Measure what matters

Since 2021 an open no longer proves that anyone read your email. What people do next does: a click, a reply, an order, a booking. Judge every email by that, count what it cost in unsubscribes and complaints, and compare it with your own usual, not with an industry table.

The short version

• Opens are the weakest number. Apple's mail apps record an open for every email they load, read or not, and Apple accounts for about three in four recorded opens.
• Judge by what people do next: the goal itself (orders, bookings, replies, donations), then money per email, then clicks. Opens come last.
• Know what each rate is divided by. The same clicks give a different rate out of delivered, out of opens or out of sent.
• Benchmarks disagree on opens, from 21.5% to 43.46%, and agree on clicks, at about 2% to 2.6%.
• Judge sequences and campaigns apart. A welcome email at 5% clicks is weak; a newsletter at 5% is excellent.
• Unsubscribes are a healthy exit; complaints do the damage. Keep complaints under 1 in 1,000.
• Your own usual beats any benchmark: the median of your last 10 campaigns, and a weekly graph on a scale you set once and hold.
• Five numbers to watch: click rate, the goal, unsubscribes, complaints and net list growth.

Why an open no longer proves a read

Apple's Mail Privacy Protection, announced in June 2021, has Mail download each email's pictures in the background, "regardless of whether you engage with the email" (Apple, 2025). Every Apple Mail reader with it switched on records an open for every email delivered.

• Litmus, September 2026, from over 1.5 billion opens: Apple's apps recorded 76.56% of opens, Gmail 16.10%, Outlook 3.48% and Yahoo 2.35%. Litmus says the privacy change touches "roughly 55-60% of all email opens".
• Open rates keep rising while clicks don't. Omnisend's rose from 26.6% in 2024 to 30.7% in 2025, and MailerLite's median from 42.35% to 43.46% over the same years. Brevo reports 20.73% with machine opens taken out and 33.87% with them in (2026).
• So on a 40% open rate, 22 to 24 points can't be checked. Some are real reads, some are machines, and nothing tells them apart.
• Open times say little about when people read, because machine opens arrive soon after delivery. Judge a send-time test by clicks.

Opens still show a trend within one list over time, and they work for readers who aren't on Apple Mail. They don't prove reading, they aren't a fair test of subject lines, and they don't define who's engaged.

Judge by what people do next

Zhang, Kumar and Cosguner (Journal of Marketing Research, 2017) found that "email-active customers are not necessarily active in purchases", and that judging email by opens "may be short-sighted". Klaviyo's 2026 figures show why:

Judged by opens, the two look the same. Judged by what people did, sequences get 3.3 times the clicks and 13 times the orders.

So measure in this order:

1. 1

   The goal itself: orders, bookings, donations or replies, per delivered email.

2. 2

   Money per email, where sales can be tied to email.

3. 3

   Click rate.

4. 4

   Unsubscribes and complaints, as the cost.

5. 5

   Opens, last, and never as the headline.

For a business whose sale starts with a conversation, a consultant, a trade, a clinic, a reply is the result. Count replies.

What each number is out of

• Click to open shows what inflated opens did. Campaign Monitor put it at 10.5% in 2021, GetResponse at 8.62% in 2023, MailerLite at a median of 6.81% in 2025, and HubSpot at 5.3%. Emails didn't get worse; the number of recorded opens they're divided by got bigger.
• Clicks aren't perfectly clean either. Some company mail systems click every link to check it before the person sees the email, which inflates clicks on business lists.
• Prefer a median. A few large, very engaged senders lift an average, so a median is the fairer comparison for a small business. MailerLite reports medians; most others report averages or don't say.

Why benchmarks disagree

Open rates differ by a factor of two. Clicks agree: the six general-purpose sources all fall between 2.03% and 2.62%. Clicks are the number the sources agree on.

They differ because each measures its own customers (shops at Klaviyo and Omnisend; small businesses, creators and nonprofits at Mailchimp, MailerLite and Constant Contact), counts different list sizes, uses averages or medians, takes machine opens out or leaves them in, and measures a different year. And figures travel: HubSpot's 42.35% is exactly MailerLite's 2024 median, passed on. A figure that has lost its source has lost what made it useful.

By industry

MailerLite's 2025 medians, all from one source, so the rows compare fairly:

Across four sources (MailerLite, Mailchimp, Campaign Monitor and HubSpot), the pattern holds even where the numbers don't:

• Nonprofits beat their source's average on opens and clicks, in all four.
• Shops click below their source's average, in all four. They send more often and each email asks for less, so they earn less per email and more in total.
• Government, education and the professions click most: government 4.58% at Mailchimp, education 4.4% at Campaign Monitor, legal 4.90% and manufacturing 4.22% at MailerLite.
• Restaurants open and leave: high opens (44.32%), low clicks (1.06%), and the highest unsubscribe rate of any industry at MailerLite (0.39%).
• The source matters more than the industry. Nonprofit opens run from 26.6% at Campaign Monitor to 52.38% at MailerLite, a wider gap than between any two industries in MailerLite's own table.

Sequences and campaigns, apart

GetResponse, 2023 data:

• Klaviyo, 2026: sequences were 5.3% of the emails sent and about 41% of email revenue, earning about 18 times as much per recipient as campaigns. The top tenth of sequences reached 10.48% clicks and 4.3% placed orders.
• Abandoned-cart sequences earned $6.77 per recipient with 6% clicks, and the top tenth $13.70 and 11.3% (Klaviyo, 2026).

So a welcome email with a 5% click rate is weak, and a weekly newsletter with 5% is excellent. Judge each email against its own kind, and each email of a sequence against the same place in your past sequences, because the first email of a sequence always does best.

The cost: unsubscribes, complaints and bounces

• Unsubscribes per campaign: 0.1% (Campaign Monitor, 2021), 0.15% (GetResponse, 2023), 0.22% (Mailchimp, 2023, and MailerLite, 2025). At MailerLite they range from 0.09% for legal to 0.39% for restaurants.
• MailerLite's median unsubscribe rate rose from 0.08% in 2024 to 0.22% in 2025. Since 2024 Gmail and Yahoo have required a one-tap unsubscribe, so people who once ignored an email, or reported it as spam, now leave cleanly. That rise is good news.
• Complaints: 0.06% on average in Validity's 2025 data, and under 0.01% at GetResponse. Stay under 0.1%, one in 1,000; Gmail and Yahoo's hard limit is 0.3%: Get into the inbox.
• Bounces: 2.33% at GetResponse. Keep hard bounces under 2%.

An unsubscribe is a healthy exit; a complaint does the damage. Unsubscribes rising while complaints stay flat means people choosing. Complaints rising means people who don't recognise you, or never asked.

Growth has to outrun loss. M+R's nonprofit lists lost 12% of their addresses to unsubscribes and 4% to bounces in 2025, and still grew 5%. To do that, new sign-ups had to equal about a fifth of the list they started the year with.

Money per email and per subscriber

• Per email: each automated email earned $2.87 per send, and each scheduled campaign email $0.18 (Omnisend, 2025 data).
• Per subscriber: nonprofit fundraising email raised $54 per 1,000 emails, and $2.40 per subscriber over the year (M+R, 2025).

The sums:

• Revenue per recipient = an email's revenue ÷ the emails delivered.
• Revenue per subscriber a year = 12 months' email revenue ÷ the average list size.
• How long people stay, in years, is roughly 1 ÷ the share of the list lost each year: lose 16% a year and people stay about 6 years.
• What one more subscriber is worth = revenue per subscriber a year × how long they stay × your profit margin. That's the most to pay to win one sign-up.

Worked through: Fern & Fig has 5,000 subscribers and sends four campaigns a month. At Omnisend's average of $0.18 an email, that's 5,000 × 48 × $0.18 = $43,200 a year, or $8.64 per subscriber, before any sequence earns a cent. People stay about 6 years and the margin is 40%, so one more subscriber is worth about $8.64 × 6 × 40%, about $21.

Count an order as email's with care. Most tools count an order as an email's if it follows an open or a click within a set number of days. A window that counts opens also counts machine opens, so it gives email orders that would have happened anyway. Count from clicks, and for a firm answer keep a holdout: a slice of the list that gets no email, compared with the rest: Test your emails.

Return-on-investment headlines

• The DMA's 2026 UK tracker puts the return at about £41 for every £1, up from about £38 (read in Action Rocket's summary of it).
• Litmus, 2025, nearly 500 marketers: 35% get $10 to $36 per $1, 30% get $36 to $50, and 5% more than $50. 21% don't measure it, down from 36% in 2023.

These are surveys of marketers. The ones who measure are mostly those whose email works, a fifth don't measure at all, and the answers are beliefs, not audited accounts. What they do show: email costs little to send, so it's usually among the cheapest ways to sell to people who already know you. Never treat a figure as a forecast. Work out your own revenue per subscriber once you have the data.

Your own usual beats any benchmark

A benchmark answers "am I roughly normal?". Your own usual answers "did this email beat my usual?", which is what changes what you do next.

1. 1

   Keep campaigns and sequences apart, and each email of a sequence apart from the others.

2. 2

   Take the median of your last 10 campaigns: click rate, the goal (orders, bookings or replies), revenue per recipient, unsubscribes and complaints.

3. 3

   Judge each new send against that median, never against an industry table.

4. 4

   Graph each number weekly.

Graph it on a scale you set once:

• 13 weeks on one graph: the week you start and the 12 after it, each shown by the date it ends.
• The top is the highest the number can believably reach in those 12 weeks, not this week's high and not a hope. The bottom is the lowest it's expected to go: zero when zero is possible, or when the number is small.
• Even lines between them, at a round amount: 1, 2, 2.5 or 5, or those times 10, 100 and so on, with ten lines at most. Write the amount on the graph.
• Hold the scale. Don't rescale it every week, or after a set time. Re-figure it only when a week goes above the top or below the bottom, and note which week moved it.
• A target line never stretches the scale. It's cut off at the edge.

A scale drawn to today's high has to change on the first good week, and a scale far above anything believable flattens the line. Either way the graph misleads: every week looks like a boom or a crash, or every week looks flat.

Worked through: Harbour Yoga's emails bring 4 to 7 class bookings a week. The most they can believably bring in the next 12 weeks is about 20, so the graph runs from 0 to 20, at 2 a line. A good week of 9 then shows as a clear rise, not a new scale.

Small lists move by chance. On a list of 500, one click moves the click rate by 0.2 points, so a difference of a few clicks is noise. How many people a test needs: Test your emails.

The day of the week matters less than it seems. Campaign Monitor's best day for opens, Monday at 22.0%, beat its worst, Sunday at 20.3%, by under two points, and MailerLite puts its best and worst days about 3.2 points apart. A bad week is rarely the day's fault.

Five numbers to watch

1. 1

   Click rate of each campaign, against your own median.

2. 2

   The goal: orders, bookings, replies or donations per campaign, and revenue per recipient where sales can be tied to email.

3. 3

   Unsubscribe rate.

4. 4

   Complaint rate: the one with a hard limit.

5. 5

   Net list growth, each month.

Opens, if you look at them at all, sit below clicks. Bounces belong with the health of the list: Get into the inbox.

Four questions to ask of any benchmark

1. 1

   Whose senders? Shops, nonprofits, small businesses, or a mix.

2. 2

   When? Before or after September 2021, when Apple's change began inflating opens.

3. 3

   Average or median, and are machine opens taken out?

4. 4

   Divided by what? Delivered, sent, opens or clicks.

If any answer is missing, treat the figure as weak.

Doing it here

• Your Results page counts people, not machines: engaged opens leave out the scanners that open mail the moment it arrives, and each person's clicks count once: What your results count. How machines are told apart: Real people, not machines.
• Its rates: engaged opens are out of what was delivered; clicks, bounces and unsubscribes out of the campaign's whole audience at launch; click to open out of the people who opened. Keep that in mind when you set a click rate here beside another tool's clicks ÷ delivered.
• Opens here are lower than many tools show, because machines are left out: Why are my open rates lower than I expected? Even so, lead with clicks.
• Choose 30 days, 90 days or 12 months, and each number is compared with the same stretch before it. All time shows everything.
• Each campaign's report has a Variations tab. It shows every open the way other tools count them, beside the opens by people, and once its versions have had at least 200 sends each, it names the one that leads on clicks by people.
• A form filled in on a landing page from a campaign becomes a lead linked to that campaign, and replies come to one inbox, filed by campaign, so you can count the enquiries and replies each campaign brought: From a click to a lead and One inbox for every reply.
• Your deliverability score rates out of 100 how much of your mail is getting through: Your deliverability score.
• What stays yours: the orders, bookings and money each email brings, counted wherever they're recorded, and the weekly graph of your own usual.