Decisions under uncertainty should be judged by the quality of the decision making, not by the quality of the consequences. Robert F. O’Keeffe, the retired head of claims for INA (now CIGNA), one of the largest U.S. property and casualty insurance companies, understands this distinction well (perhaps in part because he’s an avid poker player). In a recent conversation, Bob stated his philosophy:
If I try to settle a major liability claim out of court and the other side’s final offer exceeds what my analysis shows to be a fair value, I take the case to trial. Often I either win outright, or the jury awards the plaintiff less than my calculated value or less than the plaintiff’s final offer. But sometimes the jury award exceeds what I could have settled for. The difference can be tens of thousands of dollars or even hundreds of thousands. In these cases, was it a mistake to refuse the offer? No. I just remind myself that another jury who heard the same evidence might have made a more favorable award.
O’Keeffe’s overall record attests to the quality of his decisions, yet over the course of his career he has encountered many surprises and upsets. The best that O’Keeffe or any of us can do in making an important decision is to ensure we use a sound process that enables us to identify and think clearly about uncertainty. We can’t make uncertainty disappear, but we can address it explicitly in our decision-making process.
Use Risk Profiles to Simplify Decisions Involving Uncertainty
Uncertainty adds a new layer of complexity to decision making. A single decision may involve many different uncertainties, of varying levels of importance, and they may all interact, in tangled ways, to determine the ultimate consequences. To make sense of uncertainty, you need to find a way to simplify it—to isolate its elements and evaluate them one by one. You can do this by using risk profiles.
A risk profile captures the essential information about the way uncertainty affects an alternative. It answers four key questions:
•What are the key uncertainties?
•What are the possible outcomes of these uncertainties?
•What are the chances of occurrence of each possible outcome?
•What are the consequences of each outcome?
By providing a consistent basis for comparing the uncertainties affecting each of your alternatives, risk profiles allow you to focus in on the key factors that should influence your choice, ignoring peripheral factors. Consider this simple example. Joe Lazzarino has kept his small consulting firm in business over the last five years by bidding on many small public and private engineering projects. His company consistently makes a modest profit, but Joe’s starting to get bored—he’s eager for new and bigger challenges. One day, he receives word that a government agency has issued a request for proposals for a large, multiyear contract. Joe sees that winning the contract would provide enormous benefits, but the huge costs associated with preparing a proposal could deplete his firm’s resources. And, of course, the agency’s response to his proposal is uncertain. He might be granted a full contract, a partial contract, or no contract at all.
Joe creates a risk profile for the alternative of preparing and submitting a proposal. He succinctly describes the possible outcomes, their chances of occurring, and the associated consequences. He writes them up in a simple table, as shown below. Studying the risk profile, Joe sees a clear choice. Winning a partial contract (outcome B ) or a full contract (outcome C ) are much more likely outcomes than losing the bid altogether (outcome A), and both B and C would lead to consequences that are much more desirable than the current situation. Joe decides to go for it.
Joe’s experience with bidding, together with the limited number of alternatives and possible outcomes, made it fairly easy for him to draw up the risk profile. Many decision problems involving uncertainty will present greater challenges. In all cases, though, the development of clear, thorough risk profiles is the all-important first step.
Joe’s Risk Profile for Preparing and Submitting a Proposal
Uncertainty: Government response to bid
Outcome
Chance
Consequences
A. No contract Least likely Bad. Will need to reduce staff, borrow heavily, and scramble for some small contracts.
B. Partial contract Most likely Pretty good. More firm stability. Will make a decent profit.
C. Full contract Somewhat likely Wonderful. Not only very profitable, but also professionally interesting. Will greatly enhance our reputation.
How to Construct a Risk Profile
Now let’s look at how you’d go about constructing a more complex risk profile. Janet Ellingwood, the owner of a mail order firm in Denver, is planning a summer party for her 55 employees. They’ve worked very hard over the past year, and she wants to use this party to recognize and celebrate their efforts. Her objectives for the party are fun, family involvement, and reasonable expense. She informally polls her employees and finds that they favor two alternatives: a picnic at a mountain retreat with a swimming pool and a ball field, or a dinner dance at a downtown hotel.
When Janet looks at her three objectives, the picnic seems the better choice: Everyone would enjoy the games and facilities, it would involve employees’ kids, and the cost would be low. But the success of the picnic, much more than that of the hotel dance, would depend on the weather. While Janet knows that a sunny day would be more likely than a rainy one at that time of year, she also knows that Denver could experience one of its occasional summer downpours. If it did rain, the picnic would likely be a flop. Food could be served under a tent—for an added cost—but most other activities would be curtailed, and many people would stay home or leave early. On the other hand, few people would pass up the dance because of rain, and although the hotel’s outdoor patio—a memorable place on a nice evening—would be unusable, the ballroom would still be elegant and spacious enough for a pleasant evening.
In thinking quickly through the two alternatives, Janet has already roughly answered the four risk profile questions. She’s identified the uncertainty (weather), the possible outcomes (rain or shine), their chances (rain less likely), and the consequences (picnic a flop in the rain). In some cases, such brief, informal descriptions may be adequate to make a final decision, but Janet doesn’t feel that the information is sufficient to allow her to make a smart choice. She proceeds systematically to clarify the uncertainties, outcomes, chances, and consequences impinging on her decision.
Identify the key uncertainties. Virtually any decision involves uncertainties, but most uncertainties don’t influence consequences enough to matter. Selecting the uncertainties important enough to include in a risk profile requires just two steps:
•List all the uncertainties that might significantly influence the consequences of any alternatives.
•Consider these uncertainties one at a time and determine whether and to what degree their various possible outcomes might influence the decision. When there are many possible uncertainties, winnow them down to the few that are likely to matter most.
Janet’s decision presents a number of uncertainties in addition to the weather, including attendance and cost. In considering the possible outcomes for attendance, Janet concludes that nearly all employees would plan to attend either event and that knowing the exact number wouldn’t influence her choice. To evaluate costs, Janet asks the events managers of the two sites for estimates. She learns that the picnic would cost approximately $6,000 and the dinner dance roughly $12,500. These estimates would vary slightly depending on the exact number of guests and their food, beverage, and entertainment choices, but the variation would not significantly influence Janet’s thinking. So, even though attendance and cost are subject to some uncertainty, the possible outcomes would not impact the ultimate consequences enough to make a difference in Janet’s choice.
That leaves weather as the key uncertainty. No matter how appealing the picnic, if it rains many people will not attend or will leave early. The picnic would be a washout.
Define outcomes.
The possible outcomes of each uncertainty must now be specified. This requires answering two questions:
•How many possible outcomes need to be defined to express the extent of each uncertainty?
•How can each outcome best be defined?
The number of outcomes you’ll need to specify will depend on the kind of uncertainty you’re addressing. Some uncertainties inherently have a small number of clearly defined potential outcomes: Which of the two contestants will win the chess match? Will the pending legislation pass or be voted down? Others entail a large number of potential outcomes: How many people will attend next Saturday’s football game? How much money will I make or lose from buying this stock?
When there are many possible outcomes, you should simplify your expression of them by organizing them into ranges, or categories. The categories can be either quantitative ($10,000 to $20,000, $20,000 to $30,000, and so on) or descriptive (high, medium, low; successful, unsuccessful, neutral). In some cases, it may be helpful to assign a representative value to a numerical range—for example, using $25,000 as a stand-in for the range $20,000 to $30,000—to make calculations and comparisons easier.
Because complexity increases as the number of categories increases, you should always seek to narrow the set of outcomes down to the fewest possible—enough to fully describe the uncertainty, but no more. Start by defining a small number of outcomes, and then add more only as needed. If you’re projecting the possible outcomes of a new product launch, for instance, you might start with just two categories: ‘‘High sales’’ and ‘‘Low sales.’’ If they are insufficient to capture the range of outcomes, you would then create a new category, ‘‘Medium sales,’’ containing part of what was previously in both the high and the low categories.
However many outcomes are designated, they must meet three further criteria. First, the categories must differ clearly from one another, with no overlaps (that is, they must be mutually exclusive). ‘‘Widely scattered showers’’ shouldn’t be included in both ‘‘Rain’’ and ‘‘Shine.’’ Second, the outcomes must include all possibilities, with every possible contingency falling within one or another category (that is, they must be collectively exhaustive). ‘‘Widely scattered showers’’ must be included in either ‘‘Rain’’ or ‘‘Shine.’’ Third, the outcomes must be unambiguously defined, so that when the uncertainty is resolved, the event can be clearly recognized as falling within one or another of the defined categories. If widely scattered showers occurred, was the weather rain or shine?
Assign chances. Clearly defining the possible outcomes or categories of outcomes will help you in judging the chance, or likelihood, that each outcome will occur. Still, though, assigning chances can be one of the toughest and most nerve-wracking tasks in decision making, especially when you don’t know very much about the subject or when you’re under time pressure. But you can help ensure that your assessments are both reasonable and useful by following these suggestions:
•Use your judgment. Often, you can make a reasonable assessment of the chances of a given outcome based on your own knowledge and experience. Oddsmakers do it all the time in sports betting. Friends do it when they arrange blind dates. We all do it almost unconsciously in daily life: What are the chances I’ll encounter delays on my homeward commute this Friday?
•Consult existing information. There will often be information available that will help you assign chances to outcomes. You should carefully consider all the potential sources of information—libraries, the Internet, documents in your organization, research data, professional publications—that might shed light on the potential outcomes. Janet, for example, might get climatological data from the weather bureau to help her assess whether it will rain on a summer afternoon or evening.
•Collect new data. Sometimes the particular data you need may not be available off the shelf—you may need to collect them yourself. A food company might estimate the percentage of families who will buy a new brand of coffee by conducting a market trial or a telephone survey.
•Ask experts. For most uncertainties, there will probably be someone out there who knows more about it than you do. Seek out an expert—your doctor, lawyer, or accountant, an economist—and elicit his or her judgment. In Janet’s case, a local meteorologist would be an appropriate expert.
•Break uncertainties into their components. Sometimes dividing an uncertainty into its components, thinking about the components, and then combining the results will help in establishing probabilities. An entrepreneur recognizes that the success of a new car wash in an area currently undergoing development will depend on the relative number of cars brought to the area by the different proposals for the adjoining site: a shopping mall or an office park. He can assign chances to various ranges of washes per day assuming the mall is built, and do likewise assuming the office park is constructed. He can then blend the results in proportion to the chances he assigns to the construction of a mall and of offices, to get an overall assessment of washes per day.
When expressing chances, qualitative terms may come first to mind. In casual conversation, people often describe chances using phrases such as ‘‘unlikely,’’ ‘‘toss-up,’’ ‘‘barely possible,’’ ‘‘fairly likely,’’ ‘‘pretty good chance,’’ ‘‘almost sure,’’ and so on. They do this not only because it’s easy, but also because they think they’re really communicating their judgments about likelihood. But one person’s ‘‘fairly likely’’ may or may not be the same as the next person’s. Such subjective phrases may be sufficient for personal decisions that will not need to be justified to others, but they’re not precise enough for most decisions. In most cases, therefore, you will want to express chances quantitatively, as actual probabilities, using either a decimal (0.2) or a percentage (20 percent). Using numbers reduces the likelihood of miscommunication and sharpens decisions.
If you are having trouble expressing your judgment quantitatively, or getting someone else to do so, zero in from the extremes. If you ask the hostess at a busy, no-reservations restaurant the chances of getting seated at 5:30 P.M. on Thursday, she might respond, ‘‘I haven’t a clue; either you will or you won’t.’’ (Ah, frustration!) Countering with the question, ‘‘Is the chance better than 25 percent?’’ will very often elicit something more useful: ‘‘Oh, much more than that.’’ ‘‘More than 50 percent?’’ ‘‘Yes.’’ ‘‘As much as 90 percent?’’ ‘‘Too high.’’ The range has been narrowed to between 50 percent and 90 percent; a few more questions might provide an even more precise range.
Pinpoint precision usually isn’t required in assigning chances. Frequently, knowing that a chance falls within a certain range is sufficient for guiding a decision. (See ‘‘Which Flight?’’ below.) If the estimated chance of some outcome falls between 30 percent and 50 percent, for example, compare the alternatives using 40 percent. Then reconsider them using 30 percent or 50 percent. More often than not, the change won’t matter; the decision will remain the same.
However they are expressed, the probabilities for the outcomes of an uncertainty should always add up to 100 percent (or, if you express them as decimals, to 1.0). If the two categories for weather are ‘‘Rain’’ and ‘‘Shine’’ and if the probability of rain is 35 percent, then the probability of shine necessarily is 65 percent. Also remember that your assessment of the chances of an outcome occurring may change as circumstances change or as new information becomes available. As you proceed through your decision process, regularly reexamine the chances you’ve assigned to ensure their reasonableness based on your current information.
Resolving a Decision with an Estimate of Uncertainty: Which Flight?
Mark Hata has a dilemma. Months ago, he arranged to take his 62-year-old mother on a week-long trip to London in October. Mark lives in Phoenix, his mother in Pittsburgh. They plan to meet at Dulles Airport in Washington, D.C., on a Saturday evening in time for a leisurely dinner, before taking the 10:00 P.M. flight to London.
But Mar
k has just learned that his daughter’s soccer team has earned a spot in the league championship game, which is scheduled for 9:00 A.M. that same Saturday, and he would really love to attend. What to do?
Mark sees three alternatives:
1.Attend the game and reschedule his departure to London for Sunday, cutting a day off the trip. (Reticketing would cost $400, but plenty of seats are available.)
2.Stick with the original plan and miss the soccer game.
3.Attend the game and take a later flight to Dulles. If this flight is on time or no more than 30 minutes late, Mark will just have time to meet his mother and make the flight to London. That nixes dinner but otherwise leaves their plan intact.
After some soul-searching, Mark decides he’d rather miss the game than shorten his mom’s London vacation. But should he attend the game and gamble on getting to Washington on time? After more thought, he decides he’d take the chance if the risk of missing the London flight is less than 15 percent.
With his decision boiled down to assessing the probability that he will arrive at Dulles no more than 30 minutes late, Mark checks with his travel agent and learns that Dulles has an 80 percent on-time arrival record, with ‘‘on-time’’ defined as arriving within 15 minutes of the scheduled time. After asking a few more questions of the agent, Mark figures his odds of arriving within 30 minutes are much better than 80 percent, for three reasons. First, many late flights arrive within 30 minutes of the scheduled time. Second, Saturday flights encounter fewer air traffic delays than do weekday flights. And, third, Phoenix has few weather-related departure delays. He concludes that he has at least a 90 percent chance of making the London flight. His choice is now easy, though still worrisome. He attends his daughter’s game—a 2-2 tie, cochampions—and arrives in Washington 15 minutes early. Not only did Mark make a smart choice, he enjoyed a good consequence.
Smart Choices Page 10