Two hundred and forty people visited your page last month. The dashboard draws you a line chart at daily resolution, which means it is drawing a chart of the numbers eight, three, eleven, nought, six, two, nine. It colours the line. It puts an arrow next to a percentage.
That chart is not information. It is weather.
This is the awkward middle of a small project. You have enough traffic to feel like you should be learning something from it, and not enough for most of what a dashboard tells you to be true. The answer is not to ignore numbers. Some questions survive small samples perfectly well. It is to work out which three questions those are, and to stop looking at the rest of the screen.
Why small numbers lie, in one paragraph of arithmetic
Say 300 people visit and two links are roughly similar in appeal. One gets 9 taps, the other gets 14. That reads like a 55% advantage for the second link. Run the same page past another 300 people and the numbers could easily land at 13 and 11, reversing it. Counts in single and low double digits bounce around by several either way for no reason at all, and the smaller the count, the louder the bouncing. A gap of 9 against 14 is not a finding. A gap of 4 against 61 is a finding.
That is the whole test, and it is cruder than any statistics course would like. If one number is several times another number, believe it. If you need a percentage to see the difference, you do not have a difference yet.
The useful consequence is that most of your dashboard becomes irrelevant and you only have to look at three things.
Question one: which link gets tapped
Small numbers handle this one well, as long as you only read the extremes.
You are not trying to rank seven links from best to worst. You are trying to find out which one or two carry almost everything, and which ones nobody has touched in a month. Both of those survive the noise. If your shop link has 70 taps and your Spotify has 2, no amount of randomness is going to flip that. If your bottom four links have 3, 2, 2 and 1, they are not in an order. They are all the same link as far as your visitors are concerned, which is to say, furniture.
What to do about it is less obvious than it sounds. The tempting move is to delete everything cold, and sometimes that is right. But a link can be cold because nobody wants it, or cold because it is the sixth thing down and nobody scrolled, or cold because the label is a brand name that means nothing to a stranger. A link called Luminara behaves differently from a link called Shop the hair oil, 100ml, and you will not learn which problem you have by staring at the count. You learn it by changing the label and watching whether the number moves by a factor rather than a percent.
There is also the position effect, which at this scale will swamp almost anything else you test. The top link gets tapped because it is the top link. If you want to know whether a link is wanted on its own merits, move it to the top for a while and see whether it behaves like your old top link did, or worse. That is a clumsy experiment. It is still better than reading the order of four numbers between one and three.
Worth saying plainly: if your page has fourteen links on it, this question has no clean answer at any traffic level, because attention is divided fourteen ways before anything else happens. What your link page should say before anyone clicks is a more useful thing to fix first than any measurement.
Question two: where visitors came from
Traffic sources are categorical rather than continuous, which is why they hold up better at low volumes. The difference between 9 and 14 taps is noise. The difference between "280 of these people came from TikTok" and "four came from Instagram" is a fact about your life, even with only 300 visitors in total.
It answers things you cannot see from inside the app. You posted on three platforms. One of them is doing the work. If you are also running small ads, you can check whether the spend produced anything resembling arrivals before you look at anything further downstream. That matters because a lot of early ad money dies at a step nobody checks, and the fix is usually boring rather than creative. If your ad never ran at all, there is a separate and slightly maddening explanation for that in why your ad was rejected.
Now the caveats, because referrer data is dirtier than it looks. Social apps open links in their own in-app browsers, and those browsers vary in how much they report. Shorteners and redirects can strip the original source. A lot of traffic lands in a bucket labelled direct or unknown, which does not mean people typed your URL from memory. It means the chain of custody broke somewhere. If your direct bucket is forty percent, that is not a finding about loyal fans. It is a finding about plumbing.
The way round it is to stop relying on what the browser volunteers and tag the links yourself. Put a different parameter on the link in each bio and each post that allows one, something as plain as ?from=tiktok or ?from=newsletter. Then the source travels in the address rather than being inferred. Five minutes of work, and it converts a fuzzy pie chart into something you can act on. A printed QR code can carry its own tag too, which is the only way you will ever know whether the card on the counter is doing anything.
Countries and devices belong in this question as well, mostly as a sanity check rather than a strategy. If you ship a physical product within Poland only and two thirds of your visitors are elsewhere, that explains a quiet month better than any tweak to your button copy.
Question three: did the page help anybody act
This is the question that matters and the one your link page is worst placed to answer, because the act happens somewhere else. The page can tell you that someone opened the door. It cannot tell you what happened in the next room.
So the measurement is a ratio plus a handoff. The ratio is visits to taps. If 300 people arrive and 40 tap something, roughly one in seven found a reason to continue. If 300 arrive and 9 tap anything at all, the page is a dead end, and reordering links will not rescue a page that nobody engages with in the first place. That second number is big enough to trust even at this size, because it is the difference between a trickle and nothing.
The handoff is the part people skip. Your shop has its own analytics. Your booking tool sends you confirmation emails. Your email list has a signup count you can read with your eyes. Count the ends, not the middles. One sale, two bookings and five signups in a month is a real result, and it is a result you can hold next to "280 visitors, 70 taps on the shop link" to get a rough sense of where things leak. Seventy taps and one sale means the shop page or the price is doing the losing, not your bio. Nine taps and one sale means your bio is doing the losing and the shop is fine.
Keep that in a note or a spreadsheet, monthly, by hand. It takes two minutes and it is the only record that spans the whole journey, because no single tool sees all of it.
Where Relay fits, and where it does not
Our Insights panel gives you page views, card opens, visitor countries, devices and traffic sources, with a statistics export if you want the raw rows. That maps onto two and a half of the three questions above and then stops. It is part of Pro, which is $49.99 a year or $8 a month plus applicable VAT, and also brings all 59 themes, more fonts, custom colours, a background image, per-link images, scheduled links, inline players, email capture and removal of the Relay badge.
The honest limit is the third question. We do not process product payments, so we cannot tell you that a tap became a sale. Our numbers end at the edge of our page, which means you will always be doing that final join yourself, in your head or in a spreadsheet. Our counts will not match Shopify's exactly either, because two systems counting different events on different devices never agree to the unit. If they ever do line up perfectly, something is broken.
When to say "this data is too thin"
Out loud, to yourself, more often than feels comfortable.
Thin data looks like this: a change you made four days ago that seems to be working, a country breakdown built on nine people, a conversion rate that moved from 4% to 6% on a base of 150 visits, any week-on-week arrow at all. None of these are lies exactly. They are numbers that would look completely different if you could run the same month twice.
The rule I use is to decide the sample before looking at the result. Say in advance that you will judge the new headline after 500 visitors, or after a fortnight, whichever comes later, and then do not check daily, because checking daily is how you talk yourself into a conclusion the data does not contain. And when the 500 visitors arrive and the difference is 31 taps against 36, accept the actual finding, which is that the headline was not the thing holding you back.
Small traffic is not a measurement problem waiting for better instruments. It is a quantity problem. The fastest route to more informative analytics is usually more visitors, which is an unglamorous answer, and it is why most of the time you are better off posting again, or setting up the email signup properly, than refreshing a chart of the number six.