ZACH Insights / Decisions, learning and uncertainty
Decision-making
under uncertainty.
A choice is the visible part. Attention, knowledge, habits, feedback and the surrounding organisation influence how the choice is made. Explore what sits beneath it.
By Zacharias Razvi · Updated 12 November 2026
The central distinction
A good outcome and a good process are different questions.
Imagine a leader choosing a supplier. The familiar supplier is reliable, but an alternative might offer a better fit. The leader needs to act before every consequence can be known. An unfavourable result does not automatically show that the original process was poor; a favourable result does not prove that it was sound.
Decision-making connects an objective, an estimate of what may happen, the available options and a rule for choosing. Learning connects the choice with later feedback. When those parts are mixed together, it becomes easy to mistake confidence for information or luck for skill.
After Super Deciders, Figure 11.2
The decision iceberg.
Select a layer to connect its factors with the supplier example.
The observable choice.
The contract, action or commitment is visible. The reasoning and alternatives often are not. Record what was known at the time before judging the choice by its outcome.
Supplier example: renewing the familiar supplier could be sensible because disruption is costly, even if a competitor later proves successful.
Research connection: forecasting studies measure the quality of probability estimates; they do not capture every value, cost or organisational consequence of a decision.
Two things sit above the waterline in the model. Open each one.
Exploit or explore Super Deciders · ch. 3
The book treats most decisions under uncertainty as a point on a line: keep improving what is known, or try something new, with many positions in between. It states that no general rule solves the dilemma. People lean instead on seven rules of thumb: objectives, opportunities, time, social context, personality, skills and mood.
Decision performance Super Deciders · ch. 11
In the research the book draws on, performance is measured with a multi-armed bandit task: repeated choices among options whose pay-offs drift over time. A higher score means the person balanced reward now against information for later more effectively in that task.
Limit: a laboratory measure of one learning problem. It is not a verdict on a person or a career.
The state in which thinking happens.
Alertness, sleep loss, fatigue, acute stress, distraction and competing demands can change the conditions for attention and learning. Effects vary with the task; these factors do not add up to a validated readiness score.
Supplier example: a tired leader may have less room to examine a surprising delivery failure. A scheduled review and a colleague's independent check can be practical process choices.
Research connection: a sleep-deprivation experiment examined feedback-based updating; a recent review found task-dependent and inconsistent decision effects. Explore sleep ↗
The book’s first group of factors. Its claim: states can be regulated relatively quickly.
Sleep Super Deciders · ch. 11
The book reports that about six hours a night for more than two weeks lowered mental performance to a level comparable with 48 hours awake (citing Durmer and Dinges, 2005). It adds that sleep loss disrupts prefrontal function, so emotionally driven processes weigh more and choices drift towards too risky or too cautious (Chee and Chuah, 2008; Killgore, 2015).
A study behind it · Whitney et al., 2015
Participants learned by trial and error which signals to respond to; then the rule was reversed without warning. Rested participants adjusted. After total sleep deprivation, people kept responding as before and learned little from the feedback telling them they were wrong. The authors call this feedback blunting.
Limit: a laboratory task after a full night or more without sleep. The scoping review further down shows that findings vary with the task and the kind of sleep loss.
Physical condition Super Deciders · ch. 11
The book links regular movement and fitness with better cognitive performance and wellbeing, and with slower age-related decline in the brain (citing Tomporowski et al., 2008; Mandolesi et al., 2018).
Limit: this does not show that a training session improves a particular business decision.
Nutrition Super Deciders · ch. 11
The book states that poor nutrition can bring stress and fatigue and affect the ability to think and work. It mentions research on how the balance of carbohydrate and protein may influence decision behaviour through serotonin (Liu et al., 2021).
ZACH note: treat this as a hypothesis to follow, not as dietary advice.
Focus and time to think Super Deciders · ch. 11
Persistent high cognitive load with little time to reflect is the book’s fourth strain on state. It points to meditation research (Congleton et al., 2015) and to the value of unfocused, diffuse thinking for creativity and learning (Kounios et al., 2008).
A study behind it · Basso et al., 2019
Adults with no meditation experience were assigned to a daily 13-minute guided meditation or to 13 minutes of listening to a podcast. After eight weeks the meditation group reported lower negative mood and anxiety and did better on tests of attention, working memory and recognition memory. After four weeks the differences had not yet appeared.
Limit: one trial in beginners, measuring mood and laboratory tests, not decisions at work.
Feeling safe or under pressure Super Deciders · ch. 11 and 13
The book describes that people who feel tired, low, stressed or under pressure take in negative information more readily and tend to play safe, while feeling safe, relaxed and energised makes exploring more likely (citing Sharot, 2017). Under tiredness, stress and pressure, it adds, fast emotional responses take a larger share of the decision.
Social connectedness and purpose Super Deciders · ch. 11
Lack of purpose and of connection with others is the fifth strain. The book cites work suggesting that helping others and belonging to a wider community support wellbeing (Post, 2014; Brown et al., 2012).
What the book suggests
Seven to eight hours of uninterrupted sleep, several short breaks in the day, two to three hours of vigorous activity a week, varied nutrition, mindfulness practice, less multitasking, talking problems through with someone, and daily social contact. Its point is that any combination that suits the person helps. Nobody needs all of it.
Skills that support adaptation.
Ask what would change your view. Update estimates when relevant evidence arrives. Test assumptions, compare alternatives and distinguish a useful signal from a noisy outcome. Flexible attention helps connect learning with action.
Supplier example: use a limited trial with agreed quality and delivery criteria. Decide in advance which information would justify expanding, changing or stopping it.
Research connection: forecasting interventions and a managerial bandit study examine specific skills and conditions. They do not validate this iceberg as an assessment.
The book’s second group: three skills that most people have to differing degrees and can strengthen with practice.
Attention control Super Deciders · ch. 11
The ability to focus and stay focused, weigh alternative courses of action and choose among them. The book describes it as selecting information, holding it in mind, thinking in abstractions and analogies, and starting or stopping actions, with the prefrontal cortex as a key region that matures in early adulthood.
It reports that the skill can be trained through practice on demanding tasks, physical exercise, meditation and reflection on one’s own thinking (Posner et al., 2014; Basso et al., 2019; Fleming, 2014).
A study behind it · Laureiro-Martínez et al., 2015
Sixty-three people with managerial experience played a four-option bandit task in a brain scanner. Choices that exploited the best-known option went with activity in regions linked to reward. Choices that explored went with activity in regions linked to attention control. Participants who engaged those attention-control regions more strongly when switching earned more points in the task.
Limit: brain activity is an indirect signal, and the outcome was points in a game. You can try the same kind of task further down.
Cognitive flexibility Super Deciders · ch. 11
Changing thoughts or actions when the environment changes, and seeing a problem from several perspectives (Geurts et al., 2009; Diamond, 2013). The book says it develops through life and can decline with age.
It is supported, the book reports, by learning new skills, by working with people from different professional and social backgrounds, and by methods that move repeatedly between exploring and exploiting, such as design thinking (Randhawa et al., 2021). The book’s example of learning by testing, an experiment with start-up founders, is described under the DOCTOR method below.
Empathy Super Deciders · ch. 11
Taking another person’s perspective brings in information a decision-maker would otherwise miss. The book separates affective empathy, sharing another’s feeling, from cognitive empathy, understanding what someone thinks without feeling it oneself. It describes them as independent abilities that rely on different brain networks.
It warns that failing to separate one’s own feelings from another’s can lead to empathic distress, and reports that empathy grows with experience and deliberate practice (Teding van Berkhout and Malouff, 2016).
A study behind it · Teding van Berkhout and Malouff, 2016
A meta-analysis pooled 18 randomised controlled trials of empathy training with 1,018 participants. Training raised measured empathy by a medium amount (g = 0.63; 0.51 after adjusting for likely publication bias). Effects were larger for health professionals and students and when empathy was measured objectively instead of by self-report.
Limit: the trials measure empathy, not the quality of later decisions.
Patterns a person brings.
Prior knowledge, cognitive abilities, thinking styles, motivation and openness to revision can influence how someone approaches a task. These patterns interact with practice and context; they are not a destiny or a genetic ranking.
Supplier example: industry experience may help identify a hidden compatibility issue. It can also make a familiar option feel more certain than the available data justify.
Research connection: superforecaster research examines abilities, task-specific skills, commitment and enriched environments. Read why genetic associations do not define a person ↗
The book calls this group cognitive-behavioural inclinations: deepest in the model, least conscious and hardest to change. It also calls them assets.
Personality and character Super Deciders · ch. 13
The book notes the Big Five model and works with the Temperament and Character Inventory (Cloninger): four temperaments (novelty seeking, harm avoidance, reward dependence, persistence) and three character dimensions (self-directedness, cooperativeness, self-transcendence). It describes temperament as largely inherited and visible from early childhood, and character as shaped more by environment.
Curiosity, openness and comfort with risk lean a person towards exploring. Valuing tradition, conscientiousness and caution lean towards exploiting.
ZACH note: a description of tendencies. It does not rank people or fix what they can do.
Values Super Deciders · ch. 13
An internal signpost for what is good, desirable and important, more abstract than a norm. The book reports that most values form in three periods of childhood and youth (Massey, 1973), that they strongly influence behaviour (Rokeach, 1973), and that the person someone wants to be also steers their decisions.
Reactive emotional tendencies Super Deciders · ch. 13
Habitual ways of acting when stress and pressure arouse strong emotion. Building on Karen Horney, Anderson and Adams (2015) describe three.
- Controlling pushing for achievement, setting exacting standards, taking charge
- Protecting holding back, analysing what is wrong, keeping distance
- Complying meeting others’ expectations before one’s own, testing whether it is safe to speak
The layers are connected. A changing state may affect how a skill is expressed. A team can provide information a person did not have. A review process can make updating easier. The model is useful for asking better questions about the process, without pretending to measure the whole person.
From the book / chapter 11
Three layers.
Three speeds of change.
The book closes its chapter on the individual with one distinction: the three groups of factors do not change at the same pace. That decides where effort pays first.
Cognitive state
Days.
Sleep, breaks, movement, food, fewer tasks at once and contact with other people can shift state relatively quickly, the book reports.
Adaptive skills
Months of practice.
Attention control, cognitive flexibility and empathy take regular, continued practice. They cost more time and effort, and they are learnable.
Dispositions
Slowly, if at all.
Personality, values and emotional tendencies are much harder to change. The book’s advice is to know them and use them as a guidepost.
Summary of the book’s argument in ZACH’s words. The labels “days” and “months” are ZACH’s shorthand for “relatively quickly” and “regular and continuous practice”; the book gives no fixed durations.
Using knowledge and creating knowledge
Should we stay with the known option?
Exploitation means using an option that current knowledge suggests is valuable. Exploration means investigating an alternative to learn whether it could be useful. Neither is inherently superior. A choice can create an immediate result and information for later choices.
Use what you know.
Renewing the familiar supplier preserves a workable process and avoids switching costs. That can be appropriate when the time horizon is short or disruption would be costly.
Learn what you do not.
A defined trial with another supplier can reveal quality, reliability or compatibility. The trial has a cost, and its value depends on how the information can be used later.
The balance changes with the remaining time, the cost of learning, the stability of the environment and the quality of existing information. A short horizon can favour immediate use of a reliable option. A longer horizon may create more opportunity to benefit from learning. High-stakes or irreversible decisions also change which experiments are acceptable.
From the book / chapter 3 · an illustration
Seven rules of thumb pull at once.
The book names seven things a person or a company weighs when placing a decision between exploit and explore. Set each one for the supplier choice and watch where they pull together.
All seven are neutral.
The seven rules are the book’s. The sliders and the marker are a ZACH illustration of the idea that several considerations pull at the same time. The marker is a simple average. It is not a score, a recommendation or a measurement.
Uncertainty is not one thing.
Sometimes the options are clear but their outcomes vary. Sometimes useful options are missing from the initial list. Sometimes the environment changes before yesterday's information can be used. State which uncertainty you are trying to reduce.
A supplier choice involves more than a reward number. Price, quality, implementation cost, values and obligations all matter. A laboratory model deliberately compresses that complexity so one learning problem becomes visible.
An educational laboratory
A choice returns an outcome and a lesson.
Try the task · a simplified four-armed bandit
Four options. Sixty decisions.
Each option pays between 1 and 100 points. What an option typically pays drifts slowly, and each one drifts on its own, so the best option now may not be the best later. Collect as many points as you can.
In the style of the four-armed bandit task introduced by Daw et al. (2006) and used with managers by Laureiro-Martínez et al. (2015), the kind of task the book calls the MAB test. The drift settings, the sixty decisions and the summary are ZACH simplifications. It is not the research instrument, it has no norms, and the result says nothing about you as a decision-maker.
What the model makes visible.
The multi-armed bandit model describes repeated choices among options with uncertain rewards. Sampling one option teaches you about that option, while leaving other options less understood. A high result may be luck. A low result may be noise. The observed average is an estimate, and the environment can change.
Directed exploration seeks useful missing information. Random exploration introduces variation without choosing solely on the best current estimate. These are different strategies, and the appropriate balance depends on the task. Real decisions also include dependencies and constraints that a small bandit model omits.
A reading route, with attribution
Where Super Deciders fits.
Super Deciders (Feser, Laureiro-Martinez, Frankenberger and Brusoni, 2024) connects management decisions with scientific ideas about uncertainty. This section summarises its decision model and its opening chapter in ZACH’s words. The supplier example and the interactive models are ZACH teaching material.
From the book / chapter 7
The book’s method: six steps it calls DOCTOR.
The authors build a decision model for dynamic and uncertain situations across several chapters. It treats options as hypotheses and assumptions as things to test.
Name the dilemma.
State the tension the decision contains: keep improving what works, or try something new. The book’s point is that such dilemmas look like a choice between two options but are a choice among many positions on a line.
Supplier example: keep the reliable supplier, or try one that might fit better.
Widen the options.
The book argues that people are inclined to think in either/or. The second step is to refuse that: combine, split and sequence until there are at least three or four options along the line.
Supplier example: renew fully, renew for a shorter term, run a limited trial alongside, or switch.
Write down the conditions.
For each option, ask what would have to be true for it to be right, using the seven rules of thumb as a checklist. The book gives two ways to surface assumptions nobody has stated: check whether the option really follows from the assumptions on the table and, if it does not, name what is missing; and keep asking why until nothing new appears.
Supplier example: the trial only makes sense if a second supplier can meet the quality criteria at small volume.
Test what matters.
Not every assumption needs a test. The book sets aside those that are widely accepted and those that would not change the conclusion even if false, and concentrates on the ones with the largest effect. It then ranks tests by cost: analogies from another industry or country, public data, existing studies and a few expert interviews are cheap; prototypes, large surveys and regional pilots are expensive.
Supplier example: one sample order answers the quality question before any contract changes.
Change the boundaries.
If no option is satisfactory, look for a limiting assumption that can itself be changed, and let new options appear.
Supplier example: if switching costs block every option, ask whether the notice period can be renegotiated.
Decide.
Resolution comes last, after the assumptions behind the leading option have been examined. The book describes the whole sequence as combining two ways of reasoning: bottom-up from observations, which is fast and creative and prone to bias, and top-down from a hypothesis that is then tested, which is slower and easier to defend.
Supplier example: decide, and record what was known on the day.
The study the book uses · Camuffo et al., 2020
Founders who treated their idea as a hypothesis.
116 Italian start-ups took the same eight-week course. By random assignment, 59 were also taught to state their business idea as hypotheses and to test them rigorously; 57 were not. The researchers followed all of them for a year.
The trained founders changed direction about twice as often and were quicker to drop ideas that did not hold. Their average revenue was higher.
Limit: early-stage start-ups in one country, and only a handful of firms in either group earned revenue, so the revenue average rests on few companies. Counts are from the authors’ own summary; the book reports the pivots as shares of each group (49% and 21%).
The six steps and the acronym are the book’s (chapter 7). Wording and supplier examples are ZACH’s. The book’s own worked cases are not reproduced here.
From the book / chapter 1
What made forecasters accurate?
The opening chapter summarises the forecasting tournaments studied by Mellers and colleagues. It sorts the drivers of accuracy into three groups and reports which mattered most.
What forecasters did.
Gathering information, time and effort, and above all how often they updated a forecast when new information arrived. The book calls updating the single most important predictor across all three groups.
How the work was set up.
Training in probabilistic thinking, and being placed in a team to debate forecasts. Teams were the strongest driver in this group.
What forecasters brought.
Intelligence was the strongest trait: the best forecasters scored at least one standard deviation above the general population. They also tended to be open-minded, to enjoy thinking, and not to see events as predetermined.
Order as reported in the book, citing Mellers et al. (2015). Bar lengths show rank only; they are not measured effect sizes. The findings concern defined forecasting questions, not every kind of decision.
Scope: based on a reading of chapters 1, 3, 5, 7, 11 and 13 and the publisher’s contents. Ideas are summarised in ZACH’s words and attributed; figures, cases and passages from the book are not reproduced. Studies named “as cited in the book” were not checked independently. The iceberg follows the book’s Figure 11.2 and is not a validated assessment.
Independent evidence
What was actually studied?
A forecasting tournament, a brain-imaging task and a workplace decision answer different questions. Keep the participants, comparison, outcome and setting visible before transferring a result.
Mellers et al. / 2014 and 2015
Forecasting interventions.
Geopolitical tournaments examined probability training, collaboration and tracking strong performers. The research supports studying skills and environments alongside cognitive characteristics.
Limit: better probability forecasts are not identical to better organisational decisions. Selection into elite teams also differs from a randomised intervention.
Read the original research ↗Laureiro-Martínez et al. / 2015
Switching attention.
63 healthy participants with managerial experience completed a four-option bandit task during fMRI. Exploration, exploitation and task performance were studied through choices and brain-activity associations.
Limit: fMRI is an indirect signal, and the laboratory outcome was accumulated points. This was not an exercise intervention or proof of improved leadership.
Read the original research ↗Almaatouq et al. / 2021
When teams help.
A preregistered two-phase experiment assessed 1,200 individuals and then allocated them to solve tasks in groups or independently. Task complexity changed the comparison between group and individual performance.
Limit: a controlled task is not every workplace situation. More discussion can add time and does not automatically improve a simple decision.
Read the original research ↗Whitney et al. / 2015
Learning from feedback.
An experiment used a reversal-learning task to examine how total sleep deprivation affected acquisition and updating from outcome feedback.
Limit: the sleep-loss exposure and task are specific. This does not quantify the effect of an ordinary busy week on a particular leader.
Read the study ↗Agyapong-Opoku et al. / 2025
A varied decision literature.
A scoping review mapped 25 studies with 2,276 participants. Sleep-loss findings varied by task and exposure; some studies reported no significant effects in particular settings.
Limit: a scoping review maps the field. It does not supply one universal effect size or validate a readiness score.
Read the review ↗The transfer question
Does movement improve a business decision?
Physical activity has established broad health benefits. That alone does not prove that a ZACH session improves supplier choices, leadership or company results.
This is an academic area of interest for Zacharias. The website describes current movement and education services without presenting that interest as a validated business-outcome intervention.
Explore movement and the brain ↗A mechanism is a research route, not a guarantee.
Connecting movement, sleep, attention and decision-making can generate a useful question. Testing a benefit requires an appropriate intervention, comparison and outcome. A plausible pathway does not replace that study.
Use the illustrated evidence hierarchy to evaluate the design ↗
Original ZACH planning framework
Make the process reviewable.
Use a supplier choice to make the thinking concrete. The following prompts organise a conversation; they do not turn uncertainty into certainty.
- Define the actual decision. What must be chosen, by whom and by when? Which outcomes and constraints matter?
- Record the current view. Write the options, information sources and assumptions before the result is known. Separate a fact from a judgement.
- Look for an alternative explanation. What evidence would make the familiar supplier less attractive? What would make the new option unsuitable?
- Choose a proportionate way to learn. Define a feasible trial or information request. Set its scope, cost, safeguards and review point.
- Act with a revision rule. State what would justify continuing, changing or stopping. A rule makes later updating easier to discuss.
- Review process and outcome separately. Was the information useful and the reasoning appropriate at the time? What happened, and what can that result reasonably teach?
Feedback needs interpretation.
A late delivery could reflect the supplier, an unusual external disruption, an unclear brief or a reporting error. Repeating comparable observations can help distinguish patterns, but repeating a biased measurement does not remove its bias. Useful learning depends on the quality and timing of the feedback.
The same thinking can help a working week.
A movement routine also meets changing conditions. Prepare a usual version and a smaller suitable option, then review what fitted the week. This is a practical planning analogy rather than proof that training improves strategic decision-making.
Sources and reading scope
Follow the evidence.
- Feser et al. (2024). Super Deciders. Publisher and contents. Public first-chapter excerpt.
- Mellers et al. (2014). Psychological strategies for winning a geopolitical forecasting tournament. Original paper.
- Mellers et al. (2015). Identifying and cultivating superforecasters. Publisher abstract.
- Laureiro-Martínez et al. (2015). Exploration–exploitation, attention control and decision-making. Original paper; the author-hosted full report was consulted.
- Almaatouq et al. (2021). Task complexity moderates group synergy. Study record.
- Whitney et al. (2015). Feedback blunting after total sleep deprivation. Study abstract.
- Agyapong-Opoku et al. (2025). Sleep deprivation and decision-making: a scoping review. Review abstract.
- Cohen, McClure and Yu (2007). The exploration–exploitation trade-off. Research review.
- World Health Organization. Physical activity.
- Camuffo, A., Cordova, A., Gambardella, A. and Spina, C. (2020). A scientific approach to entrepreneurial decision making: evidence from a randomized control trial. Authors’ summary.
- Basso, J. C. et al. (2019). Brief, daily meditation enhances attention, memory, mood, and emotional regulation in non-experienced meditators. Study abstract.
- Teding van Berkhout, E. and Malouff, J. M. (2016). The efficacy of empathy training: a meta-analysis of randomized controlled trials. Record and abstract.
- Daw, N. D. et al. (2006). Cortical substrates for exploratory decisions in humans. Full text.
- Brusoni, S. et al. (2023), cited in Super Deciders as the source of the seven-factor figure (Figure 11.2). Not consulted directly.
- Further references are given as cited in Super Deciders and were not checked independently: Durmer and Dinges (2005); Chee and Chuah (2008); Killgore (2015); Tomporowski et al. (2008); Mandolesi et al. (2018); Liu et al. (2021); Congleton et al. (2015); Kounios et al. (2008); Sharot (2017); Post (2014); Brown et al. (2012); Posner et al. (2014); Fleming (2014); Geurts et al. (2009); Diamond (2013); Randhawa et al. (2021); Cloninger et al. (2016); Massey (1973); Rokeach (1973); Anderson and Adams (2015).
General educational material. Models, fictional examples and practical prompts are labelled. Zacharias Razvi is a final-semester physiotherapy student and is not yet an authorised physiotherapist.