scenario-planning

name	scenario-planning
description	Builds a small set of divergent, equally plausible futures on the two most consequential uncertainties — the Shell/GBN 2×2 scenario method — and stress-tests a strategy against every one of them, ending with robust moves and signposts to watch. Use when the horizon is 3–10 years and a single forecast would be false precision — "scenario planning for this market", "what are the plausible futures for X by 2030?", "build 2x2 scenarios", "alternative futures for the sector". Not for one dated prediction about a single trend (use `foresight`) or for a macro-factor inventory on its own (use `steep-pestle-analysis`).
license	MIT
metadata	{"category":"decision-strategy","method":"Scenario planning (Shell/GBN deductive 2×2)","origin":"Pierre Wack, Royal Dutch/Shell, 1985; Peter Schwartz / Global Business Network, 1991","version":"2.0.0"}

Scenario Planning

Scenario planning builds a small set of divergent but equally plausible stories about how the environment around a decision could unfold, then tests the strategy against each. It comes from Pierre Wack's planning group at Royal Dutch/Shell (Wack, Harvard Business Review, 1985), codified by Peter Schwartz and Global Business Network (Schwartz 1991; Ogilvy & Schwartz 2004) — the Shell/GBN tradition; the deductive 2×2 on two critical uncertainties is the GBN form. Its core principle: scenarios are not forecasts — no probabilities are assigned, no scenario is "most likely"; their job is to change the decision-maker's mental model and expose strategies that work in only one future. It prevents the official-future failure: betting on a single extrapolation and being surprised.

When to invoke

Invoke when:

• A decision depends on how the environment evolves over 3–10 years and at least two material uncertainties could swing either way — "how will enterprise AI procurement look in 2030?", "plausible futures for grid storage".

Do NOT invoke when:

• The question is "when will X happen?", has one uncertainty, or a horizon under 12 months — one dated prediction is foresight; a trend's rate and curvature is trend-analysis.
• The question is about the present — five-forces-analysis (industry structure) or position-competitor (market map).
• A probability on a specific variable is needed — bayesian-update or delphi-method estimate it; scenarios do not.
• Only the macro-factor inventory is wanted — steep-pestle-analysis produces it and feeds step 2.
• The consequences of one already-decided change are wanted — futures-wheel.

Procedure — the deductive 2×2 in seven steps

1 — Frame the focal question and horizon

Name the decision the scenarios must inform, the decision-maker and the horizon (3–10 years). "Commit €40 M to an in-house claims platform over 2027–2030?" is a focal question; "the future of AI" is not. A scenario set with no decision behind it is entertainment.

2 — Identify the driving forces

List 10–20 forces that could shape the outcome across social, technological, economic, environmental and political/legal categories (steep-pestle-analysis is the front-end). For each write which way it could swing over the horizon, not just what it is — "model capability: saturates vs keeps compounding", not "AI models".

3 — Rank by impact and uncertainty

Rate every force Low / Medium / High on impact on the focal question and on uncertainty over the horizon. High impact, low uncertainty = predetermined elements, constants in every scenario. High impact, high uncertainty = the critical uncertainties the scenarios branch on. Low-impact forces leave the axes but may reappear as texture.

4 — Choose two critical uncertainties as axes

Pick two that are independent (one end does not predict the other), distinct (they drive different dynamics) and bipolar (two clear, plausible ends each). If "axis X already implies axis Y" can be argued, the corners collapse into a diagonal — pick again. Two axes give four scenarios — the deliberate limit of the deductive form.

5 — Build and narrate the four scenarios

Each corner is a scenario with a memorable name carrying its logic, a present-tense vignette of the world at the horizon (3–5 sentences, internally consistent, containing the predetermined elements), two or three observable leading indicators that would show the world heading there, and the implication for the focal question. All four are equally plausible: "Do not assign probabilities to the scenarios. Do not categorize them as either the most or least likely" (Ogilvy & Schwartz 2004); Wack insisted on the same (Millett 2009). A corner nobody believes means the axes are wrong.

6 — Wind-tunnel the strategy

Test each candidate move against all four. Moves that hold in three or four are robust. Moves that pay off in one corner are options, taken only when that corner's indicators appear. Name at least one assumption the current plan makes that only one scenario supports, and set a monitoring plan: which indicator, how often, pointing to which scenario.

7 — Report

Fill the output template below, stating explicitly that the scenarios are equally plausible and carry no probabilities. Every implication must answer the focal question, not a general one.

Output template

## Scenario set — {focal question}

**Focal question:** {decision} · **Decision-maker:** {who} · **Horizon:** {year}
**Forces reviewed:** {N} · **Predetermined elements:** {element — why baked in}; …

**Axes:** X: {uncertainty} — {end A} ↔ {end B} · Y: {uncertainty} — {end A} ↔ {end B}
**Independence check:** {why X does not predict Y}

| | X = {end A} | X = {end B} |
|---|---|---|
| **Y = {end A}** | **{Name 1}** — {one-line logic} | **{Name 2}** — {one-line logic} |
| **Y = {end B}** | **{Name 3}** — {one-line logic} | **{Name 4}** — {one-line logic} |

Per scenario: **Vignette** {3–5 sentences, present tense} · **Leading indicators** {2–3 signposts} · **Implication** {what the decision-maker does}

**Robust moves (hold in ≥ 3 scenarios):** {move — scenarios}; …
**Options (one scenario):** {move — take when {indicator} appears}; …
**Assumption only {Name} supports:** {…}
**Monitoring plan:** {indicator — cadence — scenario}; …
**Plausibility statement:** all four scenarios are equally plausible; no probabilities are assigned.

Mandatory: focal question and horizon, predetermined elements, both axes with the independence check, four named scenarios with indicators and implications, robust moves, monitoring plan, plausibility statement.

Worked example

Illustrative case (all figures invented): a mid-size European insurer must decide whether to commit €40 M over 2027–2030 to an in-house claims-automation platform or rent one. Fourteen forces reviewed; two predetermined elements — EU AI Act high-risk obligations from August 2026, and claims volume growing about 3 % a year — appear in every scenario. Axes: X = model capability (saturating ↔ compounding); Y = regulatory regime for automated decisions (permissive ↔ restrictive). Independence check: 2024–2026 capability gains arrived under both permissive and restrictive regimes.

	Capability compounding	Capability saturating
Permissive regime	Cambrian Agents — thousands of vertical agents; AI-native insurers underprice incumbents by 15–20 % on expenses	Utility Plateau — models a commodity layer; competition on data, distribution and brand
Restrictive regime	Walled Garden — three or four certified vendors dominate; capability concentrated in the largest platforms	Frozen Winter — investment dries up; deployment stalls in compliance review

Leading indicators (examples): Cambrian — two or more agent-native insurers above 100,000 policies by 2028; Walled Garden — a certification scheme with under five approved vendors by 2029; Frozen Winter — sector AI capex down 30 % year on year.

Robust moves: build the claims-data pipeline and labelling capability (needed in all four); rent the model layer (wins in three corners; not fatal in Cambrian Agents). Option: a 20-engineer in-house agent team only if two Cambrian indicators appear before 2028. Assumption only Cambrian Agents supports: a proprietary model still differentiates in 2030. Monitoring: vendor-certification counts (quarterly), sector capex (annually), agent-native competitor policies (quarterly). All four scenarios are equally plausible; no probabilities are assigned.

Verification

Before the set ships:

• Independence re-checked: state one plausible way each corner could exist; near-duplicate corners mean correlated axes — re-pick.
• No probabilities, weights or "most likely" labels appear; the plausibility statement is present.
• Every scenario contains the same predetermined elements and at least two observable leading indicators.
• Every robust move was re-read against all four scenarios; each option names its triggering indicator.
• Focal question, decision-maker and horizon are stated; every implication answers that question.

Pair with adjacent skills

• steep-pestle-analysis — the coverage-disciplined driver inventory that feeds steps 2–3.
• foresight — one dated, falsifiable prediction for single-branch or short-horizon questions.
• cross-impact-analysis — checks the internal consistency of the set when forces interact strongly.
• futures-wheel — consequences within one chosen scenario; backcasting — the route to a preferred one.
• bayesian-update / delphi-method — a probability on a specific variable; the scenarios themselves stay unweighted.
• premortem-analysis — stress-tests the plan chosen after wind-tunnelling.
• Methodology counterpart: methodologies/foresight/scenario-planning.md — history, variants, facilitation detail.

Anti-patterns

• Do not assign probabilities to scenarios or rank them most-to-least likely — the Shell/GBN doctrine is explicit (Ogilvy & Schwartz 2004): a weighted set collapses back into a forecast, and the least expected corner is often the one that matters.
• Do not build on correlated axes — a diagonal is one story told twice — nor a best/worst/middle trio, where the middle becomes the forecast. Two axes, four corners.
• Do not name scenarios "A/B/C/D" — the name carries the memory.
• Do not stop at the narratives — robust moves, options and the monitoring plan are the deliverable.

Reference

• P. Wack, "Scenarios: Uncharted Waters Ahead," Harvard Business Review, vol. 63, no. 5, pp. 73–89, Sept.–Oct. 1985. https://hbr.org/1985/09/scenarios-uncharted-waters-ahead
• P. Schwartz, The Art of the Long View. New York: Doubleday/Currency, 1991. ISBN 0-385-26731-2.
• J. Ogilvy and P. Schwartz, Plotting Your Scenarios. Global Business Network, 2004 (first published in L. Fahey and R. Randall, eds., Learning from the Future, Wiley, 1998).
• K. van der Heijden, Scenarios: The Art of Strategic Conversation. Chichester: Wiley, 1996. ISBN 0-471-96639-8.
• P. J. H. Schoemaker, "Scenario Planning: A Tool for Strategic Thinking," Sloan Management Review, vol. 36, no. 2, pp. 25–40, 1995. https://sloanreview.mit.edu/article/scenario-planning-a-tool-for-strategic-thinking/
• S. M. Millett, "Should Probabilities Be Used with Scenarios?," Journal of Futures Studies, vol. 13, no. 4, pp. 61–68, May 2009. https://jfsdigital.org/wp-content/uploads/2014/01/134-AE04.pdf — argues probabilities can be used with care; documents the Wack no-probability stance it challenges.