Decision analysis in projects

Decision analysis helps people realize faster, more confident decisions. Uncertainties abound, and this general problem-solving approach explicitly incorporates judgments about uncertainties into the analysis.

The techniques are applicable to all types of project decisions and valuations. The feasibility study is often the first analysis of a project. Most decisions involve committing resources, such as time and money. During construction or development, there are many decisions to be made. As illustrated in Figure 1, forecast models can be continually updated as the situation moves from project management to operations to eventual abandonment.

Retrospective

This series has described the approach and principal techniques of decision analysis; the sidebar catalogs installments.

After constructing several decision analysis models, most people see a pattern emerging. Typical analyses follow these

steps:

• Determine the decision alternatives.
• For each alternative, identify the possible outcomes and the outcome values.
• Assess the probabilities (or distributions) for the various chance events.
• Solve for (usually) the expected monetary value¹ (EMV) for each alternative.
• Implement the best alternative.

Figure 1. Decision Analysis Modes

This final installment describes quick-and-dirty decision models; common errors in implementing decision analysis; strategies to help ensure success; mitigating and avoiding risks; and analysis tool selection.

Quick-and-Dirty Decisions

A common misconception is that decision analysis is time-consuming. However, many of my evaluations actually take less than ten minutes. Once you are clear about the best course of action— stop the analysis! No further value will be added. Decisions that warrant more extensive analyses have one or more of these characteristics:

• Outcomes are difficult to value (requiring detailed cash-flow modeling).
• Outcome values are significant and warrant serious attention to the decision.
• The best alternatives are possibly similar in value.
• There are too many variables, perhaps with correlations, to process in one's head.

Common Simple Situation. Suppose you have created a base project plan. Upon further reflection, you now recognize a contingency that could affect project cost (and/or schedule). A simple tree-like model of this contingency is shown in Figure 2.

Figure 2. Simple Contingency

Brainstorming about the possibilities is an important management function. Corporate planners call it S.W.O.T. analysis: identifying Strengths, Weaknesses, Opportunities², and Threats. Note that contingencies can be either unfavorable (threats or risks) or favorable (opportunities). A large project may have hundreds of identified risks and opportunities.

After a contingency has been identified, the manager should seek actions to exploit or mitigate the situation. This is where project management can be proactive rather than waiting to become reactive. The actions may change the probability that the contingency event will occur; affect the project cost distribution (i.e., impact) should the contingency event occur; or affect both the probability and the impact of the contingency (least common).

Contingencies often have several abatement (or exploitive) candidate actions. An action can preempt the contingency event. More often, actions result in only partial control. Flexibility and contingency plans are among the ways to reduce the impact of contingencies that do occur.

Following is a typical situation where a quick-and-dirty decision model is adequate.

Example. Suppose your project involves building several electromechanical instruments. Alignment is critical for certain components mounted on a base plate. The intended aluminum stock for fabricating the base plate is easy to work but may prove too-flexible and nondurable in the application.

Your team assesses a 12 percent chance that the planned aluminum base plate fabrication will be inadequate. There are three upgrade alternatives shown in Table 1.

You could upgrade to one of these other choices now or see how the originally planned aluminum stock works out. Assume that if an initial base plate is not satisfactory, you will have sufficient information then to know what solution is required.

We can use Figure 2 as a template to express judgments about the risk and impact of the contingency. Figure 3 shows how the contingency could be appraised. The expected value (EV) cost of the contingency is $1,200. Whatever alternatives we consider must reduce the EV cost by more than the cost of implementing action. Decision analysts call this a value of imperfect control³ problem.

An Alternative. Your company can decide now to use thicker aluminum stock in the initial fabrication. This will cost $500 more initially. Is this prudent? Since the $500 is less than the $1,200 EV cost of plate inadequacy, the action alternative passes the first feasibility test.

Figure 4 shows the decision model. You want to evaluate whether this conservative approach is a justifiable precaution.

The thicker base plate reduces⁴ the contingency probability from 12 to 7.2 percent. If it is inadequate, then the probability that Premium Stock is needed increases from 1/2 to 5/6. Similarly, the probability of Need Casting increases from 10 to 16.7 percent. In this case, the candidate action (Use Thicker Stock Initially) affects both the probability and the impact of the contingency.

Figure 3. Contingency Model

Figure 4. Decision Model

The EV cost of refabrication drops from $1,200 to $960 when the thicker plate is used initially. However, using the thicker plate costs $500. Thus the comparable EV cost of using the thicker plate initially is $1,460. Therefore, the originally planned aluminum stock is most appropriate⁵. The Thicker Stock alternative is a worse option, having a $260 higher EV cost.

Thus we obtain a logical, defensible basis for decision. This type of quick-and-dirty decision analysis fits many situations and should be in every project manager's tool kit.

Common Implementation Errors

I often assist clients with rationalizing their evaluation practices and decision policy. Some frequently observed shortcomings include:

• Confusion about the company's decision policy. Often, there are unvoiced criteria and hurdles⁶.
• Using decision criteria that do not correspond to the company's objective.
• Using higher hurdles in an effort to establish a conservative risk policy. A better approach is to establish a company's utility function.
• Not assessing incremental project value. A common error, for example, is including fixed overhead costs in an evaluation.
• Biasing against long-term projects by using present value discount rate that is too high.
• Recognizing sunk costs and benefits (beyond their possible impact on contracts and taxes).
• Assuming chance variables are independent.
• Not defining the problem properly. Treating a symptom rather than a root cause. Providing the decision maker with the solution to the wrong problem.
• Not being objective in individual risk assessments. Some cultures or disciplines are typically conservative or optimistic. This degrades the integrity of the analysis and usurps the risk management function.

Fortunately, each of these shortcomings is easy to remedy. The sidebar outlines some steps and situation characteristics that can help ensure the success of implementing decision analysis.

One of the key benefits of decision analysis is communications. The process, by its clarity, often proves effective in keeping teams focused and directed.

Mitigating and Avoiding Risks

It is an unhappy fact of life that there are usually many more things that can go wrong with a project than can unexpectedly go right. Although this section focuses on the downside, the project manager should be constantly vigilant for potential opportunities.

There are three principal ways to manage risk: avoidance, including waiting; reduction (mitigation); transfer. The following outline describes some of the possible ways you might find useful in mitigating or avoiding risks:

Portfolio Risks

• Share risks by having partners (dilution or diffusion).
• Spread risks over time.
• Participate in many ventures.
• Group complementary risks into portfolios.
• Seek lower-risk ventures.
• Specialize and concentrate in a single, well-known area.
• Increase the company's capitalization.

Commodity Prices

• Hedge or fix in the futures markets.
• Use long- or short-term sales (price and volume) contracts.
• Tailor contracts for risk sharing.

Interest Rate and Exchange Rate

• Use swaps, floors, ceilings, collars, and other hedging instruments.
• Restructure the balance sheet.
• Denominate or index certain transactions in a foreign currency.

Environmental Hazards

• Buy insurance.
• Increase safety margins.
• Develop and test an incident response program.

Operational Risks

• Hire contractors under turnkey contracts.
• Tailor risk-sharing contract clauses.
• Use safety margins; overbuild and overspecify designs.
• Have backup and redundant equipment.
• Increase training.
• Operate with redirect and bail-out options.
• Conduct tests, pilot programs, and trials.

Analysis Risks (reducing evaluation error)

• Use better techniques (i.e., decision analysis).
• Seek additional information.
• Monitor key and indicator variables.
• Validate models.
• Include evaluation practices along with project post-reviews.
• Develop redundant models with alternative approaches and people.
• Involve multiple disciplines, and communicate cross-discipline.
• Provide better training and tools.

Analysis Tool Selection

This section provides guidance about whether stochastic analysis is even needed and, if so, what type of calculation tool is most appropriate.

Deterministic Models. This series has been about stochastic decision models. However, conventional deterministic models will suffice in some situations, such as when the decision is:

• Not very important
• Low risk: reasonably assured volumes, costs, and prices
• Preliminary or doing a base-case evaluation
• Clear choice of the best alternative (perhaps the analysis forecast is for a budget plan).

Stochastic Models. People often ask: Which tool is better—decision trees or Monte Carlo simulation? The answer depends upon the problem. Neither technique is inherently superior. However, each technique is better at solving certain problems.

Table 2 lists situation criteria and check marks showing, with all other things equal, whether decision tree analysis or Monte Carlo simulation would be preferred. There are a few other calculation methods, usually for special situations, but trees and simulation remain the workhorses of decision analysis.

Summary

What are the characteristics of successful executives? Surveys consistently show decision-making ability at or near the top of the lists. Perhaps no management activity is more important. Surprisingly, little training investment is made toward developing this important skill.

Most decisions are about resource allocation: where do we put our time, money and other resources so that they will do us the most good? Decision analysis is the discipline that helps decision makers choose wisely under uncertainty.

Probability is the language of uncertainty. Decision analysis provides the only logical, consistent way to incorporate judgments about uncertainties into an analysis. Expected value is perhaps the most powerful management concept since the invention of the organization hierarchy.

Good decisions increase the likelihood of, but do not guarantee, good outcomes. However, making good decisions over the long term maximizes the likelihood of good progress toward the organization's objective. People using decision analysis can sleep well at night knowing that they have made the best possible choices under the circumstances.

There is a strong bias for action. There is no reason to delay—unless “Delay” is the best alternative in your analysis. Often a ten-minute, quick-and-dirty calculation is sufficient to see whether a decision needs to be made now.

Central to the decision analysis approach are:

• Value function. This measures “goodness” toward the organization's objective( s). For most purposes, dollars are a suitable measure (non-monetary objectives can be translated into dollar-equivalents). Multi-criteria value functions translate multiple important facets of a problem into a single value measure.
• Probability distributions. These express judgments about risks and uncertainties. Distributions can also be used in presenting the results of a decision analysis.
• Expected value. This is how risk and value are combined for making decisions. Maximizing expected monetary value expresses a complete, succinct decision policy appropriate for many organizations.

The concepts are straightforward once you know about them. The methodology is proven and accessible. With practice, decision analysis is easily integrated into the professional's problem-solving approach. ■

Notes

1. Expected Monetary Value is normally the probability-weighted present value of the after-tax incremental net cash flow.

2. Risk most often means chance of an undesirable outcome. English does not have a good word meaning possibility of a desirable outcome. While not wholly satisfactory, opportunity is used to mean the fortuity of a desirable outcome.

3. In the fourth installment (Oct. 1993), “Value of Information” problems were introduced. Value of information problems are evaluated similarly to value of control problems.

4. In this situation, the revised probabilities are easy to calculate. More often, new judgments or Bayes theorem would be used.

5. For a risk-neutral decision maker or organization that is using EMV as the decision policy.

6. A hurdle rate is a minimum (or maximum) acceptable level of some decision criterion, e.g., demanding a minimum 15 percent internal rate of return.

John R. Schuyler, PE, CMA, is a principal of Decision Precision® Group, specializing in project risk and economic decision analysis. He recently co-founded Coaching For Success^SM Inc., a training and consulting group.