ANALYTICS & REPORTING

Seasonality adjusted comparison

Use when a year-on-year or month-on-month comparison might be seasonality rather than performance.

seasonality-adjusted-comparison.md

Download .md

You are checking whether a change in {{METRIC}} is real performance or seasonality.

Current period: {{CURRENT_PERIOD}} value {{CURRENT_VALUE}}
Comparison period: {{COMPARISON_PERIOD}} value {{COMPARISON_VALUE}}
Historical series, as long as available: {{HISTORICAL_SERIES}}
Business calendar (promotions, holidays, term dates, industry events): {{BUSINESS_CALENDAR}}

Do this:
1. State the seasonal pattern visible in the history, with the specific months or weeks that are consistently high and low, and how many years of evidence support it.
2. Calculate what the current period would have been expected to do on seasonality alone. Show the working.
3. Give the residual: how much of the change is not explained by seasonality.
4. List calendar effects that make the two periods not directly comparable, including day-count and weekday-count differences.
5. State how confident you are, given the years of history available.

Rules:
- With fewer than two full years of history, say the seasonal estimate is weak and treat the residual as unreliable.
- Show every calculation. Do not present an adjusted figure without the arithmetic.
- Do not describe a residual smaller than normal week-to-week variation as a change. Say it is within noise.
- If a promotion or one-off event sits in either period, quantify its effect separately or say you cannot.

Fill in before running

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

{{METRIC}}
{{CURRENT_PERIOD}}
{{CURRENT_VALUE}}
{{COMPARISON_PERIOD}}
{{COMPARISON_VALUE}}
{{HISTORICAL_SERIES}}
{{BUSINESS_CALENDAR}}

Getting a better result

Day-count differences alone can explain a few percent; it is the first thing to rule out.
Three years of history makes this useful. One year makes it guesswork, and it should tell you that.
Keep the seasonal profile it produces and reuse it rather than recalculating each month.

Questions about this prompt

When do I use this rather than reporting the year on year figure?

When that comparison is about to be read as performance. Year on year only adjusts for seasonality if the two periods were genuinely comparable, and they rarely are: different day counts, a different weekday mix, a promotion sitting in one of them. This separates the seasonal expectation from what is left over.

What do I need in front of me?

Both period values, the longest historical series you can get, and a business calendar covering promotions, holidays, term dates and industry events. Two full years is the minimum for a usable seasonal estimate and three makes it good. With one year the prompt should tell you the residual is unreliable, and it will.

What comes back?

The seasonal pattern with the specific high and low periods and how many years support it, what the current period was expected to do on seasonality alone with the arithmetic shown, the residual, calendar effects making the periods incomparable, and a confidence line. The residual is the number you report, not the raw change.

What is the mistake that costs me here?

Presenting a residual smaller than normal week to week variation as a change. The prompt calls that noise and so should you. Rule out day count and weekday count differences first. They can account for a few percent on their own and they are the cheapest item on the whole list to check.