ANALYTICS & REPORTING

Forecast with stated assumptions

Use when someone wants a number for next quarter and you need the assumptions visible.

forecast-with-stated-assumptions.md

Download .md

You are producing a forecast for {{METRIC}} for {{FORECAST_PERIOD}}.

Historical data: {{HISTORICAL_SERIES}}
Known upcoming events: {{PLANNED_EVENTS}}
Known constraints (budget, headcount, capacity): {{CONSTRAINTS}}

Produce:
1. An assumptions table before any number: Assumption | Value used | Source (measured, stated by stakeholder, or guessed) | Impact if wrong (high/medium/low).
2. Three scenarios - low, expected, high - each with the single number, the assumptions that differ from the expected case, and a rough probability with the reasoning for that probability in one clause.
3. A month-by-month or week-by-week table for the expected case.
4. The two assumptions that drive the most variance, and what evidence would tighten them.
5. A plain sentence on how much to trust this, given the length of history available.

Rules:
- Do not present a single point estimate anywhere without its scenario label attached.
- If the history is shorter than two full cycles of the seasonality involved, say the forecast is weak and say why.
- Do not smooth over an anomaly in the history. Name it and say whether you included or excluded it.
- Round to a sensible precision. Do not give four significant figures on a guessed input.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{METRIC}}
{{FORECAST_PERIOD}}
{{HISTORICAL_SERIES}}
{{PLANNED_EVENTS}}
{{CONSTRAINTS}}

Getting a better result

Mark stakeholder-stated assumptions honestly - they are usually the ones that break.
Ask for the forecast to be restated after one month against actuals to see which assumption failed.
If it will not commit to a probability, that is a signal the history is too short.

Questions about this prompt

When do I use this rather than extending the trend line?

When somebody will commit spend or headcount against the number. A trend line buries its assumptions inside the fit. This puts them in a table before any figure appears, each labelled measured, stated by a stakeholder, or guessed, which is what makes the forecast reviewable rather than merely wrong later.

What do I need in front of me?

The historical series, planned events inside the forecast period, and real constraints on budget, headcount or capacity. Be honest about which inputs came from a stakeholder rather than from data. Those are the ones that break, and the source column is worthless if you launder a guess into a measurement.

What comes back?

The assumptions table first, then low, expected and high scenarios each with a probability and the reasoning for it, a period by period table for the expected case, the two assumptions driving most of the variance, and a plain sentence on how far to trust it. Those two assumptions are where your next hour goes.

What is the mistake that costs me here?

Quoting the expected case without its label. Every number here is meant to travel with its scenario, and the one that gets pasted into a plan is always the middle figure stripped of context. If it will not commit to a probability, read that as the history being too short rather than as evasion.