Consequences Under Uncertainty

Uncertainty means more than one future remains possible and you do not know which will occur. Good decision-making represents that uncertainty explicitly instead of replacing it with one confident forecast.

“The project will earn £500,000” hides variation. “There is a plausible range from a £150,000 loss to an £800,000 gain, with the central cases between £200,000 and £500,000” gives the decision owner something to examine.

Uncertainty is not a reason to avoid choosing. Waiting is also an action with consequences, and information often remains incomplete at the deadline. The aim is to make a choice whose logic survives plausible futures and whose downside you can bear.

Begin by replacing every unsupported “will” with “could,” then state what would have to be true for that consequence to occur.

A consequence map traces how an option interacts with uncertain events to produce outcomes. It separates what you control—the action—from what you only estimate—the state of the world.

For a café considering a second location:

For every option, map:

1. immediate controllable actions;
2. external uncertainties such as demand, regulation, prices, or competitor response;
3. internal uncertainties such as delivery capability or adoption;
4. first-order consequences;
5. delayed or second-order consequences;
6. the values each consequence helps or harms.

Avoid sprawling trees. Split only where branches would change the decision. “Rain on opening day” does not deserve a branch unless it materially affects the choice or mitigation.

Write causal links as testable claims: “If peak demand exceeds 400 orders daily, the current kitchen becomes the capacity bottleneck.” This is more useful than “growth could be difficult.”

A probability is a numerical expression of uncertainty given current information. It is not a promise, and a 70% event can fail without the estimate having been dishonest.

Use plain anchors:

Probability	Plain-language interpretation	Repetition test
10%	Unlikely, but serious enough to imagine	About 1 in 10 similar cases
30%	Less likely than not	About 3 in 10
50%	Balanced uncertainty	About half
70%	More likely than not	About 7 in 10
90%	Very likely, not guaranteed	About 9 in 10

Never use 100% for an empirical future unless failure is logically impossible. “Near certain” still needs a failure branch when consequences are severe.

Estimate with a range when evidence is weak: “40–60%” is more honest than “52%.” Then ask whether any value in that range changes the preferred action. If not, further precision has little decision value.

Keep assumptions beside estimates:

Estimate	Evidence	Assumption	Confidence
60% chance of 2,000 customers	Comparable store traffic and local survey	Survey respondents behave like buyers	Medium
20% chance of build delay	Three similar projects	Supplier capacity remains stable	Low

The explanation is as important as the number because it shows where new evidence should update the estimate.

A base rate is the observed frequency of an outcome in a relevant comparison class. It provides an outside view before the vivid details of your own plan pull the estimate toward optimism or fear.

If 30% of comparable projects finish on time, “our team is excellent” should not automatically produce 90%. Identify evidence that distinguishes this case: team continuity, scope, dependencies, supplier commitments, or a completed prototype.

Choose the comparison class carefully:

Too broad	Too narrow	More useful
All businesses	One friend’s café	New cafés in similar UK cities and rent bands
All software projects	This team’s last sprint	Cross-team migrations of similar size
All hires	One exceptional employee	Senior product hires in this role and market

Then show the adjustment: “The base rate for on-time completion is 35%. Completed integration tests and a reserved migration window move our estimate to 50–65%; an unfilled security role prevents a larger increase.”

When no reliable base rate exists, say so. Use analogous cases, ranges, and scenarios rather than inventing statistical authority.

Expected value is the probability-weighted average of possible numerical outcomes. It helps compare uncertain options on one measure, especially when choices repeat, but it is not a complete decision rule.

Suppose a company can run a marketing campaign costing £40,000:

Outcome	Probability	Gross value	Probability × value
Strong response	20%	£200,000	£40,000
Moderate response	50%	£80,000	£40,000
Weak response	30%	£10,000	£3,000
Expected gross value	100%		£83,000

Expected net value is £83,000 − £40,000 = £43,000.

Compare a smaller campaign costing £15,000 with expected gross value of £45,000: its expected net value is £45,000 − £15,000 = £30,000. The larger campaign leads on expected money, but the calculation has not considered cash constraints, learning, reputation, workload, or the ability to survive loss.

Check three conditions before relying heavily on expected value:

• the numbers represent the outcome you actually value;
• probability estimates are grounded enough for the ranking;
• an adverse outcome would not cause unacceptable or irreversible harm.

Expected value is an aid to comparison, not a machine that knows what you should care about.

Sensitivity analysis asks whether the preferred option changes when uncertain inputs move across plausible ranges. It reveals which assumptions deserve research, mitigation, or explicit acceptance.

Imagine Option A has expected net value £120,000, based on a 60% success probability. Option B has a steadier £85,000. Recalculate A:

Success probability for A	Expected net value	Leader
70%	£155,000	A
60%	£120,000	A
50%	£85,000	Tie
40%	£50,000	B

The switching threshold is 50%. The useful question is no longer “What is the exact probability?” but “Is success more likely than 50%, and what evidence separates the estimate from that threshold?”

Vary one input first to see its influence, then combine adverse assumptions into a stress scenario. Avoid changing every number at once without recording which assumption caused the reversal.

Tail risk is a low-probability, high-impact outcome at the extreme of a distribution. Ruin means an outcome that removes your ability to continue—bankruptcy, catastrophic harm, loss of a licence, or damage that cannot be repaired.

Expected value can recommend a gamble you cannot survive. A 99% chance to gain £10,000 and a 1% chance to lose everything may have positive expected money for some balance sheets, yet be unacceptable when “everything” includes a home, a life, or the company’s existence.

Use three filters:

Filter	Question	Possible response
Survival	Could this outcome end the ability to continue?	Reject, insure, cap exposure, or stage commitment
Recoverability	Can harm be reversed within acceptable time and cost?	Add backup, rollback, reserve, or escape clause
Concentration	Is too much dependent on one uncertain event?	Diversify, sequence, or reduce stake

Distinguish risk capacity from risk appetite. A founder may feel comfortable risking the payroll, but comfort does not create the cash needed after failure. Capacity is about resources and obligations; appetite is about preference.

Set a hard loss limit before enthusiasm grows: “No experiment may expose more than £25,000 or customer records,” “Maintain twelve months of operating runway,” or “Do not proceed without a tested rollback.”

A worked example connects consequence maps, base rates, expected value, sensitivity, and ruin. A small company must choose a full launch, a limited pilot, or a three-month delay for an unproven service.

Option	Success case	Adverse case	Estimated probabilities	Direct cost
Full launch	£500,000 gross value	£220,000 loss plus reputation damage	55% success; 45% adverse	Included in values
Limited pilot	£140,000 gross value	£45,000 loss	65% success; 35% adverse	Included
Delay	£260,000 later value	£60,000 opportunity loss	70% improved launch; 30% weak market	Included

Simplifying to the stated monetary outcomes:

• Full launch expected value: (0.55 × £500,000) + (0.45 × −£220,000) = £176,000.
• Pilot expected value: (0.65 × £140,000) + (0.35 × −£45,000) = £75,250.
• Delay expected value: (0.70 × £260,000) + (0.30 × −£60,000) = £164,000.

Full launch leads on expected money. But the company has only £250,000 of runway, so a £220,000 loss approaches ruin. Its 55% success estimate is also above the 35% base rate for comparable launches; the team’s completed prototype supports some uplift, but not enough to establish 55% confidently.

Sensitivity shows the full launch falls below delay if its success probability is below roughly 53%. The company chooses the pilot because it buys evidence about demand and reliability while capping loss. The pilot is not highest in immediate expected value; it best protects survival and future choice.

The decision statement records both logics: “Pilot to 500 users for six weeks; proceed only if weekly retention exceeds 40%, severe incidents remain below two, and projected acquisition cost stays under £80.”

An uncertainty worksheet keeps forecasts auditable and action-focused. Use it after options are real but before a preferred option becomes emotionally fixed.

Element	Prompt
Consequence map	Which controllable action meets which external and internal uncertainties?
Probability	What range represents current belief, and why?
Base rate	What happened in the most relevant comparable cases?
Expected value	What is the probability-weighted result on a useful measure?
Sensitivity	Which input changes the ranking, and at what threshold?
Tail risk	Which low-probability outcome causes severe harm?
Ruin limit	What must remain protected even in the adverse case?
Learning	Which test would most improve the choice?

Final checklist:

• [ ] Each serious option has favorable, central, and adverse consequences.
• [ ] Causal links explain how the consequence would occur.
• [ ] Probabilities use ranges when evidence is weak.
• [ ] Every major estimate names evidence and assumptions.
• [ ] A relevant outside-view base rate was sought.
• [ ] Case-specific adjustments to the base rate are explicit.
• [ ] Expected-value arithmetic is shown and units are consistent.
• [ ] Sensitivity identifies switching thresholds.
• [ ] Tail risks are considered even when they barely affect the average.
• [ ] No option threatens ruin without a cap, safeguard, or compelling necessity.

Uncertainty never disappears from a consequential choice. The practical skill is to represent it honestly, find what matters to the ranking, and protect the ability to act again.