How we define reply rate, acceptance rate and meetings

Reply rate is replies divided by people contacted, but both parts need a definition. We count a human reply in the thread, refusals included, and exclude automatic replies. Acceptance rate is accepted invitations divided by invitations sent. A meeting is booked, completed or held, three different numbers. State the denominator whenever you quote a rate.

Outreach numbers are slippery. A vendor says 40%, a benchmark says 3%, and both may be right, because they divide different things by different things. This post sets out how we define the three numbers that matter most in outbound, why we define them that way and how to make your own figures comparable to anyone else's.

Why do definitions matter?

Because the same campaign can honestly produce several reply rates. Two published sources show the spread on cold email:

A factor of seven separates the two, and neither is wrong. They measure different senders with different rules. If you hold your own result up against either without asking how it was counted, you may think you are doing far better or far worse than you are.

What is a reply?

Our rule, applied in every case study on this site:

• Counts: a human reply in the thread. A refusal counts. A question counts. A forward to a colleague counts.
• Does not count: automatic replies, out of office messages, "no longer works here" notices, helpdesk acknowledgements, satisfaction surveys and empty messages.

The Hotel Pirin study says it directly: a human reply in the thread, refusals count, auto replies do not. In Sellinger these automatic messages are marked as machine mail. They never count as a human reply, the agent does not answer them and the lead is sent to a person with a named reason. That keeps an out of office note from inflating a rate or starting a conversation with a robot.

Why count refusals? Because a "no" is a reply that tells you the message reached a person. It is also information. But refusals are not what you want, so we also report a second number.

What is a positive reply?

A reply that moves toward a conversation: a question about price or scope, an agreement to talk, a request for information, a forward to the right person. The Green Shield case study separated these: named buyers produced 17 of the 18 positive replies. The Hotel Pirin study went further and hand read 138 first replies, on a slightly narrower base of 5,190 contacts:

Of the 163 replies (3.0% of 5,499 contacts), the hand read found 23 concrete group inquiries. A reply rate alone would not tell you that. Always pair it with a positive count.

What is the denominator?

This is where the largest errors hide. Choose one and say it.

• Contacted means everyone you sent a first message to.
• Delivered excludes bounces. Hotel Pirin's 5,499 contacts exclude bounces.
• Unique people counts a person once, however many messages they received.
• Companies counts each company once, however many people at it were written to.

Each gives a different percentage. Do not mix them in one sentence, for example "163 replies from 5,190 companies". Name the unit.

The referral network case study shows another trap, quick replies. On LinkedIn many people answer "thanks for connecting". The campaign counted 4,392 replies from 9,808 professionals (44.8%), but only 3,119 of the 9,264 the agent wrote to first replied with more than a pleasantry (33.7%). Both numbers are real. The second says more about conversations. See the case study.

What is an acceptance rate?

On LinkedIn, acceptance rate is accepted invitations divided by invitations sent. It depends on who you invite, whether you add a note and how well the profile reads. Our glossary entry on connection acceptance rate covers the details. Quote the window and the base: invitations sent in a given period, not "ever".

Two cautions. Pending invitations are neither accepted nor refused, so a rate computed too early understates the result. And withdrawn invitations change the base. Say whether you included them.

What counts as a meeting?

Three numbers hide under one word.

A calendar can say completed when nobody came. In the Hotel Pirin and Shelly studies, bookings are called bookings, not held calls. The Sellinger case study reports 351 showcase meetings booked through the agents, one per person, as the reproducible subset, and an owner statement of over 1,500 calls booked in 12 months on average 7.5 LinkedIn accounts for the wider number. Both say "booked". Our post on counting meetings honestly goes through how to track the other two.

How do you make your numbers comparable?

Use this template whenever you quote a rate:

"[X] of [Y] [unit] replied ([Z]%), counted as [definition], from [date] to [date]."

For example: "163 of 5,499 contacts replied (3.0%), counted as a human reply in the thread with refusals included and bounces excluded, from 22 July to 11 October 2026." Anyone can check that, and nobody can mistake it for a different kind of number.

Then compare like with like:

1. Same unit: contacts, people or companies.
2. Same reply definition.
3. Same time window, after follow ups have run.
4. Same audience type. A formal regulated buyer and a casual founder do not reply alike.

Can you compute these yourself?

Yes. The free reply rate and acceptance rate calculator takes sent, delivered, replied and positive counts and shows each rate with its base, so the arithmetic stays consistent. The glossary entry on reply rate gives the short definition.

What are the usual mistakes?

• Quoting a vendor figure as a benchmark. A platform average includes every sender, good and bad, and says nothing about your audience.
• Dividing replies by emails sent, not people. A person who got four emails and replied once is one reply from four sends, which cuts the rate.
• Stopping the clock early. A rate measured after the first email only leaves out the follow ups. The Shelly installer programme drew 3.3% from its first email and 13.6% from the whole sequence, so the same campaign looks four times better after the follow ups have run.
• Counting opens. Opens are distorted by privacy features and say little about interest.
• Treating one good week as the rate. Report the whole window, and say when results were measured.
• Hiding the base. A percentage without its numerator and denominator cannot be checked.

Our own results show why this matters. In the Sellinger case study, the reply rate held at about 40% across a year while meetings per person messaged fell below 1%. Reply rate alone would have said all was well. See the case study, and keep the reply rate beside a booked count.

A checklist before you publish a number

• [ ] I named the unit: contacts, people or companies.
• [ ] I said whether bounces and auto replies are excluded.
• [ ] I reported positive replies beside total replies.
• [ ] I waited until follow ups had run.
• [ ] I said booked, completed or held.
• [ ] I wrote the dates.