Three open rates for the same week: 26%, 20%, 17%

Same week, three answers

Last week our outreach delivered 111 emails. The scorecard reports an open rate of 17.3%. It could have reported 26.1% instead, and it would not have been lying.

Here is the whole week, in the numbers the log actually holds. There were 111 delivery events, landing in 104 distinct mailboxes, because some properties received a second message inside the window. The tracking pixel was fetched 29 times, from 21 distinct mailboxes. Three of those 21 fetched the image within two minutes of delivery and never again, which is a mail gateway loading a picture, not a person reading anything. That leaves 18 mailboxes where a human looked.

Four counts, three rates:

• 29 fetches over 111 messages sent: 26.1%
• 21 mailboxes over 104 mailboxes: 20.2%
• 18 mailboxes over 104 mailboxes: 17.3%

The top number is 1.5 times the bottom one. Nobody cheated and nothing is broken. All three are defensible, and which one you pick decides whether you keep this week's subject line or throw it away.

Events and people are two different columns

Everything you track exists twice: as an event, and as the person it happened to. The system records events, because that is what arrives. You care about people, because people book.

Over the trailing 30 days our own numbers read like this. 323 delivered events, reaching 231 distinct properties: 1.40 messages per property. 63 open events, from 33 distinct mailboxes: 1.91 events per mailbox.

The inflation is not the same at both stages, and that is what ruins a funnel rather than just shifting it. An event-based funnel is optimistic by a different multiple at every step, so the errors do not cancel when you divide one stage by the next. Delivered to opened comes out at 19.5% counting events (63 over 323) and 14.3% counting people (33 over 231). Five points, from nothing but the choice of column.

Ten openers with no usable time

There is a third problem underneath those two. Of the 36 mailboxes that recorded an open at all in 30 days, 10 arrive with no event we can place in time: the mailbox exists, the event exists, the moment does not. Every trailing-seven-day rate therefore makes a silent decision about those 10. Count them and the window is not a window; drop them and the rate has a hole in it. You do not get to not decide, only to know that you decided.

Your occupancy has the same three answers

Take a 40-bed property and a week in which you sold 96 bed nights. The occupancy rate is one of these:

• 96 over 280 (40 beds, 7 nights): 34.3%
• 96 over 266 (two beds held for staff): 36.1%
• 96 over 248 (a six-bed dorm closed three nights for a leak): 38.7%

Four points of spread on the same week, on a metric you take pricing decisions from daily. Our engine at Maka, our hostel in Oaxaca, chases a target occupancy of 75% in low season and reads tonight's real occupancy from the property management system every morning. Across the runs we have logged, that reading came back between 11% and 47%. Whether a morning at 41% is far short of target or mid-range for the season depends on which of the three denominators above your system uses, and most owners have never gone to look.

Enquiries, enquirers, and the conversion rate you quote

This is the one that costs real money. Somebody wants 12 beds for a group in February. They write on Monday, again on Wednesday with different dates, and once more on Friday through Instagram. You book them.

Counted as events: three enquiries, one booking, 33%. Counted as people: one enquirer, one booking, 100%. Your inbox holds three threads either way. If you report the first number, your most interested prospects are the ones dragging your conversion rate down, and the conclusion you reach from it, that your replies are not persuasive enough, is the reverse of what actually happened.

Repeat guests bend the other way. A guest on his fourth stay is four bookings and one person: repeat rate on bookings counts him four times and flatters you, repeat rate on guests counts him once. Only the second tells you how many separate people decided to come back, which is what a loyalty or referral decision needs.

The number no denominator rescues

All of that is the top of a funnel, and the honest part of this post is that none of it helped at the bottom. Over the same 30 days our state file publishes both counts for clicks: 20 click events, one of them a machine, and 7 mailboxes. Replies: zero, on either count. Trials started: zero. Paying customers from outreach: zero.

The field in our own scorecard for emails per customer reads null. Not a small number, not a bad number. Null, because the divisor is zero, and no choice of numerator fixes that. Every refinement above moved a rate by a few points; none moved that.

One more caution on that zero. Our reply route was only confirmed as registered with the mail provider this morning, and one link is still unmeasured: whether mail sent to the reply address reaches the parser at all. Until that is established the zero is a measurement, not a verdict from the market. A number whose measurement path you have not walked end to end is not evidence yet, however round it looks.

Four things to write down this week

• Next to every rate you report, write its denominator in words. Not "20% open rate" but "20% of mailboxes that took delivery". If you cannot write the phrase, you do not have the number.
• Decide once per metric whether it counts events or people, then put the answer in the name. Opens and openers. Enquiries and enquirers. Bookings and guests.
• Never compare two rates whose denominators you cannot both name aloud. Occupancy this month against last, with a dorm out of service five nights, is two metrics wearing one label.
• Count machine events separately before you subtract them. That count moves when your sending pattern moves, and it is the early warning that your open rate changed for a reason unrelated to your copy.

What this does not say

One week and 111 messages is a small base: eighteen humans means a single mailbox moves the rate by about a point, so none of the three figures deserves a decimal place in a plan. Nor is the claim that 17.3% is true and 26.1% is a lie. A pixel fetch is a real event, and there are questions where fetches are the right unit: whether the tracker works, whether images are being blocked.

The argument is narrower. Three defensible numbers existed for one week, nine points apart, and the only thing separating them was what you divide by. That is a different problem from choosing the window you measure over, which we wrote about earlier this month: you can pick the right window and still divide by the wrong thing.

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