Scenarios Before Forecasts: Making Irreducible Uncertainty Explicit
Scenarios Before Forecasts: Making Irreducible Uncertainty Explicit
On this page
On large capital projects, investment committees, lenders and stage-gates still ask for a single cost–schedule–benefit story. Teams answer with a base case, then add optimistic and pessimistic variants that share the same hidden assumptions. Competing modes stay invisible. Price effects stay off the page. Regulatory and operating regimes sit frozen as if the world will not move. The preferred concept travels forward as if the future were already known. The failure named in this essay is not that people lacked a better spreadsheet. It is forecasting first under irreducible uncertainty — locking a most-likely narrative before plural futures have been made explicit.
1. When the Gate Pack Demands One Future
Picture a familiar capital-programme scene. A demand-sensitive asset — rail, port, power, process plant, or hospital catchment — needs a passenger, throughput or utilisation forecast before concept freeze. The pack shows a smooth base case. Upside and downside sit a few percent either side. What the pack does not show is that competing modes and price were never pressure-tested, or that the operating regime assumed in the estimate is only one of several plausible worlds. The gate treats the number as commitment-grade. The concept locks. Years later the traffic, price or policy path that actually arrived was never in the pack — not because the team was careless with arithmetic, but because a single story was asked to absorb what a small set of different futures should have made visible.
That pattern is not unique to one geography or one asset class. It appears whenever gate culture prefers one clean narrative over contested assumptions. On large capital projects, a project already encodes a view of the future — demand, commodity, technology, labour, geopolitics, policy, operating regime. Some of that uncertainty can be reduced with study. Some cannot be collapsed into one probability story at the moment of concept choice. Treating the irreducible part as if a decorated base case had absorbed it is how rooms invent false readiness.
Owners already recognise the symptoms in quieter language: “the business case assumed one world,” “we never stress-tested the concept against a different policy path,” “the estimate was careful inside a story that was never challenged.” This essay names the Decision Making discipline so the next gate can refuse a single-story forecast when uncertainty is still deep. Naming the failure as forecasting first — not as “we need more Monte Carlo” — gives sponsors language that is sharper than “uncertainty was high” and more honest than “the market surprised us.”
The vignette is deliberately light. It is not a named forensic of one airport or one corridor. It is the pattern: one extrapolated demand story, competing futures left off the page, concept freeze on a number that looked careful. Owners who have lived a demand miss will recognise the mechanism without needing a court record. The lesson for the next gate is procedural. Before the forecast cell is allowed to harden, ask which futures were never written — and whether those futures would have changed the concept.
2. Scope of This Essay
This essay stays inside scenarios before forecasts. It defines forecasts and scenario sets as different jobs, shows why irreducible uncertainty must be made explicit before a commitment number is asked to carry a gate, names the most-likely trap that converts three cosmetic cases into one story, offers a front-end practice loop that challenges narratives rather than decorating them, transfers one Oxford-linked idea — a plausibility set as a learning service, not a prediction product — and stacks that work with frame ownership, outside-view forecasting, slow reversible sequencing, and front-end definition before authorization.
Adjacent objects deserve a sharp boundary so this spine stays clean. Ambiguity diagnosis, contextual-versus-transactional sorting, learner naming, and the full reframing–reperception cycle belong in the ambiguity toolkit; this essay owns plural futures that stop a forecast from doing the wrong job. Reference-class forecasting owns class data and uniqueness bias once the object of estimation is clear; here the stack is after scenarios clarify what is being estimated — not a rebuild of base-rate tables. Reversible-versus-irreversible sequencing owns when to keep learning open; scenario work lives in the reversible window without renaming scenarios as a sequencing slogan. Front-end loading owns definition completeness before authorization; scenarios improve concept and context judgement before a single narrative is scored complete. Hypothesis and issue-tree discipline own structured analysis once the question is set; scenarios can supply rival futures to test, not a MECE method under another name. Frame ownership before choose owns whether the decision question is the right one; scenarios may reopen a badly owned question lightly. Quality at entry and strategic-versus-tactical performance own whether the concept itself is right; scenarios are one scant-information method that serves concept choice. Needs assessment owns normative, market and interest-group depth; forecasting-versus-backcasting appears here only as a cousin warning against one extrapolated story. New-information update theory, expert elicitation depth, and adaptive heuristics sit in remaining neighbours. Coalition devices under surprise and value allocation among claimants own post-commit and deal close. This essay may surface contested worldviews; those essays own the devices.
What belongs here is Decision Making under deep uncertainty: make irreducible uncertainty explicit with a small set of genuinely different futures before any class-based or bottom-up forecast is asked to carry a commitment.
3. Two Jobs — Forecasts Versus Scenarios
Working distinction for large capital projects:
-
Forecast — a commitment-grade statement about cost, schedule, demand or benefit, ideally disciplined by class data and estimation hygiene when the object of estimation is clear enough. Job = how big / how likely inside a usable frame.
-
Scenario set — a small number of plausible narrative futures that surface driving forces, rival worldviews, and what remains unknowable. Job = which futures must the concept and the estimate survive, and what uncertainty cannot be collapsed into one number yet.
Wrong tool class: asking Monte Carlo or a tighter bottom-up pack to fix a meaning contest or an unstated future; asking a scenario workshop to replace a class-based cost estimate once the frame is shared. A useful cousin distinction — forecasting as trend prolongation versus backcasting from a desired future with instruments — is not the same as plural scenarios, but it carries the same warning against one extrapolated story dressed as inevitability.
Research interpretation for owners: scenarios and forecasts are complements in sequence, not synonyms and not rivals for the same gate cell. A room that treats them as interchangeable will either skip the futures conversation or skip the class discipline. Both failures show up later as surprise that was available earlier as contested assumption.
Figure 1: Forecasts answer “how big inside a frame.” Scenarios answer “which futures must we survive — and what stays irreducible.”
Owners who ask the forecast cell to absorb worldview conflict will get false precision. Owners who ask a scenario workshop to replace an outside-view cost check once the frame is shared will get theatre without a commitment number. Keep the jobs separate. Sequence them honestly.
The distinction also clarifies what “uncertainty work” is not. It is not a thicker risk register that still assumes one future. It is not a PMI-style complexity label renamed as scenario planning. It is not probability theatre when probability cannot yet be grounded. On large capital projects, the first honest move under deep uncertainty is to say which futures the concept must survive — and which uncertainties remain irreducible — before the estimate pack pretends those questions are closed.
4. Making Irreducible Uncertainty Explicit
On large capital projects a project already encodes a view of the future. Demand paths, policy regimes, commodity bands, technology shifts, labour markets, geopolitics and operating models sit inside the concept whether or not the gate pack names them. Some of that uncertainty can be reduced with study, survey and engineering. Some is irreducible at the moment of concept choice: the world will not wait for a perfect distribution, and the decision still has to be made.
Scenario planning’s first service in that window is not prediction accuracy. It is making the irreducible part visible so the gate pack cannot pretend a single base case has absorbed it. Under scant information, plural scenarios improve appreciative judgement — the quality of how the situation is seen — by ordering driving forces, connecting disciplinary expert silos to the live Business Idea the project encodes, and holding multiple worldviews in view rather than averaging them into false precision.
Capital translation follows. Sponsors who skip this step do not “save time.” They smuggle unstated futures into the estimate and the scope freeze. The calendar looks protected. The commitment is softer than it appears. When the world that arrives was never in the pack, the organisation discovers that it authorized a concept against one story while several other stories were always live.
Boundary matters. Irreducible uncertainty is not a cue to skip appraisal. Scant information is a cue for structured methods, not an excuse for silence. Early flexibility is highest when uncertainty is highest — concept work belongs in that window; strategic-performance essays own whether the chosen concept is the right one. Here the cut is precise: make what cannot yet be collapsed explicit, so later forecasting and definition gates know what they are standing on.
Appreciative judgement also explains why expert silos must be connected early. Commercial, technical, regulatory, social and operations views of the same Business Idea often disagree. Averaging them into one base case looks collegial. It is usually a meaning contest postponed. Scenario work that brings those silos to the same table — and keeps rival futures visible — improves the quality of seeing before the quality of numbering.
Making irreducible uncertainty explicit also changes what “complete” means at an early gate. Completeness is not a thicker appendix of the same story. Completeness is a short, honest account of what remains unknowable, which driving forces were ordered, which worldviews were held in view, and which assumptions the preferred concept still depends on. A pack that cannot name those items is not ready for a commitment-grade forecast, no matter how many decimal places the estimate carries.
5. The Most-Likely Trap
Named failure: teams invent a “most likely” case, then add optimistic and pessimistic variants that share the same hidden assumptions — competing modes ignored, price effects ignored, regulatory regimes unchanged, interface risk assumed away. Imagination cannot enumerate every failure path. A lone most-likely narrative understates the combined weight of unfavourable paths. Inside-view mental simulation feels rigorous because it is detailed; it remains a single story about this project.
The cosmetic three-case pack is still one story. Tone changes. Kind does not. The preferred concept survives every branch because every branch was built from the same unstated world. That is not scenario discipline. It is decoration.
The remedy at this stage is not yet a finer spreadsheet. It is a small set of genuinely different futures that change what would have to be true for the concept to hold. Only after those futures have been made contestable should the room ask what can be estimated — and with which class — under each surviving frame.
Qualification: this is not a substitute for outside-view cost and schedule forecasting. Scenario inadequacy explains why a decorated base case fails as decision support. Class-based forecasting corrects class-blind estimates once the object of estimation is clear. Keep the footnotes separate. Use both. Sequence them.
Figure 2: A decorated base case still hides shared assumptions. Plural scenarios make those assumptions attackable.
Owners who pride themselves on “we always show upside and downside” should ask a colder question: do those branches change the world, or only the mood? If competing modes, price, policy and operating regime never move, the pack is still forecasting first under a single story.
Inside-view mental scenarios are seductive on large capital projects precisely because the asset feels unique. Teams know the route, the process train, the catchment, the interface list. Detail feels like diligence. The warning from executive forecasting research is that detail without a contest of futures still understates the weight of paths the imagination did not enumerate. Plural scenarios do not pretend to list every failure mode. They refuse the fiction that one most-likely narrative has already done that work.
6. Front-End Scenario Practice — Challenge, Don’t Decorate
Practical move at concept and pre-authorization on large capital projects:
- State the Business Idea the project encodes — what must be true in the world for this concept to create value.
- Surface driving forces and contested assumptions across disciplines — commercial, technical, regulatory, social, operations — connect silos instead of sequential mark-ups.
- Build a small set of narrative scenarios (typically a handful, not a catalogue) that are different in kind, not only in tone.
- Use scenarios as hypotheses: invite experts and challengers to show where each story is incomplete, inconsistent or wrong — attack logic; do not defend the preferred deck.
- Only then ask what can be estimated (and with which class) under each surviving frame — hand quantitative hygiene to outside-view and controls practice.
Do not invent mandatory matrices as ritual, firm-branded seven-step products, or glossary theatre. The mechanism is owner practice: Business Idea, driving forces, small set, expert challenge, then estimate.
Where ideas associated with Oxford-linked scenario practice are drawn, one transfer is enough for this section: a scenario set can serve as a plausibility set for learning, not a predicted future. Full learner naming, contextual sorting protocols and reframing cycles stay with the ambiguity toolkit. Return immediately to project front-end practice.
Figure 3: Scenarios before forecasts: connect silos, build a small set, invite challenge, only then estimate.
Challenge culture is the hinge. A workshop that produces pretty stories nobody is paid to attack is scenario theatre. A room that exempts the preferred concept from attack has already chosen decoration. Experts earn their seat by showing where a narrative fails — incomplete, inconsistent or wrong — not by polishing the preferred path until it looks inevitable.
Silos connect poorly when mark-up is sequential. Commercial comments first, then technical, then regulatory, then operations — each layer polishing a preferred story rather than stressing a shared Business Idea. The practice loop above collapses that sequence into a joint surface of driving forces. The output is not consensus theatre. The output is a small set of narratives that make disagreement usable before money hardens.
7. Plausibility Sets, Not Prediction Products
Ideas associated with the Oxford Scenario Planning Approach treat a scenario set as a service to decision readiness under turbulence: plausibility over false probability when probability cannot be grounded; scenarios not manufactured from forecasts, though they can discipline what a later forecast assumes. On large capital projects the transfer is practical and limited: use a small plausible set to test whether the preferred concept and the planned estimate still make sense if the contextual world shifts — not to claim the set predicts the future, and not to rename front-end-loading gates or reference-class forecasting as scenario method.
Hard non-overlap: this essay does not teach ambiguity taxonomies, contextual-versus-transactional sorting protocols, learner-naming workshop design, or reframing–reperception cycles. Those belong elsewhere. One credit for the plausibility-set idea, labelled as transfer into capital Decision Making, is enough. The Decision Making claim that remains is local: under irreducible uncertainty, plural futures before commitment numbers.
Lineage teasers from classic strategy scenario work are optional colour only: forecasting adequacy can collapse under turbulence. They do not authorise invented case forensics or secret corporate recipes. Stay with the project front-end mechanism — Business Idea, driving forces, challenge, then estimate — and the most-likely trap that a decorated base case still fails to fix.
8. What Comes After Scenarios (and What Does Not)
Sequence for Decision Making on large capital projects under deep uncertainty:
- Plural scenarios make irreducible uncertainty and rival assumptions explicit (this essay).
- Owned decision question / frame — if the question itself is wrong or unowned, framing discipline applies.
- Outside-view / reference-class forecast — once the object of estimation is clear, class data discipline the commitment number.
- Think-slow sequencing — keep learning reversible until the plan is tested; then compress irreversible exposure.
- Front-end definition — progressive definition completeness before full-funds authorization.
- Quality at entry / right concept — strategic performance of the chosen concept — scenarios serve it; they do not replace it.
Also keep apart: complexity vocabulary that uses the word “scenarios” without doing scenario planning; Monte Carlo without a shared frame; and optimism-bias uplifts used as decoration on a guess.
Figure 4: Stack the tools; do not merge them. Scenarios make uncertainty explicit before commitment numbers and definition scores carry the gate.
The honesty test is sequencing. A room that jumps to class forecasting while futures are still unstated is numbering a meaning contest. A room that runs scenarios forever without ever estimating inside a shared frame is avoiding commitment. Stack. Do not merge. Do not skip.
The stack also protects neighbours from being hollowed out. Outside-view forecasting stays powerful when it is not asked to choose the concept. Front-end definition stays honest when it is not scoring a single unchallenged narrative as complete. Slow sequencing stays useful when scenario work has already named what must stay reversible. Merge the tools and each one loses the job it was built to do.
9. Practice on Large Capital Projects
Owner and PMC manifestations (transfer, not invented benchmarks):
- Concept / early definition gates: require a short scenario set that changes assumptions in kind before accepting a single business-case narrative; record which uncertainties remain irreducible.
- Demand-, price- or policy-sensitive assets (rail, ports, power, process plants, hospitals with catchment risk): forbid one extrapolated forecast as the only future; show at least what breaks the concept under alternate contexts.
- Joint-venture and alliance steering: use scenarios to surface worldview conflicts early rather than averaging them into a false base case.
- Estimate packs: separate “futures we are testing” from “class-based estimate under the chosen frame”; do not let a scenario workshop replace reference-class discipline, and do not let class forecasting pretend to have chosen the concept.
- Challenge culture: experts paid to attack scenario logic; preferred concept not exempt.
Director and investment-committee questions that can fail the pack:
- Which assumptions are shared across every case we labelled upside, base and downside?
- What would have to be true in the world for this Business Idea to hold — and which of those truths are still contested?
- Which uncertainties are we calling reducible that are still irreducible at this gate?
- Under which alternate future does the preferred concept break — and is that future in the pack?
- Are we about to authorize a forecast that is careful inside a story nobody was allowed to attack?
Trap list (short): scenario theatre (pretty stories, no decision use); best/base/worst clones; probability theatre under gaps that are still Knightian in spirit; locking detailed engineering on the optimistic branch; treating “we ran scenarios” as a tick without irreducible uncertainties named; renaming scenario work as front-end loading or as outside-view forecasting; inventing hit-rate thresholds or NPV multiples that are not in ingested evidence for this essay.
Practice also means refusing false substitutes. A thicker risk register that still assumes one future is not plural scenarios. A Monte Carlo cloud around an unchallenged base case is not making irreducible uncertainty explicit. A workshop that never changes the preferred concept’s world is decoration. The moves above are allow/deny rules for the forecast cell. They work only if someone in the room is authorised — and willing — to stop the clock when the future is still a single untested story.
Application across asset classes stays labelled as transfer. Rail and ports make demand and mode competition vivid. Power and process plants make commodity, technology and regulatory regime vivid. Hospitals and catchment-sensitive social infrastructure make utilisation and policy vivid. Joint ventures make worldview conflict vivid. The Decision Making rule does not change with the asset label: plural futures before the commitment number; irreducible uncertainty named; challenge before decoration.
10. Closing — Scenarios First, Then Forecasts
Under irreducible uncertainty on large capital projects, the cheapest honest move is to make plural futures explicit before asking any forecast to carry a commitment. Protocol in one breath: name the Business Idea → surface driving forces and contested assumptions → small scenario set → expert challenge → name what remains irreducible → then class-based or bottom-up forecast and definition gates.
Forward to neighbours as ideas only: use the ambiguity toolkit when meaning itself is contested; use outside-view discipline when numbers dominate inside a shared frame; use slow reversible sequencing when lock timing is the issue; use front-end definition when completeness before authorization is the issue; use quality-at-entry thinking when the concept itself may be wrong. Forthcoming scant-information titles on needs before numbers, new information that should reopen the case, expert judgement limits, and fast heuristics under scant information sit beside this spine without replacing it.
The corrective is not more slides after the gate. It is plural futures before the commitment number — contested on purpose, attacked by experts, with irreducible uncertainty named in the pack. Forecasts still matter. They matter more when they are no longer asked to hide the world.
References
- Williams, T. M., Samset, K., & Sunnevåg, K. J. (eds.). (2009). Making Essential Choices with Scant Information. Palgrave Macmillan. (Primary — especially van der Heijden Ch.4 on scenario planning, appreciative judgement, irreducible uncertainty, Business Idea, driving forces, expert challenge and multi-perspective worldviews; Williams overview; Klakegg needs chapter on forecasting versus backcasting and passenger-forecast cautionary; Samset Ch.2 quality-at-entry adjacency; Flyvbjerg Ch.8 outside-view / reference-class boundary.)
- Lovallo, D., & Kahneman, D. (2003). “Delusions of Success: How Optimism Undermines Executives’ Decisions.” Harvard Business Review, 81(7). (Supporting — scenario inadequacy; most-likely scenario understates combined unfavourable paths; inside view versus outside view.)
- Ramírez, R., & Wilkinson, A. (2016). Strategic Reframing: The Oxford Scenario Planning Approach. Oxford University Press. (Idea-only credit — plausibility set as learning service, not prediction product; capital application labelled transfer. Full method depth owned elsewhere.)
- Ramírez, R., Churchhouse, S., Palermo, A., & Hoffmann, J. (2017). “Using Scenario Planning to Reshape Strategy.” MIT Sloan Management Review. (Light Supporting practice notes only if read as idea transfer.)
- Flyvbjerg, B., with Gardner, D. (2023). How Big Things Get Done. Currency; and Flyvbjerg, B. (2006). “From Nobel Prize to Project Management: Getting Risks Right.” Project Management Journal. (Boundary — outside view / reference-class forecasting after frames stabilize; do not rebuild those methods here.)
- Merrow, E. W., & Nandurdikar, N. S. (2018). Leading Complex Projects. Wiley. (Light adjacency — definition complete before authorization benefits from scenario conversation before locking one narrative; not a rename of front-end loading as scenario method.)
- Wack, P. (1985). “Scenarios: Uncharted Waters Ahead” / “Shooting the Rapids.” Harvard Business Review. (Teasers only — forecasting adequacy under turbulence; no invented full-article claims.)
- PMI. (2014). Navigating Complexity: A Practice Guide. (Adjacent vocabulary only — “complexity scenarios” are not synonymized with scenario planning here.)