46 views, 0 sales, and probably 0 readers: what UTM links told me that my dashboard couldn't

Five days ago I started reading my analytics dashboard every morning, which is what you do when you have published something and nobody has bought it.

The number went up. Day one, 5 views. Day two, 10. Day three, 19. Day four, 11. Today, 1. Forty-six views in total, zero sales, zero per cent conversion.

For about a day I treated that as a conversion problem. Forty-six people had arrived and none of them had bought, so the obvious suspects were the price, the page, the screenshots. I had a list of things to change.

Then I broke the number down, and it stopped being a number about people.

The breakdown that didn't fit

I sell a kit of delivery templates — spreadsheets and documents for people running migration projects — and the storefront has fourteen products on it: the kit, a larger edition, a free sample, and eleven of the templates sold individually.

Here is how 46 views were spread across those fourteen pages:

Look at the shape of that. Twelve of fourteen pages got hits, in a band between one and six, and the flagship product — the one every link I have ever published points at, the one with the best cover and the only one with a discount code — got one view in five days. An obscure lessons-learned template got five.

That is not what an audience looks like. An audience is lumpy. People arrive at one page, maybe click through to a second, and most of the catalogue never gets touched at all. What I was looking at was flat, and flat across a catalogue is the signature of something walking a list.

The referrer data agreed without proving anything: 44 of the 46 arrived with no referrer at all, filed under "Direct, email, IM", and 40 of them from one country.

Why I couldn't just conclude it

Here is the uncomfortable part, and the reason this post exists.

I had two competing stories. One: a handful of real people found the store through an article I published, browsed around, and didn't buy. Two: something automated walked the storefront.

Both of them are consistent with 46 flat views and no referrer. I could argue for either, and — this is the bit worth noticing — I could not design a future observation that would tell them apart. Whatever tomorrow's number was, I would read it through whichever story I had already picked. That is not a conclusion, it is a preference with a dashboard behind it.

A conclusion you can't falsify with the next data point isn't a conclusion. When you get there, the thing you are missing is not another interpretation. It is instrumentation.

Two links, and the question answered itself

So before arguing any further I did the boring thing: I put UTM parameters on the two links in the article I had published, and I let the week run.

The result was unambiguous in a way the 46 never could have been.

Two clicks. One on each link. Out of 46 views, the only links I had ever published in public account for two of them.

Which means the article did reach a small number of real people — the click-through rate against its own readership is reasonable — and it means that the other 44 views arrived at my store without passing through anything I wrote.

I don't know with certainty what those 44 were. I know they are not evidence of demand, and that is the thing I was about to spend a week optimising against.

The part that will bite you specifically

One detail that cost me days, and is worth knowing if you publish links anywhere.

DEV puts rel="noopener noreferrer" on outbound links. So does a lot of the web, for good security reasons. The effect is that traffic arriving from an article reaches your store carrying no referrer at all, and your analytics file it under "Direct". Real readers and automated crawlers land in the same bucket, indistinguishable, forever.

And the referrer cannot be recovered afterwards. There is no log to go back to, no report that fills it in later. A link published without parameters is a measurement that can no longer be taken — not "not taken yet", can no longer be taken. The five days I spent before adding them are simply not analysable, and they never will be.

That is the asymmetry that makes this worth a rule rather than a habit. Adding the parameters costs about ninety seconds per link. Not adding them costs the entire question.

What I'd tell myself five days ago

Four things, in the order they would have helped.

A traffic number with no breakdown is a headline, not data. Total views, by day, by source and by destination — all four, or you have learned nothing. The single total was the number that sent me off to redesign a page.

Flat distribution across a catalogue is a tell. Not proof, but the first thing to check. Humans concentrate; lists don't.

Instrument the link before you publish it, not when the number gets interesting. By the time the number gets interesting it is already too late to attribute it.

And when two stories fit the same data equally well, stop writing the third version of the argument and go add a sensor. The question I had been chewing on for a day was settled by a tool that took ten minutes to set up and a week of patience.

The underlying failure wasn't analytical, by the way. It was that I spent my effort on the part I enjoy — building the thing — and almost none on the part that decides the outcome, which is getting it in front of people who didn't already know me. Forty-six views that turn out to be mostly machines is just a precise measurement of that.

Better to know. A zero you can trust is more useful than a forty-six you can't.

I build these as a delivery manager on cloud and data migrations. The weekly status report template is free, pay what you want including zero: the status report template

The full kit, 12 templates and 50 prompts, is here if it's useful.

If you've launched something with no audience: what did your first ten sales actually come from? I'm still at zero and collecting honest answers.