Reference

Frameworks: Scenario Planning and Competitive Game Theory

Part of the War Gaming skill · loaded on demand from SKILL.md

Contents

  1. Pierre Wack and the Shell method (1985)
  2. Peter Schwartz's eight-step process (1991)
  3. The McKinsey adaptation: levels of uncertainty and the use and abuse of scenarios
  4. Choosing critical uncertainties: the procedure in detail
  5. Signpost design
  6. Brandenburger and Nalebuff: the Value Net and PARTS (1996)
  7. Game-theory mechanics for competitive dynamics
  8. Competitor-response modelling: Porter's profile and Coyne and Horn's three questions
  9. Business war games: structure and adjudication
  10. Robustness lenses: counting, minimax regret and downside

1. Pierre Wack and the Shell method (1985)

Pierre Wack led scenario planning at Royal Dutch Shell from the early 1970s and published the method in two Harvard Business Review articles: "Scenarios: Uncharted Waters Ahead" (September-October 1985) and "Scenarios: Shooting the Rapids" (November-December 1985). The first explains why forecasting failed and what scenarios are; the second explains how to make them change decisions.

The argument runs as follows. Forecasts are built by extrapolating the present and are usually right, which is why managers trust them. They fail at the moments of discontinuity, which are precisely the moments when a forecast would be most valuable. Shell's planners in 1971 and 1972 observed that the predetermined elements of the oil market (the growth of demand, the depletion of US reserves, the concentration of spare capacity in a small number of producing countries whose interests were not served by expanding output) made a supply-driven price shock not merely possible but likely, even though every price forecast in the industry showed stability. Their scenarios presented a world in which the producing countries constrained supply, and the 1973 oil shock arrived on schedule. Shell's managers, having inhabited that world in advance, moved faster than rivals.

Wack drew three lessons that define the method.

First, separate predetermined elements from uncertainties. A predetermined element is something already in the pipeline whose outcome is not in doubt over the horizon: the number of 18-year-olds in 2035 is set; the refineries under construction will be completed; the regulation passed last year will apply. Wack's image is the monsoon rain falling in the Himalayas: the flooding of the Ganges plain a few weeks later is predetermined, and anyone who bothers to look upstream can know it. The scenario team's first job is to look upstream. Uncertainties are the elements whose direction is open. Scenarios differ only in how they resolve the uncertainties; the predetermined elements appear in all of them.

Second, the target is the microcosm of decision makers, not the document. Wack's phrase is that unless scenarios change the mental image of reality held by the people who decide, they are "like water on a stone". His first-generation scenarios at Shell were intellectually sound and changed nothing, because they were presented as alternatives without consequence. His second-generation scenarios, which he called decision scenarios, were built backwards from the decisions managers actually faced and were designed to make a manager feel what it would be like to run the business in each world. The test of a scenario is whether a manager, months later, recognises the world arriving and acts.

Third, scenarios re-perceive rather than predict. The goal is not to be right about the future. It is to have already thought about the future that arrives, so that the organisation recognises it early and does not spend the first year of a discontinuity arguing about whether it is real.

The failure mode Wack built the method against is the single-point forecast, the consensus view that management commits to and that, being consensus, no one is positioned to notice failing.

2. Peter Schwartz's eight-step process (1991)

Peter Schwartz, who succeeded Wack at Shell and later co-founded Global Business Network, set out the working process in "The Art of the Long View" (1991). The eight steps are the standard operating procedure for a scenario project and this skill's workflow follows them.

StepWhat it asksThe reasoning behind it
1. Identify the focal issue or decisionWhat do you really want to know? Which decision would the scenarios inform?Scenarios built from the outside in (start with the macro world) drift into generality; built from the inside out (start with a decision) they stay useful
2. Key forces in the local environmentWhich factors close to the decision will shape its success: customers, suppliers, competitors, costs, technology?These are what a manager will recognise; the scenarios must be told in these terms
3. Driving forcesWhich larger forces (social, technological, economic, environmental, political) drive the key forces?This is where predetermined elements are found and where the team learns things it did not know
4. Rank by importance and uncertaintyWhich forces matter most to the decision and are least predictable?Only forces that are both important and uncertain can differentiate scenarios; important-and-certain forces are predetermined and belong in all scenarios
5. Selecting scenario logicsAlong which axes will the scenarios differ?The two or three top-ranked uncertainties define the logic; a 2x2 of two axes is the common form
6. Fleshing out the scenariosWhat is the narrative of each world, and how does each key force behave in it?A narrative forces internal consistency and gives managers something to inhabit
7. ImplicationsHow does the focal decision fare in each scenario? What does each scenario reveal about the decision's vulnerabilities?This step turns scenarios back into strategy
8. Selection of leading indicators and signpostsWhat would tell us, early, which scenario is unfolding?Without signposts the organisation cannot know when to act

Two practical points from Schwartz that the workflow relies on. The first is that a scenario should be a story with a plot, told with characters, chronology and causation, because people remember and reason with stories and do not with parameter sets. The second is that scenario names matter: a name that captures the logic becomes shorthand in later conversations and keeps the scenario alive after the workshop.

How the eight steps map onto this skill's workflow, and what was added:

Schwartz stepSkill stageAddition in this skill
1. Focal decisionStage 1The Courtney et al. uncertainty-level check, to confirm scenarios are the right tool
2 and 3. Key and driving forcesStage 2The explicit predetermined-versus-uncertain split from Wack, with a source per predetermined element
4. RankStage 3Numeric importance scoring from the driver tree where one exists
5. Scenario logicsStage 3Independence, discrimination and plausibility filters; composite axes
6. Flesh outStage 4Past-tense narrative; every rival's state written into every scenario; parameter table per scenario
7. ImplicationsStages 6 to 8Competitor-response matrix and war game, because implications depend on rival moves, not only on the world
8. SignpostsStage 5Seven-property acceptance test; hand-off to the scorecard

Exploratory versus decision scenarios. Wack distinguished first-generation scenarios, which explore the environment and teach the team, from second-generation or decision scenarios, which are rebuilt around the decisions managers face. Most scenario projects stop at the first generation and are ignored. The discipline here is to build the exploratory set quickly (Stages 2 to 4), then rewrite each narrative so that the focal options appear in it by name and the reader can see which option is winning in that world. A narrative that does not mention the options is not yet a decision scenario.

3. The McKinsey adaptation

Two McKinsey-authored pieces shape how strategy practices use scenarios.

Courtney, Kirkland and Viguerie, "Strategy Under Uncertainty" (HBR, November-December 1997), argue that treating all uncertainty alike leads either to false precision or to paralysis, and propose four levels of residual uncertainty after the best possible analysis:

LevelDescriptionAnalytical toolStrategic posture
1. A clear enough futureA single forecast is precise enough for the decisionThe standard strategy toolkit: market research, cost analysis, DCF with sensitivitiesUsually adapt; a single committed strategy is defensible
2. Alternate futuresA few discrete outcomes, often driven by a regulatory or competitive eventDecision analysis, game theory, option valuationShape or adapt; big bets or options on the discrete outcomes
3. A range of futuresA continuous range bounded by a few variablesScenario planning, latent-demand research, technology forecastingAdapt or shape; build flexibility
4. True ambiguityNot even the range can be identifiedAnalogies, pattern recognition, nonlinear dynamic modelsReserve the right to play; shape where possible

The three postures (shape the future, adapt to it, reserve the right to play) are the authors' terms; the tool assignments per level are theirs as well.

The point for this skill is that the scenario matrix is a level 2 and level 3 tool. At level 1 it wastes effort; at level 4 signposts and staged options matter more than a matrix, and the scenarios should be treated as provisional.

Charles Roxburgh, "The use and abuse of scenarios" (McKinsey Quarterly, November 2009), catalogues the ways scenario work fails in practice and the rules that prevent it. The failures: paralysis, where four futures are presented and no decision follows; narrow outcome sets that give false comfort; single-variable scenarios that are really sensitivities; and scenarios discarded the moment the world diverges from them. The rules: build at least four scenarios so there is no middle option to default to ("show three and people always pick the middle one"); push the extremes beyond what feels comfortable, in both directions; identify the highest-probability scenario as the base case and plan on it while covering the others; give scenarios memorable names of two to four words; include a maverick view; search for predetermined outcomes; identify no-regrets moves that hold across scenarios; and treat scenarios as living documents, adding one for each that is discarded. Roxburgh also warns that leaders should absorb the scenarios privately and communicate a single bold goal to the organisation; the scenario set is a leadership tool, not a message.

4. Choosing critical uncertainties: the procedure in detail

The two axes are the highest-stakes choice in the exercise. The procedure:

  1. Assemble the candidate list from step 3 of Schwartz. Phrase each as a question with a numeric range, so that "regulatory tightening" becomes "incremental compliance and capital cost for non-bank firms by the horizon year: 2 to 10 points of net revenue".
  2. Score importance on a 1-5 scale: how much does the resolution of this uncertainty move the value of the options? Where a driver tree exists, use the sensitivity output; a force that moves horizon-year EBITDA by more than 15% is a 5, less than 3% is a 1.
  3. Score uncertainty on a 1-5 scale: how wide is the plausible range, and how divided is informed opinion? A force on which two credible experts hold opposite views is a 5; one where the range is narrow and the consensus firm is a 1, and probably a predetermined element.
  4. Plot. Candidates scoring 4 or 5 on both are the shortlist.
  5. Test independence. Write each shortlisted uncertainty in one sentence and ask whether resolving one largely resolves the other. If tighter rules would themselves deter new entrants, and the reverse is not true, the two are partly dependent. Partial dependence is acceptable if you can name a mechanism for each off-diagonal cell (tight rules with incumbents entering anyway, loose rules with incumbents staying out). Full dependence is not; find another axis.
  6. Test decision relevance. For each end of each axis, name the option it favours. If both ends favour the same option, the axis does not discriminate.
  7. Test plausibility at the extremes. For each end, name a mechanism and a credible believer. Wack's warning applies: an extreme that the room dismisses as impossible discredits the whole set.
  8. Where several shortlisted uncertainties are correlated, combine them into a composite axis (for example "regulatory settlement: proportionate vs bank-like" can absorb both "capital requirements" and "interchange-cap extension"). The composite must still have a numeric threshold.

The threshold between low and high on each axis should be a number, chosen so that the two ends are genuinely different worlds rather than points on a gentle slope. Set the threshold where the options' relative ranking flips, if the model can tell you; otherwise set it at the point where informed opinion divides.

5. Signpost design

Schwartz's eighth step is the one most often skipped, and its absence is why scenario sets end up in drawers. A signpost is an observable event or measurement that, if it occurs, increases belief that a particular scenario is unfolding.

Properties of a good signpost:

  • Leading: it moves before the outcome it predicts. Building permits lead capacity; procurement RFP language leads channel shift; hiring patterns lead strategic moves.
  • External: observable without waiting for the client's own results. A fall in the client's revenue is a lagging consequence, not a signpost.
  • Cheap to monitor: a public filing, a trade publication, a quarterly survey, a permit database, a job board.
  • Thresholded: it has a number or a definite event, not "increasing interest".
  • Dated: a review date by which the signpost will have been checked and a judgement recorded.
  • Owned: a named person checks it.
  • Actionable: tripping it triggers a defined change in belief and a defined decision.

Design three to five per axis. Fewer than three and a single false reading dominates; more than five and monitoring lapses. Signposts should be handed to kpi-architecture for inclusion in the scorecard, with the scenario belief state reviewed at a fixed cadence, quarterly for most corporate decisions.

6. Brandenburger and Nalebuff: the Value Net and PARTS (1996)

Adam Brandenburger and Barry Nalebuff, "Co-opetition" (1996), apply game theory to business strategy. The book's central claim is that business is a game in which players simultaneously cooperate to create value and compete to capture it, and that most strategy errors come from seeing only the competition half.

The Value Net is the map of the game. The firm sits at the centre. On the vertical axis are customers and suppliers: the players with whom the firm transacts, resources flowing from supplier through the firm to the customer. On the horizontal axis are competitors and complementors: the players who affect how much customers and suppliers value the firm. The definitions are precise. A player is a competitor if customers value your product less when they also have the other player's product. A player is a complementor if customers value your product more when they also have the other player's product. The same test applies on the supplier side: a player is a competitor for suppliers if the supplier values your business less when the other player is present, and a complementor if more. Players can be both at once; a rival that shares your distribution network and lobbies with you on regulation is a competitor for customers and a complementor for rules.

The insight that matters for war-gaming is that the Value Net forces the analyst to list players whose payoffs change when the client moves, not merely the two or three rivals the client already fears. A supplier who loses volume, a complementor who gains, a customer whose switching costs fall: each is a potential mover in the game.

PARTS names the five elements of any game and therefore the five levers for changing it:

LeverWhat it isHow a firm changes it
PlayersWho is in the gameBring in a new customer, supplier or complementor; pay a player to enter or to stay out
Added valueWhat each player brings that the game loses without themRaise your own added value (differentiation, lock-in) or lower others' (build alternatives, reduce dependence)
RulesThe laws, contracts and norms that structure the gameMost-favoured-customer clauses, meet-the-competition clauses, long-term contracts, industry standards
TacticsMoves that shape how other players perceive the gameSignalling, commitments, reputation, deliberate opacity or transparency
ScopeThe boundaries of the gameLink games across markets or time; unlink them

A player's added value bounds what it can capture: nobody gets more from a game than they add to it. This is why war-game adjudication should ask, for any move that captures value, where the added value came from.

Allocentrism is the discipline Brandenburger and Nalebuff ask for: put yourself in the other players' shoes and ask what the game looks like to them, on their payoffs. The competitor-response matrix is the formal output of that discipline.

7. Game-theory mechanics for competitive dynamics

The concepts below are the working minimum; the sources are Dixit and Nalebuff, "Thinking Strategically" (1991), Schelling, "The Strategy of Conflict" (1960), and Ghemawat, "Commitment" (1991).

Sequential and simultaneous moves. In a sequential game one player commits visibly and the other responds; the rule is to look forward and reason back: anticipate the rival's best reply to each of your moves and choose the move whose best reply you can live with. Capacity additions, acquisitions and public price architectures are sequential. In a simultaneous game both players decide without seeing the other; the rule is to find the moves that are each player's best response to the other's, and to ask whether the resulting outcome is one you want to be in. Bids in a sale process and quarterly price setting are often simultaneous.

Credible commitment. A commitment changes a rival's behaviour only if it is credible, and it is credible only if it is costly or impossible to reverse. Schelling's insight is that reducing your own options can strengthen your position: a public, contractual, multi-year price to key customers removes your ability to match a rival's price cut on those accounts, and so removes the rival's incentive to cut. Ghemawat extends this to strategy generally: irreversible investments (capacity, brand, acquisitions) are what make strategies durable, and the same irreversibility is what makes them dangerous under uncertainty. The war game should test both edges.

Price wars as a repeated dilemma. Two rivals both prefer high prices, each is tempted to cut, and mutual cutting leaves both worse off. In a one-shot interaction cutting dominates; in a repeated interaction, tacit restraint can hold if each side can observe the other and punish defection. This is why price transparency (a new online channel, for example) tends to destabilise industry pricing: it shortens the observation cycle and lowers the cost of testing the other side.

An illustration in the form the war game uses. Two rivals each earn $100M of EBITDA at current prices. If one cuts price 5% and the other holds, the cutter gains share worth $20M and the holder loses $30M; if both cut, each loses $15M.

Rival holdsRival cuts
We hold100 / 10070 / 120
We cut120 / 7085 / 85

Each player's best reply to either move is to cut, so the one-shot outcome is 85 / 85, worse for both than 100 / 100. What changes the game is not persuasion but structure: a multi-year contract rate with key customers removes our option to cut on those accounts (a credible commitment), the rival then gains nothing by cutting against them, and the equilibrium on that revenue returns to hold / hold. The adjudication rule in the war game should reproduce this: contracted revenue does not respond to list-price gaps until the contract expires.

Signalling and reputation. A rival's past reactions are the base rate for its next reaction, and rivals know this, so some moves are made to build a reputation rather than to win the immediate exchange. A competitor that has matched every regional price cut for ten years is signalling; the war game should treat that history as a commitment.

First-mover and second-mover advantage. Moving first captures position and shapes the rival's options; moving second allows learning from the first mover's errors and avoids the sunk cost. Which dominates depends on how reversible the move is and how fast the rival can follow.

8. Competitor-response modelling

Two frameworks combine into the response matrix.

Porter, "Competitive Strategy" (1980), chapter 3, sets out the components of a competitor analysis: future goals (at all levels of management and across dimensions: growth, share, margin, risk, non-financial goals such as family control or technology leadership); assumptions (about itself and about the industry, including the blind spots those assumptions create); current strategy (what it is actually doing, in each function); and capabilities (strengths, weaknesses, ability to grow, quick-response capacity, ability to adapt, staying power). These combine into a response profile: what offensive moves the competitor is likely to make, and how it is likely to respond to moves by others, including what moves would provoke it most.

Coyne and Horn, "Predicting Your Competitor's Reaction" (HBR, April 2009), observe that most companies do not model rival reactions at all, and those that do usually assume the rival will react as the company itself would. They propose three questions, each answered from the rival's position:

  1. Will the competitor react at all? Consider whether it will notice the move (visibility, importance to its business), whether it will feel threatened (effect on its metrics, not ours), whether a response is a priority (management bandwidth, other battles), and whether it can respond in time.
  2. Which options will the competitor actively consider? Most firms evaluate only a few obvious counter-moves, typically those they have used before. The rival's history of responses is the base rate.
  3. Which option will the competitor most likely choose? The one with the best payoff on the rival's own metrics, judged with the rival's own assumptions and analytical habits, not with a full rational-actor model.

Applied together: the Porter profile supplies the objectives, assumptions and capabilities; the three questions convert them into a most-likely move for each of our options, with a lead time and an estimate of damage. The counter is then designed against that move, and the PARTS levers are checked for a way to change the rival's calculus before the move is made.

9. Business war games: structure and adjudication

Sources: Benjamin Gilad, "Business War Games" (2008); Daniel Oriesek and Jan Oliver Schwarz, "Business Wargaming: Securing Corporate Value" (2008); the wider practice described in competitive-intelligence literature. The standard structure:

  • Teams. A home team (the client), one team per key rival, a market team representing customers, channels and sometimes regulators, and a control team that runs the game, adjudicates outcomes and introduces injects. Four to six teams of four to six people is typical.
  • Preparation. Each team receives a briefing pack: the scenario in which the game is set, its own profile (objectives, capabilities, constraints), its payoff metrics, the rules of play and the menu of legal moves. Rival teams are briefed at least a week ahead so they can absorb the role.
  • Rounds. Three rounds, each representing a defined period (one year is common; a quarter for fast-moving markets). Each round: teams decide moves in writing within a time box; control resolves the interaction and publishes the state of the world; a short plenary surfaces what each team saw.
  • Injects. Events introduced by control at the start of a round to test specific assumptions: a rival's acquisition, a regulatory change, a demand shock, a technology release. Injects should come from the signpost list and the scenario narratives so that the game tests the scenario set as well as the options.
  • Adjudication. Control needs a stated rule, published in advance, for how moves translate into outcomes: how share moves with relative price and service, subject to capacity, lead times and switching costs; how costs respond to volume; how balance-sheet constraints bind. A simple model is better than an argument; the rule should be transparent enough that teams can anticipate it and therefore play strategically.
  • Debrief. Structured, and the most important part: what surprised each team, which moves worked and why, which options collapsed, which hidden options appeared, and what the home team would now do differently. Control keeps the surprise log and the decision list.

Gilad's emphasis is on realism over theatre: rival teams must play the rival's actual likely behaviour, based on evidence, not the most aggressive possible behaviour. Oriesek and Schwarz emphasise that the game's value lies in decisions taken afterwards, and that a game without a decision owner in the room is an expensive seminar.

10. Robustness lenses

Three lenses convert a payoff table into a verdict. None is sufficient alone.

Counting. In how many of the four scenarios is each option best, acceptable, or unacceptable? "Acceptable" must be defined against a stated floor (covenant, dividend, cash). An option acceptable in all four is a survivor; one that is best in one or two and unacceptable elsewhere is a bet.

Minimax regret. For each scenario, regret for an option is the best achievable payoff in that scenario minus the option's payoff. Each option's maximum regret is its worst case of regret across scenarios. The option with the smallest maximum regret is the one you will least wish you had not chosen, whichever world arrives. This lens favours survivors and penalises bets.

The arithmetic, on a small example with three options and four scenarios (horizon-year EBITDA, $M):

OptionS1S2S3S4WorstMax regret
X605030202025
Y504855454520
Z706035252520
Best in scenario70605545

X's regrets are 10, 10, 25, 25, so its maximum is 25. Y's are 20, 12, 0, 0, maximum 20. Z's are 0, 0, 20, 20, maximum 20. Y and Z tie on regret; the downside lens separates them. With a floor of 25, Z's worst case sits exactly on the floor and X's breaches it, while Y's worst case of 45 clears it comfortably. Y is the survivor, Z is a bet on S1 and S2, X is dominated. The point of showing the arithmetic in the deliverable is that a board member can check it in a minute and will trust the verdict more for having done so.

Why not expected value. Weighting scenarios by probability and taking the average is tempting and is what most finance teams do first. It fails here for two reasons. The probabilities are invented, and a 15% weight on the scenario that ruins the company is a way of not looking at it. And the decision is usually irreversible, so the client lives in one scenario, not in the average. Expected value is a fair tie-breaker among survivors; it should never be the lens that admits a bet.

Downside against tolerance. What is each option's worst-case payoff, and does it breach a constraint the client cannot breach? A bet whose worst case triggers a covenant default is not a bet the client can make, however attractive its best case.

The verdict combines the lenses into an adaptive plan: no-regret moves (good in every scenario), options bought cheaply now (rights of first refusal, pilots, partnerships) that convert a bet into a staged decision, and signpost-triggered decisions with dates. The verdict sentence always includes "unless", because a plan without triggers is a forecast.