision to run an advertising blitz in the spring resulted in $3 million of additional earnings from sales. Similarly, manufacturing decisions may be judged against resulting product quality changes. The decision to upgrade a particular assembly process, for instance, could be associated with a change in product quality, measured in defects per 1,000 units, from 30 defects to 10 defects per 1,000 units. Finally, hiring decisions may be judged against employee productivity or on-time delivery. Regardless of the decision, or how you measure its value outcome, you can reconstruct a picture of the decision and its outcome that you can represent as V/D. (See Figure 1.)
Estimating the outcome of a decision is more difficult when multiple decisions affect the same measurable outcome or when a single decision affects multiple outcomes. For example, many decisions made on the shop floor may jointly contribute to a single measurable product attribute. In such cases, you need to either look for intermediate measurable outcomes, such as the speed or quality of a particular process, which are outcomes you will eventually need to connect to product quality; or treat all decisions as having equal weight; or ignore the individual decisions and focus on the measurable outcome of the aggregate of all decisions affecting product quality. (Remember, the DEER cycle can be evaluated at multiple levels of granularity.)
2. Identify the alternative decisions that were considered but not taken. For example, the promotion department may have considered running a super ad blitz or a more subtle campaign instead of the standard blitz that was actually run. The manufacturer may have considered upgrading all the assembly processes or upgrading the finishing process instead of the one assembly process that was upgraded.
3. Represent the decision actually taken against this backdrop of considered-but-not-taken decisions. (See Figure 2.)
4. Determine the dimensionality and internal ordering of the set of alternative decisions. If the potential decision set has a one-dimensional ordinal or cardinal ordering, arrange the alternates from farthest to closest less-than, and from closest to farthest greater-than the decision taken. (See Figure 3.) Typical examples include pricing and quantity decisions.
If the potential decision set has a nominal ordering, try to associate an implicit ordinal or cardinal attribute with the decision choices. Unless your decision was truly binary, you will discover that your seemingly nominal decision sets have some implicit ordering. For example, the decision to promote primary colored clothing, which might appear to lie on a nominal color scale, may reflect an implicit ordinal ordering based on brightness. The decision to work with one service vendor over another, where service vendors form a seemingly nominal scale, might reflect an implicit ordinal ordering based on financial stability.
If the potential decision set has an implicit multidimensional structure beneath a seemingly nominal set - for example, if a supplier decision were based on a combination of company size, track record, product quality, and relative prices - you need to identify the realistic combinations of values within the implicit dimensions and then group them by their estimated decision value.
5. Given the actual decision and its known value embedded within an ordered set of potential decisions, estimate what the value would have been for one to three alternative decisions in each direction. For example, assuming the decision to go with primary colors was worth $3 million in earnings, estimate what the earnings would have been had one gone with one or two decision increments more or less in starkness. If your preseason, 20-percent "Give Away" sale earned $2 million, estimate what you would have earned from alternative discount decisions. That's how you define the change in value per change in decision, or ΔV/ΔD. (See Figure 4.)
6. Associate some measure of variance with each of the decision value results. Although you may think it useful to associate specific outcome values with specific decisions, even if there were a directly measurable resulting value such as the increased sales due to a promotion, repeating the same decision the next week or in another geographic location or department would likely produce a different result. The amount of difference between results of the same decision is a measure of the variability of the decision's value. (See Figure 5; this is a V+/- x/D.)
A typical decision value graph for management-level decisions has a discrete decision axis with a distribution of decision values for each decision. (See Figure 6.) A decision value graph shows the value of your current decision and the incremental value of an improvement to, or degradation of, that decision as defined by the slope of the graph at the point defined by your decision. The first thing to notice is that the slope need not be constant. The value gained by one unit improvement may not be the same as the value lost by one unit of degradation.
Note that in Figure 6 the current decision is worth $3,000,000 +/-$600,000. It is considered a pretty good decision, whereas an outstanding one would be worth $6,000,000 +/-$1,200,000. A unit improvement to the decision would be worth $3,000,000 +/-$600,000, while a unit degradation would lose $2,000,000 +/-$400,000.
For purposes of aggregation and comparison, you may want to translate the measurable outcome into a common frame comparable with other decisions. This process again involves the organizational metrics you have in place. Likewise, you may want to translate the decision axis into a common frame such as worst to best. You will, of course, blur the distinction between decision quality improvement resulting from increasing vs. decreasing your decision variable, but it will enable you to talk about the average value of improving decision quality by one step.
  
|
|
|
|
|
|