Making Decisions with Incomplete Information

Certainty is not coming; decision quality still must

Lesson 1 taught you to define problems precisely. Lesson 2 decomposed them with MECE issue trees. Lesson 3 tested hypotheses and updated beliefs with evidence. Lesson 4 forced explicit tradeoffs when everything cannot be first. None of that produces certainty. Markets shift mid-quarter. Integrations fail in staging. Competitors launch on your roadmap day. The manager's job is not to wait until the spreadsheet is perfect. The job is to make decision quality high under incomplete information: classify reversibility, estimate expected value with honest ranges, cap downside on experiments, document reasoning before outcomes arrive, and know when to optimize versus satisfice (choose good enough, Herbert Simon's term for bounded rationality).

Executives who wait for full information move after the market. Operators who ignore uncertainty bet the company on one narrative. This closing lesson integrates the unit into action under uncertainty. You will never have all data. You can still avoid the two fatal errors: treating reversible bets as if they are permanent, and treating permanent bets as if they are experiments.

Reversible vs irreversible decisions (two-way and one-way doors)

Jeff Bezos popularized a useful classification:

• Two-way door decisions are reversible with limited cost. You can walk back through the door if the experiment fails. Examples: pricing test on one segment, pilot feature behind flag, trial vendor for three months, internal reorganisation trial.
• One-way door decisions are hard or costly to reverse. Examples: major acquisitions, public pricing architecture changes, entering regulated markets, selling a core product line, committing to a multi-year lease for proprietary manufacturing.

Misclassification is expensive both directions. Treating a one-way door as two-way leads to "we can always undo the acquisition" fantasies while integration debt and culture clash compound. Treating a two-way door as one-way leads to analysis paralysis while competitors iterate.

Decision protocol: label the door type in writing before debate. Two-way: delegate closer to the edge, time-box analysis, default to experiment. One-way: escalate appropriate scrutiny, red team, decision memo, explicit kill criteria even after launch.

From Unit 1, remember transaction costs and firm boundaries: outsourcing a function is often two-way at small scale (switch vendors) but one-way once you dismantle internal capability. From Unit 2, business model choices (asset-heavy vs asset-light) shift how many one-way doors you face.

Expected value, ranges, and sensitivity (without fake precision)

When outcomes are uncertain, compare options with rough expected value (EV): the probability-weighted average of scenarios.

EV = Σ (probability × outcome)

Example: New market entry (NPV in $M):

Check: 0.30 + 0.50 + 0.20 = 1.00 ✓; EV = 6 + 1 − 2 = 5 ✓

EV of +$5M does not mean you earn $5M. It means repeated similar bets average toward that value if probabilities are calibrated. Managers still decide using EV plus downside tolerance, strategic optionality, and cash constraints.

Always pair point EV with:

• Range: 10th–90th percentile outcomes
• Sensitivity: which assumption moves EV most (often probability of failure or peak success)
• Break-even probability: how low can success probability fall before EV ≤ 0

If a positive EV bet can bankrupt the firm in the failure state, EV alone is wrong metric. Survival constraints dominate.

Barbell strategy: protect core, experiment at the edge

Nassim Taleb's barbell adapts to corporate portfolios: combine a conservative core (reliable cash generation, compliance, key customer retention) with small, diversified bets where downside is capped and learning is fast.

Avoid the mediocre middle: one $5M gamble without informational edge, or timid half-measures that spend like big bets but learn like small ones.

Barbell rules for operators:

• Cap any single experiment loss (budget, time, reputation exposure)
• Require predefined kill criteria (Lesson 3 falsification at portfolio level)
• Diversify bets across uncorrelated hypotheses
• Fund core reliability before transform theater (Lesson 4 portfolio)

Portfolio of ten $50K experiments often beats one $5M build without staged gates because variance of learning is lower and politics of sunk cost is weaker.

Decision records and post-decision learning

Write decision memos before big calls. Minimum fields:

When outcomes arrive, review process quality, not only luck. Good process + bad luck can repeat. Bad process + good luck cannot be scaled.

Post-mortems ask "what happened?" Decision reviews ask "was our reasoning sound given what we knew?" That distinction protects teams from hindsight bias where winners are geniuses and losers are fools.

Store memos where future leaders find them. Acquisitions repeated every decade often repeat the same integration mistakes because decision memory left with departing VPs.

Satisficing vs optimizing: where perfection pays

Herbert Simon noted managers satisfice because optimizing every decision is impossible. Limited attention and time force "good enough" on low-stakes choices.

The error is optimizing snack vendors while satisficing enterprise pricing. Map decisions by leverage and reversibility (Lessons 4 and 5 combined).

Integrating the unit: from alarm to action under uncertainty

Unit 3 is a single workflow, not five disconnected frameworks. Strong managers loop through it quickly and revisit steps when evidence demands.

Symptom alarm
  → Problem statement (Lesson 1)
  → MECE issue tree (Lesson 2)
  → Hypotheses + cheapest tests (Lesson 3)
  → Priority stack with opportunity cost (Lesson 4)
  → Decision under uncertainty with memo (Lesson 5)

Speed comes from skipping steps, not from skipping thinking. Teams that jump from symptom to solution skip problem definition. Teams that build 200-slide decks skip prioritization and decision classification. Teams that "move fast" on one-way doors confuse velocity with recklessness.

When information will remain incomplete at decision time, the unit still demands clarity on four questions before capital moves:

1. What problem are we solving, in one scoped sentence?
2. Which tree branch explains most of the variance today?
3. What evidence would change our mind in the next two weeks?
4. What are we not doing because we chose this path?

If you cannot answer all four, you are not ready for a one-way door. You may be ready for a two-way experiment.

Decision quality metrics for your team (track quarterly):

None of these guarantee outcomes. They guarantee repeatability. Boards fund repeatability because luck does not scale across a portfolio of bets.

As you enter Unit 4: Organizations and Execution, remember that the best problem-solving process fails if ownership is ambiguous and incentives reward activity over outcomes. Problem statements need owners. Tree branches need owners. Hypotheses need owners. Priority stacks need a single publisher. Decision memos need a signer. Execution is where defined problems become solved problems, or where they die in handoffs.

Decision memos: template and example excerpt

Use this skeleton for one-way doors and any decision above your materiality threshold:

1. Problem statement (Lesson 1, one paragraph)
2. Options (at least three, including do-nothing)
3. Issue tree summary (dominant branch, quantified)
4. Evidence and confidence (Lesson 3, branch IDs referenced)
5. Priority tradeoff (Lesson 4, opportunity cost)
6. Recommendation and door type
7. Revisit triggers and kill criteria
8. Owner and decision date

Short excerpt (fictional):

"Problem: UK enterprise prospects require in-region data residency; we lose deals without it. Options: (A) full UK entity Q1, (B) six-month lighthouse with partner host, (C) defer. Dominant branch: new-logo ARR blocked on residency (62% of stalled pipeline comments). Confidence 7/10 from CRM tags. Tradeoff: B consumes one senior solutions architect from U.S. enterprise installs (~$300K slip risk). Recommend B, two-way door, kill if <2 paid pilots by June 30. Revisit if U.S. NRR <98% two consecutive quarters."

Memos this crisp are rare and valuable. They allow dissent on facts before the decision, not after outcomes.

When a decision memo cannot be written in two pages, the problem is usually still scoped at symptom level, or the issue tree has a MECE gap. Pause and return to Lessons 1–2 rather than adding pages of prose that mask uncertainty.

Northwind Analytics (Lesson 1) validated that event pipeline collapse drove Q2 miss. Q3 debate: enter UK market ($1.8M investment) vs fix U.S. event motion ($400K). Information will stay incomplete: regulatory burden, UK bank sales cycles, and competitor response are uncertain.

Part A: Options and door classification

Part B: Scenario EV for UK full launch (illustrative)

EV = 0.25×12 + 0.45×3 + 0.30×(−2) = 3 + 1.35 − 0.6 = $3.75M

Investment $1.8M → positive EV on paper.

Sensitivity: if fail probability is 0.45 (not 0.30), EV = 3 + 1.35 − 0.9 = $3.45M still positive but thinner. Break-even fail prob solves 3 + 1.35 − 1.8p = 0 → p ≈ 0.41 if win/slow fixed.

Part C: Downside and barbell recommendation

Failure state loses $2M plus management distraction during U.S. fix window. Cash runway 14 months. Full launch risks survival constraint even with positive EV.

Recommendation (Pyramid answer): Fund U.S. event reinstatement ($400K, two-way). Run UK lighthouse pilot ($180K, six months, kill if <2 paid pilots by month 6). Defer full entity until pilot converts.

Opportunity cost explicit: "Pilot borrows one senior AE from U.S. enterprise expansion; two U.S. deals may slip one quarter."

Part D: Managerial read

Board gets decision memo with revisit trigger: "If UK pilots close ≥$500K ARR by June 30 and U.S. NRR ≥100%, approve full UK entity in July." Process quality documented before outcomes.

Helio Health needs predictive no-show modeling for clinic utilization. Options:

1. Hire lead data scientist ($250K loaded/year)
2. Agency project ($80K, 3 months deliverable)
3. Delay

Roadmap stability uncertain: telehealth regulations may shift product toward async visits, changing feature value.

Part A: Classification and unknowns

Hire is closer to one-way if wrong specialty hired; agency is two-way if scope fixed.

Part B: Two-step barbell plan

Step 1 (two-way, 4 weeks, $25K): agency spike on historical no-show data.

Step 2 decision tree:

• If lift ≥8% utilized appointments → extend agency 3 months OR hire if roadmap stable 18 months
• If lift <3% → satisfice with rules-based reminders (optimize elsewhere)

Parallel: limit hire approval until strategy memo on async vs video (one-way strategic clarity).

Part C: Expected value sketch (qualitative + partial quant)

Agency path: $80K + $25K spike, 70% chance useful model (+$400K annual margin from utilization), 30% chance discard (−$105K).

EV ≈ 0.7×400 − 0.3×105 − 105 sunk = 280 − 31.5 − 105 ≈ +$143K vs delay at 0.

Hire path: $250K year one, slower start, 50% chance fit if roadmap pivots (wasted $180K recruiting/onboarding).

Barbell chooses agency + spike before irreversible hire.

Part D: Managerial read

CFO avoids premature fixed cost. CTO gets learning without headcount politics. Decision memo records kill criteria for agency continuation.

You must choose between building in-house payments stack ($2.2M, 14 months) vs continuing Stripe at rising fees (projected +$600K/year at scale). Incomplete info: transaction volume growth 40–80% range; regulatory rules may change.

Tasks:

1. Classify door type for each option.
2. Build a three-scenario EV for in-house over 5 years (use illustrative numbers; show check line).
3. Propose a two-step barbell approach with kill criteria.
4. List two revisit triggers for a decision memo.

Solution

1. Door type

• In-house build: one-way (team, architecture, compliance ownership hard to unwind)
• Stay on Stripe: two-way at quarterly contract scale (can re-evaluate build annually)

2. Illustrative 5-yr EV (contribution after fees, $M)

In-house EV = 0.35×4 + 0.45×2.5 + 0.20×(−1) = 1.4 + 1.125 − 0.2 = $2.325M

Stripe EV = 0.35×1.5 + 0.45×2 + 0.20×1.8 = 0.525 + 0.9 + 0.36 = $1.785M

Check probabilities sum to 1 ✓

In-house wins EV but carries failure state; survival/regulatory risk must be weighed.

3. Barbell two-step

Step 1 (3 months, $120K): compliance + volume sensitivity study; negotiate Stripe volume tier.

Step 2 kill criteria: proceed to build only if (a) projected fee drag >$900K/yr at base volume with confidence ≥7/10, AND (b) regulatory scan shows no new licensing barrier in 12 months. Else renew Stripe 18 months.

4. Revisit triggers

• Volume CAGR falls below 30% two quarters (Stripe path cheaper)
• Regulator publishes new money-transmitter rule requiring license (build timeline/cost shifts)

In 250–350 words, explain why a decision can have positive expected value and still be wrong for your company. Use a fictional example with numbers. Include role of reversibility, barbell, and decision memo.

Solution

Positive expected value is an average across scenarios, not a guarantee in any single trial, and it ignores constraints like cash runway, reputation, and organizational focus. Suppose ArcVault, a fictional cybersecurity startup with 11 months cash, considers a $3M marketing blitz with 60% chance of +$8M NPV and 40% chance of −$4M NPV. EV = 0.6×8 + 0.4×(−4) = 4.8 − 1.6 = +$3.2M, attractive on spreadsheet.

Still wrong for ArcVault because failure state consumes cash and forces distressed financing, wiping out founders' control (survival constraint dominates EV). The blitz is closer to a one-way door: brand spend and team morale hit if campaigns fail publicly. A barbell alternative spends $400K on three segmented experiments with kill rules after 6 weeks; only scale if two segments exceed CAC payback targets.

Reversibility matters: experiments are two-way doors enabling learning; the $3M blitz is hard to unwind mid-flight. A decision memo before spending records problem statement ("need pipeline for Series B"), options, EV range, downside case, and revisit trigger ("pause if cash <9 months").

Process quality means the board can later judge whether rejecting the blitz was prudent even if one parallel universe shows the blitz would have worked. Decision quality is separable from luck.

Key takeaways

• Label decisions by reversibility; two-way doors deserve fast experiments, one-way doors deserve memos and red teams.
• Use expected value with ranges, sensitivity, and survival constraints, not fake precision.
• Barbell portfolios combine a protected core with capped experiments that generate learning.
• Document decisions before outcomes; review process quality to build organizational memory.

1. Identify one pending decision your team treats as two-way that is actually one-way (or the reverse). Relabel it and change the process accordingly.
2. What small experiment could you run this week to reduce uncertainty on your top priority?
3. Return to the unit page for assessments, or continue to Unit 4: Organizations and Execution.