Research & Analysis / Trend Analysis
Evaluate prediction intervals and tell you how much to trust a 30/60/90-day forecast.
Why This Prompt Exists
Forecasts are presented as single numbers — but they’re really ranges. Treating them as certainties leads to disaster.
You get:
- ordering inventory based on a point forecast, then getting stuck with excess
- hiring based on a forecast that was 30% too high
- budgeting with false precision (we’ll make exactly $10.2M, not $10.1 or $10.3)
- blaming the forecaster when actuals fall within the predicted range
- overconfidence in near-term forecasts and underconfidence in long-term
But forecasts have uncertainty signatures:
- near-term (1-30 days): relatively narrow, but outliers still happen
- medium-term (1-3 months): uncertainty grows, ranges widen
- long-term (3-12 months): very wide ranges, trends matter more than levels
- forecastability: some metrics are inherently more predictable than others
- model fit: how well does historical data predict the future?
Without confidence assessment, you treat guesses as facts.
This prompt tells you how much to trust a forecast — and what range to plan for.
The Prompt
Assume the role of a forecasting expert who quantifies prediction uncertainty. Your task is to assess the reliability of a forecast and provide actionable confidence intervals. Generate: 1. FORECAST SUMMARY - Point forecast for [horizon]: [value] - Method used (if known) - Historical accuracy of this method (MAPE, RMSE) 2. CONFIDENCE INTERVALS - 50% interval (likely range): [lower, upper] - 80% interval (plausible range): [lower, upper] - 95% interval (almost certain range): [lower, upper] 3. UNCERTAINTY DRIVERS - Primary sources of uncertainty (seasonality, trend, volatility) - External factors not in model (competitors, economy, weather) - Known upcoming events that could shift forecast 4. FORECAST HORIZON ASSESSMENT - How does uncertainty grow with time? - At what horizon does the forecast become useless (CI too wide)? 5. RECOMMENDATION - Use forecast for [purpose] with [confidence level] - For planning: use the 80% interval, not the point forecast - Update frequency needed (daily/weekly/monthly) INPUTS: Forecast (point and method if known): [E.G., "Next month revenue: $1.2M — exponential smoothing"] Historical data (for accuracy check): [PASTE PAST FORECASTS AND ACTUALS, OR DESCRIBE TYPICAL ERROR] Metric characteristics: [E.G., "Stable, low volatility" or "Highly volatile, event-driven"] Planning horizon needed: [E.G., "Need to plan 3 months out for inventory"] RULES: - Always provide ranges, not just point forecasts - Distinguish between prediction interval (future observation) and confidence interval (parameter) - Flag when historical forecast accuracy is unknown (be conservative) - Note that longer horizons = wider intervals (always) - Recommend scenario planning when 80% interval spans >50% of mean
How To Use It
- Always ask for prediction intervals, not just point forecasts.
- Use the 80% interval for operational planning (inventory, staffing).
- Use the 95% interval for risk assessment (worst-case scenarios).
- Track forecast accuracy over time — if actuals consistently fall outside 80% intervals, your model is too confident.
- For high-stakes decisions, run multiple forecasting methods and compare.
Example Input
Forecast:
“Next quarter sales: $5.0M — linear regression based on last 2 years”
Historical data:
“Past 4 quarters: forecast errors were -8%, +5%, -12%, +3% (average absolute error ~7%)”
Metric characteristics:
“Moderate seasonality, some economic sensitivity”
Planning horizon needed:
“3 months — inventory ordering”
Why It Works
Most forecasts are presented with false precision — a single number that implies certainty where none exists.
This framework improves outcomes by forcing:
- confidence intervals (ranges, not points)
- uncertainty drivers (what could make the forecast wrong)
- horizon assessment (how far out is still useful)
- historical accuracy (how trustworthy is this method?)
- actionable recommendation (how to use the forecast for planning)
Great forecast assessment doesn’t eliminate uncertainty — it quantifies it so you can plan for it.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.