Introduction
Business planning sets targets and allocates resources; decision-making chooses among options when outcomes are uncertain. Both rely on forecasts, assumptions and trade-offs, and both suffer when assumptions go unexamined.
How AI applies
AI helps by forecasting, running many scenarios and making assumptions explicit. Language models can draft plans and challenge them; quantitative models estimate ranges of outcomes. AI supports the decision-maker; it does not take responsibility for the decision.
Applications and use cases
- Forecasting: revenue, demand and costs.
- Scenario analysis: explore best, expected and worst cases.
- Assumption review: list and challenge the premises of a plan.
- Pre-mortems: imagine failure and trace causes.
- Decision memos: structure options, risks and recommendations.
Technologies involved
Spreadsheets and Python for models, time-series forecasting libraries, Monte Carlo simulation, language models for drafting and critique, and BI tools to display results to decision-makers.
Implementation guide
- State the decision and the criteria for success.
- List key assumptions and give each a plausible range.
- Simulate outcomes across those ranges.
- Ask a model to challenge the assumptions, then evaluate its points yourself.
- Record the decision, its rationale and a date to review it.
Technical resources
A small Monte Carlo simulation shows a range of outcomes instead of one number. The distributions are assumptions that must come from your own data and judgment.
import numpy as np
rng = np.random.default_rng(42)
n = 10_000
customers = rng.normal(1000, 150, n) # expected customers, uncertain
price = rng.uniform(40, 60, n) # monthly price range
churn = rng.uniform(0.02, 0.08, n) # monthly churn
revenue = customers * price * (1 - churn)
p10, p50, p90 = np.percentile(revenue, [10, 50, 90])
print(f"Monthly revenue: 10%={p10:,.0f} median={p50:,.0f} 90%={p90:,.0f}")Business benefits
Organisations aim for decisions that consider uncertainty openly, better-documented assumptions and faster planning cycles. A simulation is only as good as its inputs.
Challenges and risks
- False precision: detailed outputs can hide weak assumptions.
- Automation bias: people may trust a model's recommendation too readily.
- Black swans: models trained on the past can miss unprecedented events.
- Accountability: a named person must own each significant decision.
- Sensitive data: financial plans require careful handling.
Examples
Illustrative scenarios: a leadership team reviews a range of revenue outcomes before approving a hire; a planner uses an assistant to list reasons a budget might be wrong; a board receives a decision memo with options, risks and review dates.