---
name: scenario_planning
description: Run a full scenario planning workshop: frame a focal decision, map driving forces with STEEP, separate predetermined elements, pick two critical uncertainties as axes, write four narratives, wind-tunnel strategies, and set signposts with contingency triggers. Use when a decision depends on how the environment could change over 2 to 10 years, for strategy, capital allocation, product bets, policy, or risk work where a single forecast would be brittle. Do not use for a short term forecast, a probability estimate, or a present day market map.
---

# Scenario Planning

## Purpose

Build a small set of divergent, equally plausible stories about the world around a decision, then test the decision against each story. The point is to change how decision makers see the situation before they commit, and to leave behind early warning signs and pre-agreed responses.

Scenarios are not forecasts. Assign no probabilities. Name no scenario most likely. A best case and worst case pair is a disguised forecast, not a scenario set. Work in the Shell and Global Business Network tradition codified by Peter Schwartz, the step method set out by Paul Schoemaker, and the reframing emphasis of the Oxford approach by Rafael Ramirez and Angela Wilkinson.

Use this skill when:
- A decision is hard to reverse and its payoff depends on external change over 2 to 10 years.
- At least two material uncertainties could swing either way.
- The team keeps arguing from one shared picture of the future, usually last year extended in a straight line.

Do not use it when:
- The horizon is under 12 months. Use a forecast or a plan with risks.
- The user needs a probability for one variable. Use a forecasting or Delphi method instead.
- There is only one real uncertainty. Write a sensitivity analysis.
- The question is about current structure or competitors. Map the present first.

## Workflow

### 1. Frame the focal question

Write one decision question, one horizon year, and one decision maker.

Good form: Should [actor] commit [resource] to [option] through [year], given how [environment] could change?

Rules:
- One decision per exercise. Two decisions means two exercises.
- Put the horizon in the question. Most corporate work uses 3 to 10 years. Shorter horizons collapse into forecasting, longer ones lose decision relevance.
- List what is inside scope and outside scope: geography, customer set, product set.
- Write the official future in 3 lines: what the current plan quietly assumes. Every scenario must break at least one line of it.

Output of this step: a framing card (see references/templates.md).

### 2. Map driving forces

Brainstorm 15 to 30 external forces that could shape the focal question. Use STEEP as a checklist so one category does not crowd out the rest:

- Social: demographics, behaviour, work patterns, trust, inequality
- Technological: capability, cost curves, platforms, standards, adoption speed
- Economic: growth, rates, capital cost, input prices, labour supply
- Environmental: physical climate effects, resource limits, regulation driven by them
- Political: elections, regulation, trade, industrial policy, geopolitics. Add Legal as its own line when law moves separately from politics.

For each force write: name, one line description, and the two ways it could swing. A force with no swing is a fact, not a force.

Sources for forces: interviews with people who see the system differently, customer and frontline staff, regulators, suppliers, plus published data. Note the source next to each force. Do not invent forces from a generic list and stop there.

Helper: `bin/scenario_matrix.py` takes a CSV of forces with impact and uncertainty scores and produces the ranking, the axis candidates, and a blank 2 by 2 scaffold.

### 3. Separate predetermined elements, then score the rest

Predetermined elements: high impact forces whose direction is already largely locked in for the horizon, for example a demographic cohort that has already been born, physics of a technology already shipped, contracts already signed. They go into every scenario as shared ground. Test each candidate hard: if a thoughtful person could argue the opposite direction within the horizon, it is not predetermined.

Score every remaining force 1 to 5 on:
- Impact: if this swings, how much does the focal decision outcome change?
- Uncertainty: how honestly unpredictable is its direction or scale over the horizon?

Plot impact against uncertainty. The high impact, high uncertainty corner holds the critical uncertainties. High impact, low uncertainty forces join the predetermined list. Low impact forces are parked, however interesting they are.

Have two separate groups score, then reconcile scores that differ by 2 or more. The argument about the score is the useful part. Record it.

### 4. Choose two axes, and test them

Pick the two critical uncertainties that score highest, then run four tests before locking them:

1. Independence: the two axes must be able to move separately. If high on one almost forces high on the other, they are one axis in disguise. Pick a different second axis, or switch to the fallback below.
2. Difference: each end of each axis must describe a genuinely different world for the decision, not a small variation.
3. Plausibility at the extremes: all four combinations must be able to happen, even if one feels uncomfortable.
4. Relevance: crossing them must change what the decision maker would do. If all four quadrants imply the same action, the axes are wrong for this decision.

Name each axis as a spectrum with two labelled poles, for example: Demand: scarce and price driven vs abundant and convenience driven.

Fallbacks when the 2 by 2 fails:
- Correlated axes: use three uncertainties and prune inconsistent combinations with a consistency check (step 5), keeping 3 to 4 surviving combinations. This is the morphological route.
- Many equal uncertainties: cluster forces into 3 or 4 themes, write one inductive scenario per theme, then check the set covers the uncertainty space.
- One dominant uncertainty: build scenarios around its branches plus one wildcard.

Never force a neat matrix onto a messy problem to make the slide look tidy.

### 5. Check internal consistency

Before writing stories, check each quadrant for contradictions:
- List the key uncertainties in the quadrant and mark each pair: supports, conflicts, or unrelated. Schoemaker describes this kind of correlation matrix as a practical consistency test.
- A quadrant with a hard conflict must be repaired: change a secondary assumption, explain the mechanism that resolves the conflict, or drop the quadrant and take a scenario from the fallback route.
- Check each scenario against the predetermined elements. A scenario that quietly drops one is invalid.

### 6. Write the narratives

Write 600 to 1000 words per scenario, in this order:

1. Title: short, memorable, neutral. No good, bad, base, or worst labels.
2. End state snapshot: what the world looks like in the horizon year. Market, customers, regulation, technology, prices, winners and losers.
3. Pathway: how it got there in three phases, with turning points. A scenario with no pathway is a wish.
4. What it feels like to operate in it: a day or a deal inside that world.
5. Implications preview: what this world rewards and punishes.

Writing rules: present tense or past tense told from the horizon year, concrete details over adjectives, numbers only when they come from a source or are clearly labelled as illustrative assumptions inside the scenario. Keep all scenarios at the same level of detail and the same length. A favourite scenario written twice as rich as the others rigs the exercise.

Quality gate for the set. Each scenario must be: plausible (could happen, with a coherent pathway), distinct (a reader can tell them apart in one line), internally consistent (step 5 passed), relevant (it changes a decision), and challenging (at least one scenario makes the current plan fail). If a scenario fails a gate, fix it now.

### 7. Wind-tunnel the strategies

Test each candidate strategy, including the current plan and do nothing, against every scenario. Score each cell: works well, works with strain, or fails, with one line of evidence.

Classify the strategy:
- Robust: works acceptably in all scenarios.
- No regret move: low cost, helps in all scenarios, do it now.
- Contingent move: works in some scenarios only. Pair it with a signpost and a trigger (step 8) instead of committing now.
- Bet: excellent in one scenario, fails in others. Take it only knowingly, sized so the failure is survivable, or hedge it.

Also extract: capabilities needed in every scenario (build now), capabilities needed in one scenario (build an option: partner, pilot, hire on contract, modular design), and early moves that keep later choices open.

Prefer reversible moves early. Record maximum regret per strategy: the worst cell in its row, in plain terms. A strategy with an unacceptable worst cell needs a hedge or a smaller first step, whatever its average looks like.

### 8. Set signposts and contingency triggers

For each scenario define 3 to 5 signposts. A signpost is an observable event or metric that tells you this world is starting to happen, early enough to act.

Each signpost needs: indicator, data source, threshold or event definition, check cadence, owner, and lead time check (does the signal arrive before the decision deadline it protects?).

Turn the important signposts into triggers:
- If [signpost crosses threshold] by [date or review], then [pre-agreed action], owned by [name], costing [estimate], approved in advance by [decision maker].

Review the signpost dashboard on a fixed cadence, quarterly for most strategy work. At each review: which scenario is the evidence drifting toward, which contingent moves are now live, and do any scenarios need a rewrite because a predetermined element broke? Retire a scenario set when the horizon passes, a predetermined element fails, or the focal decision is closed.

### Workshop shape

- Half day version: steps 1 to 4 in the room with forces pre-worked, narratives drafted by small groups, wind tunnelling in plenary. Works for a narrow decision.
- Two session version (recommended): session one covers steps 1 to 5 and ends with quadrant sketches. Between sessions, writers draft full narratives. Session two covers steps 6 to 8. The gap improves the stories and lets people sit with an uncomfortable scenario.
- Participants: 8 to 15, mixed seniority and function, at least two people from outside the core team, one facilitator who does not advocate a strategy.
- Common failure modes: executives picking the scenario they like and planning only for it, axes chosen to produce a preferred quadrant, no owner for signposts so monitoring dies in a month, and treating the exercise as a one off event instead of a review cycle.

## Output Contract

Deliver, in this order:

1. Framing card: focal question, horizon, decision maker, scope, official future.
2. Driving forces table: force, STEEP category, swing in each direction, impact, uncertainty, source.
3. Predetermined elements list, with the reason each is treated as locked in.
4. Axis choice: the two axes with poles, the four test results from step 4, plus any fallback used and why.
5. Four narratives (or 3 to 4 from a fallback route), each following the structure in step 6.
6. Consistency check summary per scenario.
7. Wind tunnel matrix: strategy by scenario, with classification and maximum regret per strategy.
8. Signposts and triggers table, with owner and cadence for each.
9. Decisions made now, options bought, and the date of the first signpost review.

Use the templates in references/templates.md. A complete worked example is in references/worked-example.md.

## Operating Rules

1. No probabilities, no most likely scenario, no base case language in the scenario set.
2. Every factual claim about the present or the past needs a source. Numbers inside a future narrative are scenario assumptions: label them as such, never present them as data.
3. Equal plausibility does not mean equal comfort. Protect the uncomfortable scenario from being softened in editing.
4. Axes must pass the independence test. A tidy matrix with correlated axes is a defect, not a style choice.
5. Predetermined elements appear in every scenario. If one breaks in the real world, reopen the exercise.
6. Every contingent strategy needs a signpost, a threshold, an owner, and a pre-agreed action. A contingent strategy without a trigger is a hope.
7. Name owners and a review date before the workshop ends. Unowned signposts do not get watched.
8. Keep the set small: 3 to 4 scenarios. More than 4 and nobody holds them in memory during real decisions.