How to Make Better Decisions

Classify the door before debating the paint. A reversible choice has a cheap route back: trial software, a two-week schedule, a small campaign. An irreversible choice creates lasting cost: signing a five-year lease, deleting customer data, hiring for a role you cannot define. Most teams deliberate backwards: weeks on button colors, one afternoon on a lease.

Can we undo this within 30 days for less than 10% of the cost?
YES → owner decides, small test, deadline today
NO  → gather evidence, seek dissent, stage commitment

When unsure, make it more reversible:
annual contract → paid pilot · full rollout → one branch
permanent hire → scoped contract · migration → parallel run

“Can we afford it?” is incomplete. The real question is: what is the best thing we cannot do if we choose this? Money, calendar space, attention, and reputation are all scarce. If no alternative is named, a proposal gets compared with nothing and almost anything beats nothing.

Choice: founder attends a three-day conference
Visible cost: $1,800
Hidden cost: three sales calls + release review + recovery day
Best rejected alternative: visit two existing customers

Decision sentence:
"We choose the conference over the customer visits because
three pre-booked buyer meetings are worth the release delay."

Expected value is not fortune-telling. It forces the odds and consequences into daylight. Multiply each outcome by its probability, then add them. Use ranges when estimates are weak. A positive average is still a bad bet if one loss can end the company, so check survival separately.

Campaign costs $10,000
25% chance of $60,000 contribution:  .25 × 60,000 = 15,000
75% chance of $0:                     .75 × 0      =      0
Expected value before cost:                         15,000
Expected value after cost:                           5,000

Then ask: can we survive losing the full $10,000?

Your project feels unique because you know its details. That feeling is not evidence. Start outside: how long did the last ten similar migrations take? What percentage of restaurants at this location survived? Then adjust for genuinely different facts. A specific story should modify the base rate, not erase it.

Forecasting a data migration
Inside view: "Our team is strong; six weeks."
Outside view: last 8 similar migrations = 11, 14, 9, 18,
12, 16, 10, 15 weeks → median 13 weeks

Plan: start at 13; subtract only for measured advantages;
add a rollback checkpoint before the old system is switched off.

Money and time already spent cannot be recovered. The decision begins now: if you did not own this project today, would you buy its remaining costs for its remaining benefits? “We have come too far to stop” is not resilience. Sometimes it is a refusal to admit that the evidence changed.

Bad frame: "We already spent $240,000, so we must finish."
Decision frame:
Remaining cost:     $110,000 + 4 months
Likely future value: $70,000
Alternative use:    onboarding repair worth ~$160,000

Choose from today forward. The $240,000 is gone either way.

Before commitment, tell the room: “It is six months later and this failed badly. Write the reason.” This frees people to name risks without sounding disloyal. Do it silently first so the senior person's theory does not become everyone else's. Then convert the top risks into owners, warning signals, and prevention.

Failure: supplier misses holiday inventory
Early signal: sample approval slips more than 5 days
Prevention: approve backup supplier before deposit
Owner: Maya                 Review: September 1

Failure: customers cannot import old data
Early signal: pilot import success below 95%
Prevention: migration rehearsal with 20 messy real files

After an outcome, memory cheats. A win becomes “obvious”; a loss becomes “unlucky.” Write a one-page journal before acting: what you know, what you believe, probabilities, alternatives, and what would change your mind. Review it after the outcome. Judge the process before judging the result.

Date / owner:
Decision and deadline:
Options considered:
Facts known / assumptions:
Expected outcomes with probabilities:
Base rate used:
Biggest downside / mitigation:
What evidence would reverse this decision?
Review date:

Later: outcome · surprise · process lesson · next calibration

Do not settle disagreement by averaging confidence or letting the loudest title win. Ask what fact would make each person switch sides. Often people disagree because they hold different hidden forecasts. Put those forecasts on paper and find the cheapest test that separates them.

Marketing: "Customers will pay $49."  → predicts >8% conversion
Sales:     "Above $29 will fail."      → predicts <3% conversion

Test: show equal qualified traffic $29 and $49 offers;
predefine sample size, refund rules, and decision threshold.
No changing the metric after seeing the result.

A stop rule says what evidence ends or changes a plan. Set it while calm. Once reputation, payroll, and hope are attached, every weak signal gets explained away. Good rules include a metric, threshold, date, and action. They can also trigger expansion, not only cancellation.

Pilot runs 8 weeks.
Stop: week-4 activation below 20% after onboarding fix.
Stop: support cost above $40 per active account.
Expand: retention above 60% and gross margin above 45%.
Owner: Priya. Review dates: weeks 2, 4, and 8.

Changing a rule requires new evidence, written before results.

Use this for decisions important enough to regret but too small for a committee. Ten honest minutes beats a polished deck built to defend a choice already made.

1. What exactly must be decided, by when, and by whom?
2. Reversible? If yes, what is the cheapest test?
3. What is the best alternative we give up?
4. Low/base/high outcomes, probabilities, and survival risk?
5. What happened in comparable cases: the base rate?
6. If starting today, would we still fund it?
7. It failed: why? Owner, signal, prevention?
8. What evidence would change our mind?
9. What are the stop and expansion rules?
10. Record the forecast now; review on a fixed date.