Revenue, orders or margin falling? How to find the cause in your data

A drop in total revenue is a signal, not an explanation. A useful analysis shows which segment or period the change appeared in and which other data can confirm it.

First, make sure you are comparing the right periods

Compare periods of equal length and check that the data is complete. A week with a different number of working days or a delayed import can create a false decline. If your business is seasonal, also compare against the same period last year.

Break the result into smaller parts

Look at the trend by product, location, region, channel or team. Then split revenue into its underlying drivers, such as the number of orders and the average order value.

• Verify the quality and definitions of the source data.
• Find the segments that contributed most to the change.
• Separate an observed correlation from a cause that still needs to be confirmed.

The analysis, step by step

Start by picking the metric that changed and confirming its definition. For revenue, clarify whether you mean ordered, invoiced or paid amounts. Then choose comparable periods and check that your sources cover the same number of days and are complete. Only after that check does it make sense to break the result down by product, region or channel.

If average revenue changed, split it into the number of orders and the average order value. For margin, also examine costs, discounts and shifts in product mix. For every breakdown, record how much of the total change each group explains and whether the segments add up to the total. A gap can point to missing categories or inconsistent filters.

From correlation to a confirmed cause

Say the decline shows up mainly in one sales channel. That is a place to dig deeper, not proof that the channel caused it. Check for changes in campaigns, pricing, availability, the sales process and measurement. Talk to the people who know the process, and look for data that supports or disproves your hypothesis.

Finish by summarizing what is confirmed, what is only a possible explanation, and which next step makes sense. This distinction keeps a correlation from turning into an unsupported conclusion in a presentation. If the data is not enough to answer the question, name the missing source and add the measurement before the next review.

Write a short note to support the decision

A useful write-up has four parts: what changed, where the change is largest, what has been verified so far, and what comes next. Add the period, the metric definition and the data source. That way the reader can tell fact from hypothesis and can reproduce the result.

If you decide to act, define when and how you will evaluate the effect. Record the expected change and possible side effects, such as the impact of a discount policy change on margin. Compare the same segments and periods, and check that the measurement itself has not changed. If the result is ambiguous, do not declare success or failure too early. Sometimes the right next step is better measurement or a longer observation window, not an immediate change to your whole commercial strategy.

Look at each segment's contribution, not just the ranking

The largest percentage decline does not necessarily have the largest impact on the total. When interpreting results, show the absolute change and the size of the segment as well. A small category can swing sharply, while a modest decline in a large category may explain most of the overall gap. Both views are useful, but they answer different questions: where the relative deviation is, and what moves the company's result the most.

Check the composition and completeness of segments

Before you share a breakdown, check that every record belongs to exactly one category, or that it can legitimately belong to several. Show unknown and empty categories separately; hiding them means the segments will not add up to the total. When filtering, also keep the same time period and the same rules for canceled items. These checks help you tell a real segment contribution from a classification error.