AI Scenario Planning for Executives

Forecasting AI has a poor record: the timelines, the costs, and the winners have all surprised the people paid to predict them. Scenario planning does not require being right about the future; it requires making decisions that survive several of them. This guide gives executives a compact method: four variables, three or four scenarios, decisions tested against each, and the moves that hold up regardless.

Why scenarios rather than forecasts?

Because the variables that matter move faster than planning cycles and in directions experts disagree about. A forecast bets the program on one path. A scenario set asks which decisions are good on every path, which are good on some, and which depend on one, and it allocates confidence accordingly. FISTA's how to lead through AI uncertainty piece sets out the leadership stance; this piece gives the method.

What are the four variables?

Variable	Range	Why it matters
Model capability and cost	Rapid improvement and falling prices, or plateau and stable prices	Determines what agents can do and what they cost per task
Regulation	Strict and fast, or light and slow, by jurisdiction	Determines controls, records, and which uses are permitted
Competitor adoption	Aggressive deployment, or cautious and incremental	Determines pricing pressure and the cost of waiting
Human response	Customers and employees accept agents, or resist them	Determines where agents can be deployed and how fast

Most other uncertainties, such as vendor survival or framework popularity, are downstream of these and can be handled with replaceability rather than scenarios.

What does a scenario set look like?

Three or four named futures built from the variables, each a short paragraph:

• Fast and open: capability improves quickly, prices fall, regulation is light, competitors deploy aggressively, customers accept agents. Speed is the constraint.
• Constrained: capability improves, but regulation is strict and customers cautious. Governance records and trust design are the constraint.
• Incremental: capability plateaus, adoption is gradual, competitors are cautious. Discipline and cost control are the constraint.
• Disrupted: a competitor resets the market with agent-delivered service or outcome pricing. Response speed is the constraint.

Each scenario should be plausible enough that a reasonable executive could believe it, and different enough that decisions diverge across them.

How are decisions tested?

List the major AI decisions on the table: platform investment, vendor commitments, committed outcomes, workforce moves, pricing changes. For each, ask how it fares in every scenario:

Decision	Fast and open	Constrained	Incremental	Disrupted	Verdict
Governed platform on open standards	Good	Good	Good	Good	Robust: do it
Multi-year single-vendor model contract	Good if they lead	Neutral	Poor if overpaying	Poor if wrong vendor	Brittle: restructure for exit
Evaluation sets for committed processes	Good	Good (evidence for regulators)	Good	Good	Robust: do it
Wait for the market to settle	Poor	Neutral	Acceptable	Very poor	Brittle: reject
Outcome pricing on automated services	Good	Good with controls	Slow	Essential	Mostly robust: sequence it

Decisions robust across all scenarios are made now. Decisions robust across some are restructured until they are, often by adding reversibility. Decisions that depend on one scenario are deferred with a date or rejected.

What moves are robust?

The same list keeps appearing: a governed platform on open standards; replaceable models behind a gateway; specifications and evaluation sets; data readiness; governance records that regulators are converging on; redesigned roles and reskilling; and supervised deployment on defined work. Each pays off whether capability moves fast or slow and whether regulation is strict or light. The AI competitive advantage explained piece explains why these compound in every future.

What moves are brittle?

Single-vendor bets without exit terms; strategies that assume regulation stays light or becomes strict; waiting for stability; cutting headcount ahead of proven agents; and building generic infrastructure that standards will commoditize. Each pays off in one scenario and fails in others. The LLM vendor lock-in guide covers the first; the how to think about AI and headcount guide covers the fourth.

How is the exercise run and maintained?

A half day with the executive team, using the four variables to write the scenarios and the decision table to test the current agenda. Then twice a year, or when a variable resolves, a shorter session retires scenarios that have played out, adds any that new evidence suggests, and re-tests deferred decisions. Keep it short; it is a decision tool, not a planning ritual. The how to run an AI executive offsite guide describes a session format that can include this.

What should executives ask?

• Which of our AI decisions would fail if regulation tightened sharply next year?
• Which would fail if a competitor offered outcome pricing next quarter?
• What are we deferring that is actually robust and should be done now?
• What have we committed to that depends on a single future?
• When did we last retire a scenario that resolved?

How can FISTA Solutions help?

FISTA Solutions builds the robust moves: governed platforms on open standards, replaceable model layers, evaluation sets, data readiness, and governance records, through its AI enablement and AI agents practices, and facilitates scenario sessions as an independent practitioner. Since 2017, FISTA has delivered 150+ projects for 50+ companies across 12+ countries.

To test your AI agenda against a scenario set in a single session, talk to FISTA on WhatsApp, or read AI strategy mistakes executives make.