Fourteen clicks, three mailboxes, and not one person

The only green number on the board was the fake one

On 12 September our own outreach scorecard showed one metric above target. Click rate: 10.9 percent, against a 4 percent goal, marked "ahead". Everything around it was red. Open rate below target, replies at zero, customers at zero. And there sat one number saying the email was working.

It was not working. Fourteen of those clicks came from three mailboxes, and not one of those three mailboxes had opened the message. That is not a fast reader. A person cannot click a link in an email they have not opened, because every mail client that draws a clickable link loads the tracking pixel first. Fourteen clicks against nine opens on the same version of the email is not a good subject line. It is a machine reading the message before any human gets the chance.

What a click actually counts

Your email platform records a click when something requests the tracked URL. That is the whole definition. It cannot tell you whether the request came from a guest on a phone or from a security filter in a data centre, and a great many of them come from the filter: antivirus scanners and link checkers open every URL in a message to decide whether it is safe, and some do it at delivery, before the message reaches the inbox.

We have one in the log from 7 September: six clicks from a single address in 98 seconds. For part of that afternoon the campaign looked like it had a 12 percent click rate.

Opens have the same flaw in a gentler form. An open is recorded when a one-pixel image loads. Privacy proxies and prefetchers request that image whether or not anybody read anything, which is why an open is the weakest signal on the board and the one every dashboard puts first.

The detail that settled it

One message carried both a click on the tracked link and a fetch of the unsubscribe link, same message id, seconds apart. No person clicks the offer and the opt-out in the same breath. That is a gateway walking every URL in the message, in order. We shipped that check on 22 September at 16:42 UTC. On the first run afterwards, the week's click rate went from one click to zero over 36 delivered, which is the honest figure.

What believing it cost us

We send six versions of the same pitch and let a formula decide how much volume each one gets. It reads roughly: replies plus half the clicks, over the number sent. Feed it fourteen clicks nobody made and it does exactly what it was built to do. It gave that version the maximum weight and pushed the other five down as far as 0.15, so for about ten days somewhere near 62 percent of the only sales channel this company has went to the version a link scanner happened to like.

Nothing looked broken while it happened. No error, no alert. The report was green.

You are running the same measurement

If you send a pre-arrival email, a review request after checkout, or a low season note to past guests asking them to book direct next time, you are reading the same counters we were. The platform does not matter. They all count a fetch as a click and a pixel load as an open, and they all put those two numbers at the top of the screen.

That matters because operators decide real things on those numbers: which subject line to keep, whether the rebooking campaign is worth writing every month. A campaign that looks like it has an 18 percent click rate and produces no bookings usually gets blamed on the website, when the clicks were never people.

Four checks, about ten minutes

• Compare clicks to opens on the same campaign. Clicks should be a fraction of opens. If they are close, or higher, machines are in your numbers. Ours were 14 against 9.
• Read the unique column, not the total. Every platform reports unique clicks and unique openers somewhere, and that is the number that means something. Fourteen events across three mailboxes is not fourteen interested operators. One mailbox should not outvote 76 deliveries.
• Look at the timestamps. Export the click log and read the clock. Six fetches in 98 seconds is a scanner. Opens firing seconds after your send, before anybody could have picked up a phone, are prefetchers.
• Strike the dead and the departed. An address that hard bounced does not exist, so nobody there read anything, and any open or click attributed to it was a gateway. Same for anyone who has unsubscribed. In our current audit every rolled-up click in the window sits on a record that has unsubscribed or was a machine. Zero belong to someone still on the list.

What we count instead

We score each prospect with weights that say plainly what we think each signal is worth: an open is 1 and stops counting after three, a click is 6, a visit to the site is 4, scrolling to the pricing section is 12, pressing the call to action is 15, and a reply from a human is 100. The reply is not merely the largest weight, it is larger than any realistic pile of the others, because someone typing a sentence back is the only evidence here that a tracker did not infer.

Where that leaves us this morning, in full: over the last seven days, 36 delivered messages, 8 pixel fetches from 6 mailboxes, zero clicks from anyone still on the list, zero replies. Over 30 days, 222 delivered and 35 opens. That is a much worse looking board than the one with 10.9 percent on it. It is also the one we make decisions from, and the first honest read of how far the copy still has to go.

The cleanup is not finished either. A click we know was a machine still sits at the top of the list that decides who gets a personal email, because the per-contact counters were written to disk once and the filter that cleans the event stream never reaches them. Knowing a number is wrong and having removed it from every place it is read are two different pieces of work.

The same rule, on the pricing side

This is the reason our pricing engine judges itself on occupancy rather than on anything easier to collect. Across the logged runs at our own hostel in Oaxaca, days priced at an 8 percent discount averaged 33.7 percent occupancy over 23 days, and days at no discount averaged 42.0 percent over 5. That is a small, uncontrolled sample and we treat it as a direction, not a law. But it is a count of beds that were slept in, not a count of fetches, and when the engine saw it, it held the discount instead of digging deeper.

Pick the number closest to money that you can measure honestly, and treat everything upstream of it as a hint. An open is a hint. A click is a stronger hint. A reply is evidence. A booking is the answer.

If this sounds like your property, see the three Nightfill tiers and start a free 14-day pilot on your own data: view pricing.