Decision Making

From the role and decision map to a decision

Take out your Week 2 role and decision map. Its last row covers the company's biggest decision: which direction to test next. In the practice scenario, Mina prepares that decision, Judy makes it, and an outside adviser reviews it.

The decision concerns the three possible directions introduced in Week 1: Option A · Expansion (expand cheerleading and sports-entertainment talent management), Option B · Boutique (a small cross-border talent and advisory service for a few clients) and Option C · Content-centered return (return to content, advertising, food service and personal brand). The case does not say which one Judy will choose.

The decision will not wait for complete evidence. In the practice scenario, the professional team that wrote in Week 1, the home team, now wants an answer by tomorrow: will Judy's company bring cheer performers to its games next season? Waiting may close the opportunity. Saying yes may create promises that the company cannot keep, and your evidence record still shows many unknowns.

This week's question is: how should a manager decide when the evidence is incomplete, and how can the decision be written so that others can check it and change it later? The answer this chapter builds is a decision memo: a one-page record of the choice, the reasons, the evidence and the conditions for reviewing it.

Where this week fits

Week 2 produced a role and decision map. This week uses it to name who decides and who reviews, and adds tools for deciding under uncertainty. The result is a decision memo. In Week 4 you will list the memo's assumptions that depend on conditions outside the company, such as rules and partners.

Bounded rationality: thinking well within limits

Simple textbooks sometimes describe an ideal manager who lists every option, knows every consequence and picks the best one. Real managers never work like this. They have limited time, incomplete information and constant interruptions. Judy has one day to answer the home team, and Mina has not yet checked travel costs.

Herbert Simon (1955) called this condition bounded rationality: people reason carefully, but within limits on their information, time and attention. Bounded rationality is different from irrationality. A decision can be reasoned and responsible even when the decision maker could not know everything. So the useful question is whether the decision makers used a sensible process for the decision it actually faced.

One sensible process is satisficing. Satisficing means choosing an option that meets stated minimum requirements, instead of searching until the best possible option is found. The search for the best option may never end, and the opportunity may close while you search. Satisficing is responsible when the minimum requirements are written down before the options are compared. Then everyone can see what "good enough" means, and nobody can quietly lower the bar under pressure.

Here is an example of written minimum requirements for Judy's company. Before accepting any appearance request, the company requires three things: a signed agreement stating who pays for travel and when the fee arrives, each performer's written consent to the schedule and to the use of her image, and a named coordinator with free hours. The company may not know whether another request would bring more attention. It does know that a request failing any of the three requirements is declined, however attractive it looks.

Satisficing becomes careless when "good enough" is never defined. A founder under pressure may accept a vague promise because it feels better than waiting. Limits on time and information are real, and they make a clear process more important.

Week 2's roles apply here too: Mina prepares the evidence, Judy decides and the outside adviser reviews. If the decider cannot explain which evidence and criteria she used, the company cannot learn from the decision.

Frame the decision before comparing options

A weak question can spoil the analysis before it begins. "Should Judy's company become bigger?" could mean one contract, a new hire or a new identity, and each version needs different evidence. Compare it with this question: "Should the company run a 12-week test to see whether it can deliver two reliable game appearances to one paying team?" The second question names a period, a number of partners and a review point, so it is specific enough to check.

A useful decision frame names the choice and its deadline, says what is outside the decision, and names who may bear the costs. Framing can also hide values. "Which option earns the most?" leaves out rest, rights and cash timing, while "Which option is safest?" hides the cost of doing nothing. So ask who wrote the question and whose interests it leaves out.

Practice scenario: the question that changed the meeting

Judy's staff meet to discuss the home team's request. Judy asks, "Can we do it?" The person who manages the channel says yes, because the audience will grow. Mina says no, because the travel plan is incomplete. A performer asks whether the fee covers rehearsal time and rest.

Mina reframes the question: "By tomorrow, should we authorize a conditional negotiation, subject to a written agreement, confirmed travel support, the performer's consent and a cost ceiling?" With this question, Judy is no longer deciding whether the event will happen. She is only deciding whether Mina may spend one day checking those conditions with the home team and the performer. The final yes or no comes later, when the facts are known. Judy may still decide not to go ahead, but the discussion now focuses on conditions that can be checked.

Make the working rules explicit

In this course, every team that makes a decision states its criteria and working language at the start, and names who decides and where disagreement is recorded. Authority comes from the role and decision map, not from age, fluency or confidence.

Risk and uncertainty

Every decision contains unknowns, and each kind of unknown calls for a different method. This course separates two main kinds, risk and uncertainty. Two further kinds, deep and Knightian uncertainty, are optional reading at the end of this section.

Risk is a situation in which the possible outcomes and their chances can be estimated well enough for the decision at hand. Example: a venue may know from past seasons how often games are canceled, and an insurer may publish a fixed price for travel insurance. The estimate can still be wrong. But calculation has a sound basis, so a manager can put numbers on the risk and plan for them.

Uncertainty is a situation in which a manager lacks the knowledge to estimate the chances reliably. Example: a new team might pay on time or pay late, but Judy's company has no history with this team, so any percentage would be a guess. Under uncertainty, useful tools are ranges ("payment could arrive 30 to 60 days after the game"), similar past cases and smaller commitments. The memo should say where each assumption came from.

The classification depends on the decision, and one proposal can contain both kinds. For the home team's request, the price of a flight is close to risk, because it can be looked up. Whether fans will enjoy the performance is uncertain, and so is whether the home team will want a second season. So a decision memo classifies each important assumption, not the whole decision, and chooses a method to match.

How a manager sees uncertainty also shapes the method. In a scenario study with senior managers, Smit (2023) found that a threat frame pushed them toward forecasting, while an opportunity frame made them more willing to act on what they could control. A good memo uses both views.

Optional: deep and Knightian uncertainty

You may meet two stronger terms in further reading. Deep uncertainty describes situations in which experts do not know, or cannot agree on, how the system works, which outcomes are possible or how much each outcome matters (Marchau et al., 2019). The usual advice is to look for plans that stay workable across several possible futures. Bonjean Stanton and Roelich (2021) reviewed such methods, mostly in infrastructure planning, and found that the people who must carry out the plans were often left out of the analysis.

Knightian uncertainty, named after the economist Frank Knight, describes futures whose chances cannot be known or calculated at all, for example when a new kind of business creates situations nobody has seen before. Neither term is a fancy name for an ordinary missing number. For most decisions in this course, the difference between risk and uncertainty is enough.

Using the evidence record in a decision

Your Week 1 evidence record is the starting evidence for the memo. To use an entry in a decision, add two things to it: which criterion it affects, and what else could explain it. For example, a sudden rise in views could follow a good video, a platform recommendation or paid promotion. Check first the explanations that would change the decision.

The labels must survive the move. An inference stays an inference when it appears in a memo. A public page supports the narrow claim that a partner described a collaboration; who did the work needs other sources. Evidence also ages, so the memo gives the date each important source was checked and says when it must be checked again.

How much analysis does a decision need?

The right amount of analysis depends on the decision. Atuahene-Gima and Li (2004) studied new technology ventures and found that the value of thorough analysis depended on the kind of uncertainty the firm faced and on the result being measured. Eisenhardt (1989) studied fast-moving computer firms and found that the fastest decision makers used more real-time information and more alternatives, and relied on experienced advisers. McAllister, Mitchell and Beach (1979) found that people used more careful methods when a decision was significant, hard to reverse and something they would have to explain.

The practical lesson is to link each piece of analysis to a use. If a source will not change a criterion, a threshold or the design of a test, it is not a priority. Speed also depends on preparation. A reply such as "We can negotiate, but we cannot commit until consent and payment terms are confirmed" can be faster and safer than a rushed yes or a vague delay.

Criteria, measures, thresholds and weights

Four terms help a team compare options. A criterion is something that matters in the choice, such as delivery readiness. A measure says how the criterion will be observed. A threshold is a minimum (or maximum) condition that an option must meet before anything else is compared; it is the written minimum requirement from satisficing. A weight says how important a criterion is compared with the others.

Criteria can come from the four questions of Week 1: Target, Value, Capability and Revenue Model. Take "trust" as a criterion. On its own it cannot be observed, so a decision needs measures, such as clear agreements, repeat bookings and written consent records. A threshold might be "no public use of a person's image without written permission". A weight might give care and rights more importance than short-term reach for this decision.

Weights are choices about whose interests count most. Judy may weight growth because the company needs income. Performers may weight predictable schedules and rest, because the work affects their livelihood, and a partner may weight speed. A memo should show these differences and test whether the recommendation changes under other reasonable weights.

Worked example: a weighted decision matrix

Here is a weighted matrix for the three options. Each option is scored from 1 to 4: 1 means a major weakness nobody has resolved, 2 weak support, 3 support under stated conditions and 4 comparatively strong support. The matrix scores each option as a full commitment of the company's direction, not as a small test.

How the totals are calculated. Multiply each score by its weight, add the results and divide by 100. For Option A: (25 × 4) + (25 × 2) + (20 × 2) + (20 × 2) + (10 × 4) = 100 + 50 + 40 + 40 + 40 = 270. Dividing by 100 gives 2.70. The same steps give 345 for Option B (3.45) and 310 for Option C (3.10).

Sensitivity check. Now suppose Judy's staff care more about learning. They raise learning value to 40 and lower delivery readiness to 20, cash protection to 15 and care and rights clarity to 15, with strategic fit still at 10. The new totals are A 3.00, B 3.35 and C 2.90. B stays first, but A and C swap places. So the comparison between A and C is fragile, and a memo should present that part of the recommendation as conditional.

Threshold check. Suppose the company has written a threshold: care and rights clarity must score at least 3. Option A scores 2, so a full commitment to expansion fails, whatever its total. In the arithmetic, a high score for learning can make up for a low score for consent; in real life it cannot. That is why a weighted average can hide a red line. Check thresholds first, and average only the options that pass.

Every score needs a reason. Option A receives a 2 for care and rights clarity because no written consent process or rest rule exists yet; writing them would move it to 3. Reasons like these turn the matrix into a list of what to check next.

Compare the three options as hypotheses

At this stage, each option is a hypothesis: a statement about what might work, which evidence could support or weaken. The table compares the three options on five questions, and a decision memo should answer the same five.

The table does not rank the options. Its "test first" row shows that every option can start small, and its "hard to reverse" row shows where each option could lock the company in.

Reversibility and the size of the first step

The significance of a decision is more than its price. It includes safety, rights and the number of people affected. Reversibility is how easily a decision can be undone without serious cost or harm. A large investment can sometimes be split into steps so that the first step is easy to undo.

Bowman and Hurry (1993) describe a first investment as a way to keep an opportunity open: the company gains the chance, without the duty, to invest more after learning. In this course, a pilot counts as reversible only if its owner, budget, boundary and stop condition are written down in advance. If stopping would leave a performer without promised income, or embarrass a partner who has already announced the event, the first step already contains a promise that is hard to take back. Week 4 returns to this idea.

Measures change behavior

When people know how they will be judged, they often adjust to the rule. If a platform rewards views, creators aim for views. If Judy gives the best opportunities to whoever replies fastest, performers may reply quickly without checking details. Tsirtsis and colleagues (2024) build a mathematical model of how to set decision rules when people will change their behavior in response to them. Sometimes this improves the real work, and sometimes it only improves the number.

So a memo should ask two linked questions: what will this rule measure, and what will people learn to do because we measure it? It should also ask who must spend extra time to look eligible, and which unmeasured work, such as translation or rest, becomes less valued.

Human judgment when AI tools help

Mina might use an AI tool to summarize a Korean-language video or translate a team's message. Week 2 set a course rule: for any important AI-assisted decision, name the person who can explain it and reverse it. Research gives four reasons for the rule.

• Prediction has limits. Townsend and colleagues (2025) argue that when the future is genuinely new, a more powerful prediction tool cannot turn uncertainty into calculable risk.
• How the work is divided matters. Fügener and colleagues (2022) found that people working with AI advice tended to rely on it too much and were poor at knowing when to hand a case to it. When people chose which cases to give the AI, results were worse than the AI working alone; when the AI chose which cases to give to people, results were better. Krakowski and colleagues (2026) found that results with an AI tool depended on whether the company adapted its procedures, training and incentives to it.
• Reliance depends on the situation. Alon-Barkat and Busuioc (2023) did not find that people generally followed advice from an algorithm more than the same advice from a human expert. Parasuraman and Manzey (2010) describe two kinds of error: missing a problem because the tool did not flag it, and following advice that was wrong.
• Real correction rights help. Dietvorst, Simmons and Massey (2018) found that people used an imperfect forecasting tool more, and performed better, when they could adjust its forecasts even slightly.

Assign the human roles before using the tool

Before Mina uses an AI tool on an important decision, the memo names five human roles. One person may hold several of them.

• Task responsibility: who prepares the data, runs the tool and checks the sources? Example: Mina.
• Decision right: who may make this kind of decision, and within what limit? Example: Judy.
• Accountability: who must explain the process, make sure it is reviewed and answer for the result? Example: Judy.
• Human oversight: who watches how the tool performs over time and can stop its use? Example: Mina, reporting to Judy each month.
• Human review: who examines one particular output and can change or reverse it? Example: the translator checks every AI translation of an agreement before anyone relies on it.

A quick approval click fills none of these roles if the person clicking lacks the information or the authority to say no. Guidance from the U.S. National Institute of Standards and Technology (Autio et al., 2024) and the OECD AI Principles (OECD, 2024) supports this approach: clear human roles, fact-checking and ways for affected people to challenge AI outputs.

Writing a decision memo

The decision memo is this week's artifact. It is short enough to read in a few minutes and detailed enough for someone else to check. Use these parts:

• Decision requested: one sentence and a deadline.
• Who decides, prepares and reviews: taken from your role and decision map, with the limits of each role.
• People affected: those who gain or bear costs, including people with no say in the decision.
• Options considered: including "do nothing for now" when it is realistic.
• Criteria and thresholds: with the weights and the reason for them.
• Evidence: the key evidence, inference and unknown, with sources, dates and labels kept.
• Recommendation and first step: the choice and its conditions, the owner and budget, and whether it can be undone.
• Review trigger: the event or date that reopens the decision, who calls the review, and what can then be continued, changed or stopped.

The review trigger matters because a decision continues after it is made. Klingebiel and De Meyer (2013) followed one large project and found that managers handled decisions differently when new unknowns appeared and when old unknowns were resolved. "Review later" is too vague to act on. "Mina calls a review if the performer's consent is not confirmed within three weeks" is a trigger someone can check.

A memo should also record disagreement and write about people with care. "The performer is unreliable" may be a conclusion built from one missed message. "Our records do not show whether the performer understood the commitment" points to a problem the company can fix. In the same way, "This market is unsuitable" can become "Our current plan leaves payment timing unresolved", which points to a test.

Worked example: a decision memo for Judy's company

This memo applies the parts above to the home team's request and uses the matrix from this chapter. A team with different weights could reasonably recommend something else.

The memo and the matrix answer different questions. The matrix scores each option as a full commitment, and on those scores Option B looks strongest today. The memo decides something smaller: whether to run a test. A small, reversible test of Option A is cheap to learn from, because nothing is promised in public and stopping costs a few conversations. It also gathers evidence on A's weakest criterion, care and rights clarity.

Another group of decision makers could reasonably test B first; what matters is that the reasons are written down. Week 4 then studies the memo's unknowns that depend on outside conditions, such as the payment permission.

Class practice: a matrix and a one-page memo

This practice is done without AI tools. Your teacher will tell you which worksheet to use and how to submit it. Bring your Week 1 evidence record and your Week 2 role and decision map.

1. Frame one decision. Write one decision Judy's company could make in the next 12 weeks, with a deadline and one thing that is outside the decision.
2. Rework the matrix. Start from the matrix in this chapter. Change one weight and one score you disagree with, and give a one-line reason for each. Recalculate the three totals and apply the threshold. Does the order change?
3. Write the core of the memo. Name who decides and who reviews, using your Week 2 map. Copy one piece of evidence, one inference and one unknown from your Week 1 record, keeping their labels. Then write your recommendation and one review trigger.
4. Check each other's work. Swap with a classmate. Your classmate marks one claim with no support and one value judgment that is hidden inside a "fact". Restate the criticism in your own words before you decide whether to revise.

Label any number you invent as invented. A strong memo lets a reader see what you decided, why, on what evidence, and what would make you change your mind.

Key points and what you bring to Week 4

• Managers reason within limits. Satisficing is responsible when the minimum requirements are written down before the options are compared.
• Risk means the chances can be estimated; uncertainty means they cannot. Classify each important assumption and choose a method to match.
• Criteria, measures, thresholds and weights do different jobs. Check thresholds first, because a weighted average can hide a red line.
• Run a sensitivity check. If reasonable weights reverse the ranking, present that part of the recommendation as conditional.
• Size the first step by how easily it can be undone, and name the human roles before an AI tool helps with a decision.

For Week 4. Bring your decision memo. In Week 4 you will list the memo's assumptions that depend on conditions outside the company, such as permission to work, a partner's authority to sign and the language of the agreement. You will turn them into an environment map.