Decision Techniques – AI-Powered Decision Making | AI 102 Week 3 | How2AI

Decision Decomposition

When to use

You feel stuck in circles
The decision is framed as a false binary (“A or B?”)
The real constraint is unclear

What you should get

Clean separation of facts vs assumptions
Expanded option set
Success criteria you can actually measure

Common failure mode

Jumping into analysis before you've clarified what you're deciding.

Prompt Template

Before giving advice, ask up to 10 clarifying questions.

Decision (2–5 sentences): [paste]

Then decompose:
1) FACTS vs INTERPRETATIONS
2) ASSUMPTIONS (likely / uncertain / potentially flawed)
3) REVERSIBILITY (reversible vs hard-to-reverse)
4) ACTUAL OPTIONS (include non-obvious options)
5) CONSTRAINTS (and which may be self-imposed)
6) SUCCESS CRITERIA (3–5 measurable indicators + timeframe)

Mini Example: Hiring Decision

Decision: “Hire a full-time data analyst or use a contractor?”

Decomposition often reveals: the “true problem” is backlog variability, not headcount; hidden options like part-time hire, automation, upskilling, or shared services; and success criteria including backlog reduction, SLA, cost per analysis hour, and team satisfaction.

Pre-Mortem Analysis

When to use

You're about to commit resources (money, reputation, headcount)
Optimism bias is likely (“this will work because we're smart”)

What you should get

Failure narrative + root causes
Early warning signs (so you don't notice failure too late)
Kill criteria (so you don't keep going out of pride)

Common failure mode

Turning it into a generic “risk list” instead of a realistic chain of events.

Prompt Template

Conduct a Pre‑Mortem.

Plan/decision: [paste]

Assume it is 12 months from now and it clearly failed.
1) Failure narrative (sequence of events; concrete, not vague)
2) Root causes (top 5) with Likelihood / Impact / Detectability
3) Early warning signs for each root cause
4) Preventive actions for the top 3 root causes
5) Kill criteria: what conditions should make us stop, not "pivot"
6) Success narrative: If it succeeds, what enabling conditions made that happen?

Why Success Narrative Matters

Including a “success narrative” (item 6 above) ensures you don't only train attention on threats. It helps you identify the enabling conditions that would need to be in place for success, which you can then work to create or monitor.

Red Teaming Your Own Ideas

When to use

You have a preferred option
You need to defend a decision to stakeholders

What you should get

A steel-man case against your plan
Hidden costs + second-order effects
Conditions under which the plan is wrong

Common failure mode

Asking for critique, but prompting in a way that leads to agreement.

Prompt Template

RED TEAM this decision.

Decision: [state]
My reasoning: [paste]

Rules:
- Steel‑man the opposition (no strawmen)
- Assume my preferred option is emotionally appealing; compensate for that bias

Deliver:
1) Strongest case AGAINST
2) Hidden costs (financial, relational, opportunity, reputational)
3) Second-order effects (2–3 moves ahead)
4) Who loses + how they might react
5) Reversal test: if we were already in the opposite position, would we switch?
6) Verdict: proceed / proceed with modifications / reconsider
7) What would make this decision catastrophically wrong?

The Reversal Test

The reversal test (item 5 above) is one of the most powerful debiasing tools in this guide. It works like this: if you're considering switching from A to B, ask yourself — if you were already at B, would you actively switch to A? If the answer is “no, I'd stay at B,” that confirms the switch makes sense. If the answer is “no, I'd also stay at A,” then the desire to switch may be driven by novelty bias, frustration with the status quo, or sunk-cost avoidance rather than genuine improvement. It forces you to separate “this option is actually better” from “I'm just tired of where I am.”

Scenario Planning & Stress Testing

When to use

Uncertainty is high
Multiple external variables could swing outcomes

What you should get

A small set of distinct scenarios (not “best/likely/worst” only)
Signposts that tell you which scenario is unfolding
No-regret moves + adaptive actions

Common failure mode

Creating scenarios that are just mood labels (“good/bad”) without actionable triggers.

Prompt Template

Stress-test this decision across scenarios.

Decision/strategy: [paste]
Time horizon: [6 months / 1 year / 3 years]
Key uncertainties (2–3): [list]

Create 4 scenarios:
1) Base case
2) Upside case
3) Downside case (realistic, not apocalyptic)
4) Wildcard disruption

For each scenario:
- Describe in 3–5 sentences
- Rate decision performance (1–5)
- Identify what we'd wish we had done differently
- Provide 3 signposts (early indicators) that this scenario is unfolding
- Suggest 1 adaptive action we can take now that helps here without hurting others

Then:
- Identify 1–3 no‑regret moves that help across all scenarios
- Recommend monitoring cadence (weekly/monthly/quarterly) per signpost

Stakeholder Simulation

When to use

Decisions affect people (teams, customers, partners)
Implementation risk is high

What you should get

Likely reactions + objections
What stakeholders need to see to support
Communication sequencing

Common failure mode

Treating simulations as “mind-reading” rather than hypotheses to validate.

Prompt Template

Stakeholder Simulation (hypotheses to validate, not mind-reading).

Decision: [paste]
Stakeholders:
1) [Role/name] — priorities/concerns/context
2) ...
3) ...

For each stakeholder:
1) First reaction (emotion + thought)
2) Primary concern
3) What they'll say vs what they'll think (if different)
4) What would win them over (evidence, mitigation, trade, framing)
5) Risk if I mishandle communication
6) The best single sentence to say to them first

Then:
- Recommend announcement sequence and why order matters
- List 5 questions I should ask stakeholders to validate these hypotheses

Structured Frameworks on Demand

When to use

You want systematic tradeoff evaluation
You need to justify reasoning transparently

What you should get

A framework fit to your decision type
Clear criteria + constraints
Explicit tradeoffs

Framework Menu

Framework	Best For	What It Does
SWOT	Strategic positioning	Strengths/Weaknesses/Opportunities/Threats
Pre‑mortem	Risk assessment	Failure-first root causes + mitigations
RICE	Prioritization	Reach/Impact/Confidence/Effort
Second-order thinking	Consequences	“And then what?”
Opportunity cost analysis	Tradeoffs	Names what you're giving up
Weighted decision matrix	Multi-criteria	Scores options across weighted criteria
Inversion	Debiasing	“How would we guarantee failure?”
10/10/10	Emotional distance	Impact in 10 min / 10 months / 10 years
OODA loop	Fast iteration	Observe/Orient/Decide/Act cycles
Decision tree / Expected value	Uncertainty + payoffs	Quantifies branches + probabilities
Sensitivity analysis	Key drivers	Shows which assumption matters most
Regret minimization	Life/career choices	Minimizes future regret under uncertainty

The “Recommend a Framework” Prompt

Here's the decision (2–5 sentences): [paste]

1) Recommend 2–3 frameworks that fit this situation and why.
2) Apply the best one.
3) Then run a Sensitivity Analysis: which 1–2 assumptions drive the answer most?

The “Apply This Framework” Prompt

Apply [FRAMEWORK] to this decision: [paste]

Output:
- The full framework analysis
- The strongest insight revealed
- What this framework misses (limitations)
- What we should verify externally

Key Takeaways

✓Decision Decomposition prevents answering the wrong question — always separate facts from assumptions first
✓Pre-Mortems bypass optimism bias — always define kill criteria before committing, and include success narratives to identify enabling conditions
✓Red Teaming counters confirmation bias — use the Reversal Test to validate your reasoning and compensate for emotional appeal
✓Scenario Planning identifies signposts and no-regret moves — define early indicators that tell you which scenario is unfolding, not just mood labels
✓Stakeholder Simulation treats reactions as hypotheses to validate — not mind-reading, but structured preparation for communication
✓Ask AI to recommend the right framework — you don't need to know them all, just how to apply them with sensitivity analysis