Your Website's Conversion Rate Is a Range, Not a Number
If your site had 300 visitors last month and made 6 sales, your conversion rate is not 2 percent. It sits somewhere between 0.9 percent and 4.3 percent, and the honest position is that you cannot yet tell where.
That range is not caution or hedging. It is what the arithmetic returns. Six is a small number, and small numbers carry a lot of uncertainty with them.
This matters because the whole small business analytics ritual assumes the opposite. You check the dashboard, you see 2 percent, you compare it against a benchmark you read somewhere, and you conclude something about your website. Every step of that treats a rough estimate as a precise fact.
What "2 percent" actually means at 300 visitors
Start with what the platform is doing. Squarespace calculates the purchase funnel figure as checkouts divided by total visits, times 100. That division is exact. Six out of 300 really is 2 percent, and nothing is broken.
The trouble starts the moment you use it to say something about your website rather than about those 300 visits. For that, 6 sales is a sample, and a thin one.
Run the numbers and the picture changes. Here is the same measured 2 percent at different traffic levels, with the range the underlying rate could plausibly sit in.
| Visitors | Sales | Measured rate | Plausible range | Width |
|---|---|---|---|---|
| 300 | 6 | 2.0% | 0.92% to 4.29% | 3.4 points |
| 1,000 | 20 | 2.0% | 1.30% to 3.07% | 1.8 points |
| 3,000 | 60 | 2.0% | 1.56% to 2.57% | 1.0 points |
| 10,000 | 200 | 2.0% | 1.74% to 2.29% | 0.6 points |
At 300 visitors the range spans almost everything a small store could realistically be. The benchmark articles that tell you a good rate is 1 to 3 percent are describing a band narrower than your own measurement error. Comparing yourself to them is not a meaningful exercise yet.
Those ranges use the Wilson method rather than the formula most people were taught. That choice is deliberate: Brown, Cai and DasGupta showed in Statistical Science that the textbook interval behaves erratically at exactly these sample sizes, and they recommend Wilson for small samples.
Worth running your own version. Take your visits and your orders, divide, then work out the range around it. I check the arithmetic at math-solver.io because it lays out each step rather than handing back a figure, which is useful when you are trying to understand the calculation and not just get past it. Do not expect it to run a full test analysis for you. It handles the percentages and the z-score, and that is the part most people get wrong anyway.
The month that went "up 54 percent"
Here is the version of this that costs money.
Two months, with a homepage rebuild in between. The dashboard tells what looks like a clear story, and you changed the homepage in late March, so the rebuild worked.
| Month | Visitors | Sales | Conversion rate |
|---|---|---|---|
| March | 250 | 5 | 2.00% |
| April | 260 | 8 | 3.08% |
What actually happened: Three extra sales. The rate moved 1.08 percentage points. Run a comparison of the two months and the result is a z of 0.77, which gives a probability of roughly 0.44 that you would see a swing at least this large with nothing changed at all.
Put plainly, a difference that size turns up by chance close to half the time. It is not evidence about your homepage. It is what two small months look like next to each other.
The expensive part is not the wrong conclusion. It is what follows: you keep the change, you build the next decision on top of it, and you never find out.
Why you probably cannot A/B test this
The standard advice at this point is to test properly. So let us price that out.
Say you convert at 2 percent and you want to spot an improvement worth having, something like a fifth better than where you are now. The number of visitors that takes is the part nobody quotes.
The sample size Detecting a lift from 2.0 percent to 2.4 percent, at the conventional 95 percent confidence and 80 percent power, needs 21,109 visitors per variant. That is 42,218 visitors in total.
At 1,000 visitors a month, the test finishes in about 42 months.
Most small business sites cannot run a real A/B test on purchases. Not because the owner lacks discipline, but because the traffic is not there, and no amount of tooling fixes that.
Which leads to the trap that follows. When a test drags on, people watch it and stop as soon as it looks significant. Evan Miller worked out the cost of that habit in How Not To Run an A/B Test: stopping the moment you see significance at the 5 percent level pushes your real false positive rate to 26.1 percent. Roughly one in four "winners" is nothing.
What to measure instead
None of this means stop improving your website. It means stop asking a rare event to referee your decisions, and move to events you have enough of.
Purchases are the rarest thing on your site. Almost everything upstream happens more often: button clicks, form starts, pricing page views, add to cart. Squarespace reports form and button conversions as their own panel, and those events can be ten to fifty times more frequent than orders on the same site. More events means a tighter range and a faster read.
There is a floor worth knowing. The NIST Engineering Statistics Handbook gives a working rule that your sample should satisfy min(Np, N(1-p)) of at least 5. At a 1 percent rate that means 500 observations before the standard test is even appropriate. Below that you are not measuring, you are reading tea leaves.
Three habits that survive small traffic:
Report counts, not just percentages. "8 orders, up from 5" is honest. "Conversion up 54 percent" is not, at these volumes.
Pool longer windows. A quarter of data beats a month. Add the orders, add the visits, divide once. Do not average three monthly percentages together, because months with different traffic do not carry equal weight.
Look for changes big enough to see. At small scale you can detect a doubling. You cannot detect a 20 percent lift. Spend your effort on changes that might plausibly double something.
Reading your own numbers honestly
The uncomfortable part is that this takes away a source of reassurance. A dashboard reading 2.4 percent feels like knowledge. A range of 0.9 to 4.3 percent feels like ignorance, even though it describes what you actually know more accurately.
It is still the better place to work from. Owners who know their numbers are fuzzy make decisions for reasons that hold up, because customers said something, because the checkout genuinely is confusing, because a page takes six seconds to load. Owners who trust a precise-looking figure end up chasing noise around and calling it optimisation.
Before you act on a conversion number
How many sales is this actually built on? Under 30, treat it as a hint
Have I written down the count next to the percentage?
Would a swing this size happen anyway in a quiet month?
Am I comparing two periods with similar traffic, or averaging percentages from months that are not comparable?
Is there an upstream event with more volume that answers the same question?
If this is a test, did I fix the sample size before starting, or am I stopping because it currently looks good?
Your conversion rate was never a single number. Reporting it as a range is not a lowering of standards. It is the first honest thing most small business dashboards could say.
