Trend Analysis and Forecasting

Look Beyond Current Numbers

Descriptive statistics tell you what happened. Trend analysis tells you what is happening. Forecasting tells you what could happen. These three levels of analysis answer three different questions in a meeting, and confusing them is costly: no one makes the same decision based on a historical fact as they do on a projection. With ChatGPT, you can perform all three without writing a single line of code.

Identifying a Trend

The simplest form of trend is linear: are your data going up or down, and at what speed? The useful reflex is to ask for the trend line and its reliability indicator in the same breath; otherwise, you will be commenting on a slope whose solidity you cannot verify.

Plot monthly revenue for the last 12 months with a trend line.
Indicate the average monthly growth rate and whether the trend
is statistically significant.

ChatGPT calculates a linear regression, plots the line, and gives you the coefficient of determination (R²), which measures the reliability of the trend. An R² close to 1 means the line summarizes your twelve points well; a low R² warns you that the “trend” you were about to announce is mostly noise, and a thirteenth month could reverse it.

Moreover, many business datasets follow cycles, and a rise in December is not necessarily good news: it may just be the usual December. To decide, ask for a decomposition.

Decompose monthly sales into 3 components:
underlying trend, seasonality, and residuals.
Display the 3 curves separately.

ChatGPT applies a seasonal decomposition method, which it can specify for you, to isolate each component. It needs at least two full cycles—two years of monthly data for annual seasonality—for the decomposition to make sense. You then clearly see whether the December rise stems from a seasonal effect or real growth—and the residual curve flags the months that neither explanation covers.

Moving Averages

Moving averages smooth out variations and highlight the underlying trend. On daily revenue, the weekend dip almost always masks the real movement; overlaying two different windows lets you read both the short and medium term on the same chart.

Plot raw daily revenue and overlay a 7-day moving average
and a 30-day moving average.

You choose the window based on your data granularity and the cycle you want to eliminate.

Simple Forecasts

For a quick estimate, extending the observed trend is often enough—provided you acknowledge uncertainty rather than hide it.

Project revenue for the next 3 months by extending
the linear trend of the last 12 months.
Display the 95% confidence interval.

The confidence interval is the decisive part of the result: it shows the margin of uncertainty, and the more it widens as it moves into the future, the less reliable the forecast. This is also your best protection in a committee meeting. Announcing “between €480k and €620k” commits you less than a single figure that no one can guarantee, and it triggers the right conversation: the one about what would push results from one end of the range to the other.

As soon as your data show recurring patterns, linear projection becomes misleading, and you need a model that integrates the cycle.

Use a forecasting model that accounts for seasonality
to estimate sales for the next 6 months.
Compare the result with a simple linear projection.

ChatGPT can use models that integrate seasonality, such as Holt-Winters exponential smoothing or a seasonal ARIMA model, which capture both trend and cycle. The comparison requested at the end of the prompt is instructive in itself: the gap between the two curves measures exactly the weight of seasonality in your business.

Breakpoint Analysis

Sometimes, a trend does not deform gradually: it breaks. A launch, a price change, or an external event shifts the trajectory from one month to the next, and a trend line calculated over the entire period then describes nothing real.

Identify breakpoints in the evolution of monthly revenue.
Are there months where the trend changes direction significantly?
For each breakpoint, indicate the date and the magnitude of the change.

This analysis is valuable for understanding the impact of specific events: product launches, price changes, external events. Then, match the dates obtained with your internal calendar—pricing overhaul, campaign, departure of a sales rep. This matching, and this alone, turns a curve into an explanation.

Period-over-Period Comparison

Two readings complement each other, and disagreement often arises because two stakeholders are not using the same one. Year-over-year comparison neutralizes seasonality, since each month is confronted with its counterpart from the previous fiscal year.

Compare sales month by month between 2025 and 2026.
Calculate the percentage variation for each month
and flag the months with a decline.

Sequential growth, on the other hand, measures the immediate movement from one month to the next. It reacts much faster to recent inflections, at the cost of high sensitivity to seasonality.

Calculate sequential growth (month vs. previous month)
and display a bar chart with growth months in green
and decline months in red.

More Reasoning for Forecasts

Some forecasts cannot be reduced to a calculation: they require integrating factors that your historical data do not yet contain, such as a product that did not exist last quarter.

Taking into account seasonality, the underlying trend,
and the fact that we are launching a new product in April,
estimate sales for the next quarter.
Explain your reasoning step by step.

A higher level of reasoning—High or Extra High in the ChatGPT selector, depending on your plan—helps in this type of multi-factor reasoning, where you must combine quantitative analysis with contextual judgment. The result remains an estimate based on your assumptions, not a fact. Demanding a step-by-step explanation is not a whim: it allows you to challenge a specific assumption—the ramp-up pace of the new product, for example—rather than rejecting the figure wholesale because you do not know where it came from.

Your Turn

Your first forecast.

1. Plot your monthly revenue with a trend line and its R²
2. If you have at least two years of history, decompose trend, seasonality, and residuals
3. Project the next three months with a 95% confidence interval
4. Compare with a model that integrates seasonality
5. Present a range rather than a single figure, along with its assumptions

Official sources: Extracting Insights with ChatGPT Data Analysis · Data analysis with ChatGPT (OpenAI Help Center)

Key Takeaways

• The trend line and R² measure the direction and reliability of the trend
• Seasonal decomposition separates trend, seasonality, and noise
• Moving averages smooth variations—choose the window based on your granularity
• Every forecast must include a confidence interval—no single figures
• Breakpoints identify moments when everything changed