Forecast {{metric, e.g. revenue}} for the next {{period, e.g. quarter}}.
Inputs:
- Historical data: {{paste CSV or describe — at least 2x the forecast period of history}}
- Known seasonality patterns: {{paste — weekly, monthly, quarterly, annual}}
- Known events that aren't repeating: {{paste — one-time campaigns, outages, big customer wins/losses}}
- Known events that ARE upcoming: {{paste — planned launches, marketing, headwinds}}
Output:
## Method
Which forecasting approach you used and why. One of:
- Naive baseline (last period × growth)
- Trend + seasonality decomposition
- Simple time-series (ARIMA, exponential smoothing)
- Regression with leading indicators
For most planning, simpler is better — flag if you used something fancier than the question warranted.
## Forecast
| Period | Point estimate | 50% CI | 80% CI | 95% CI |
|---|---|---|---|---|
The CIs are load-bearing. A point estimate without uncertainty is a guess pretending to be analysis.
## What drives the range
Top 3 sources of variance in the forecast:
- Source 1 (e.g. churn assumption): if it moves from X to Y, forecast moves by Z
- Source 2: ...
- Source 3: ...
## What this forecast assumes
Explicit assumptions, each one falsifiable:
- Growth rate continues at {{X%}}
- No major customer churn events
- Marketing spend stays at current level
- No platform / pricing change
## How to update this forecast monthly
The 3 numbers to refresh and the rule for when to re-forecast vs. stay the course (e.g. "re-forecast if any single assumption breaks by >20%").
## What this forecast is bad at
Honest.
- Tail events (big wins or losses outside historical range)
- Regime changes (new product, new pricing)
- Anything not in the input data
Hard rule: if your point estimate has 5-significant-figure precision, you're hiding uncertainty. Round to what the CI actually supports.forecastingplanningfinance