At small volumes, random variation produces swings that look like trends. The practical response is to lengthen the period, watch counts rather than percentages, and act on patterns that persist across several months rather than on any single month.

The problem with percentages on small counts

A business with four enquiries in January and two in February has a fifty percent decline, and it also has two people.

The percentage is arithmetically correct and practically meaningless, because two enquiries is well within the range you would expect from chance alone with no change to anything.

Reporting tools present percentages by default, which makes ordinary variation look like a crisis and makes ordinary good luck look like a success worth repeating.

What normal variation looks like

Larger than most people expect.

If your true average is five enquiries a month, months of three and months of eight will happen regularly with nothing having changed. That is not a signal, it is what a small average looks like when you sample it monthly.

The practical implication is that a single month tells you almost nothing, and two consecutive months tell you only slightly more.

The intuition worth building is that the smaller the number, the longer you have to watch before a change means anything.

What to do instead

Lengthen the period

Quarterly rather than monthly, or a rolling three-month average. Both smooth the noise without hiding a genuine trend.

A rolling average is the more useful of the two, because it updates every month while still reflecting a longer window.

Watch counts, not percentages

Enquiries went from four to two is honest. Enquiries fell fifty percent is technically true and misleading. Use the count in your own reporting.

Compare against the same period last year

Which removes the seasonal component that explains most small-business variation.

Look at leading indicators with more volume

Enquiries are few; page views and search impressions are many. Those move earlier and are less noisy, which makes them better for spotting a real change.

Require persistence before acting

A useful rule: do not change anything on the strength of one month. Three consecutive months moving the same direction is a signal; one is weather.

Split testing on a small site

Worth addressing plainly, because it is commonly recommended and rarely applicable.

Comparing two versions of a page requires enough conversions in each version to distinguish a real difference from chance. For a site producing a handful of enquiries a month, that takes many months, by which time other things have changed and the comparison is spoiled.

The honest position is that formal split testing is not available to most small businesses, and pretending otherwise produces confident conclusions from noise.

What is available: making a clearly reasoned improvement, leaving it in place, and comparing a long period before against a long period after, while accepting that the comparison is suggestive rather than proof.

The changes worth making anyway

The consequence of all this is that small businesses should act on reasoning more than on data.

A tappable phone number, a published price, a form with fewer fields, and a page that loads quickly are improvements whether or not you can measure them. They are known to remove friction, and waiting for statistical confirmation you will never accumulate is a way of never improving anything.

Measure to catch things going badly wrong, which is what small-sample data is genuinely good for. A drop to zero means something. A drop from five to three does not.

What small numbers are good for

Being fair, since this is not an argument against measuring.

Detecting breakage. Zero enquiries where there were normally some is unambiguous and worth investigating immediately.

Large differences. One over four is noise; forty over four is not. Big effects show through small samples.

Accumulating a baseline, so that in two years you have a history worth comparing against.

Qualitative detail, such as which pages people actually land on and what they searched. That does not need volume to be informative.

The habit worth building

Record the counts monthly, look at them quarterly, and compare annually.

That cadence matches what the numbers can actually support, and it prevents the most expensive mistake in small-business measurement, which is abandoning something that was working because of a quiet month. That is the same error as reading a seasonal dip as a competitor problem, described in when a competitor opens nearby.

Frequently asked questions

Why are percentages misleading on small numbers?

Four enquiries falling to two is a fifty percent decline and also two people. The percentage is correct and practically meaningless, since that swing is well within chance.

How much variation is normal?

More than expected. With a true average of five a month, months of three and eight will happen regularly with nothing having changed.

What should I look at instead?

Counts rather than percentages, a rolling three-month average, the same period last year, and leading indicators like impressions that have more volume and less noise.

When should I act on a change?

When three consecutive months move the same direction. One month is weather, and changing things on the strength of it is how working approaches get abandoned.

Can I run split tests on a small site?

Usually not. Distinguishing a real difference from chance needs more conversions than a handful a month provides, and by the time you have them other things have changed.

What are small numbers actually good for?

Detecting breakage, since zero where there were normally some is unambiguous; spotting large differences; building a baseline; and qualitative detail like which pages people land on.

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.

Reacting to a forty percent drop that was two people?

We set up reporting at a cadence the numbers can actually support, so you stop cancelling things that were working.

Start a Conversation