Revenue trends: telling signal from noise month to month

Monthly revenue for a small business moves around a great deal, and most of that movement means nothing. The number of working days differs between months. A festival falls in one month this year and the next month last year. One large customer placed an order in the last week of the month rather than the first week of the following one. Weather affected a fortnight. Someone was on leave. None of these tell you anything about the health of the business, and together they can easily produce swings large enough to look like a trend.

The consequence is that reacting to a single month's figure is worse than not looking, because it generates action without information — discounting after a soft month that was only short on working days, or hiring after a strong one that was a single large order. The skill is not sensitivity to changes; it is the discipline to distinguish a change from a fluctuation, and that distinction cannot be made from one data point no matter how carefully it is examined.

The standard tool is a moving average: for each month, the average of that month and the two before it. Plot it alongside the actual monthly figures and the picture separates into two lines — a jagged one showing what happened, and a smooth one showing the underlying direction. Because each point of the average includes three months, a single unusual month is diluted to a third of its influence rather than dominating the view.

The trade-off is worth stating plainly: smoothing costs responsiveness. A moving average turns later than the actual data does, so a genuine change in the business shows up in the smooth line a month or two after it began. That lag is the price of not reacting to noise, and for most small-business decisions it is a good trade, because the decisions in question take weeks to implement anyway. A longer window smooths more and lags more; three months suits businesses with monthly rhythms, while a longer window suits lumpier ones. What matters is choosing the window before looking at the data, since choosing it afterwards means selecting the version that tells the story you already believe.

Comparing March with February is close to meaningless for a seasonal business, because the difference between them is mostly a property of the calendar rather than of anything you did. If March is reliably stronger than February every year, March being stronger this year is not news, and February being weaker is not a decline. Comparing March this year with March last year holds the season constant and isolates the part that is actually about your business.

This is the most useful single comparison available to a seasonal business and it has a real cost: you need at least two years of history, and it is sensitive to whatever was unusual in the comparison period. A strong March last year makes an ordinary March this year look like a decline; a disrupted March last year flatters this one. So look at more than one year where you have it, and note anything exceptional in the base period rather than letting it silently drive the conclusion. Where the festival calendar shifts between months from year to year, comparing individual months can mislead in both directions, and grouping the affected months together is more honest than pretending the boundary is meaningful.

Formal statistical tests exist for deciding whether a change is real, and applying them honestly to small-business monthly revenue is mostly not possible. They require assumptions the data does not satisfy — independent observations, a stable underlying process, enough data points — and with a couple of dozen months of a business that is itself changing, a test that produces a confident verdict is producing false comfort. It is more honest to use judgement, and to be explicit about what is informing it.

Three things make a change credible in practice. Persistence: several consecutive periods moving the same way is far more convincing than one large move, because noise does not usually line up in a row. Size relative to normal variation: look at how much your monthly figure has historically bounced around, and treat a movement inside that band as unremarkable — a swing of a size you have seen repeatedly in stable years is not evidence. And a mechanism: a change you can explain, and whose explanation predicts something else you can check, is much stronger than an unexplained movement. Where you have all three, act. Where you have only one, keep watching and say so, rather than converting a weak signal into a decision because a decision felt overdue.

A few habits protect against the ordinary ways people mislead themselves with their own revenue chart. Fix the start date of the chart before looking at it, because a period chosen after the fact can support almost any conclusion, and a chart beginning at an unusually low month makes anything look like growth. Look at the whole series you have rather than a window, and if you present a window, say why it starts there.

Separate volume from price and from mix, since revenue is their product and three different stories produce the same total: more customers, higher prices, or a shift toward expensive items. Revenue alone cannot distinguish them and they call for different responses. Watch the count of customers or orders alongside revenue for that reason. Then check whether concentration is changing, because revenue held up by fewer, larger customers is more fragile than the same figure spread widely, and the top-line number conceals that entirely. And write down your reading at the time, with the date. Recollection reliably reshapes itself to fit what happened next, and a written note from three months ago is the only reliable way to find out whether your judgement about trends is any good.

A trend describes the past and carries no obligation toward the future. An upward line continues until the thing causing it stops, and the line contains no information about what that thing is or how durable it might be — which is why extrapolating several months forward from a short series is a hopeful act rather than an analytical one. Nor does a trend explain itself: revenue rising while margins fall is a worse position than flat revenue with stable margins, and the revenue line looks better in the first case.

The deeper limit is that all of this describes what happened, not why, and the why is where the decisions live. A moving average that turns downward tells you something changed a month or two ago and says nothing about whether it was a competitor, a price, a product, a staff departure or the weather. Narrowing that down means going back to the underlying transactions — which customers, which products, which channels — and often talking to people. The smoothing and comparison techniques here are for deciding whether an investigation is warranted. They are not the investigation, and treating a clean chart as a conclusion is how a business ends up confident about a trend it has not explained.