How do you count AI SDR meetings honestly: booked, completed or held?

A meeting counts as held only when you have evidence that someone joined: a stored recording or transcript, or a host who marked it held. Keep booked, held and no show as separate numbers. A calendar marks a meeting completed when its time slot ends, whether or not anyone joined, so completed is not the same as held.

What is the difference between booked, completed and held?

• Booked: a person picked a time through a booking link, or agreed one in a message. It is a promise.
• Completed: a status many calendars set automatically when the time slot ends. It says the clock ran out. It does not say anyone attended.
• Held: at least one prospect and one host were in the meeting. You can show evidence: a recording, a transcript, or a host who marked it held.
• No show, cancelled, rescheduled: each is a real outcome. They belong in the report, not in the bin.

If you only report booked, the number is high and means little. If you only report completed, you may be counting empty rooms.

Why does a calendar say completed when nobody came?

Because the calendar knows the schedule, not the room. When the slot ends, the status moves on. Nobody has to join for that to happen. It is a reasonable default for a calendar and a poor definition for a sales report.

That is why we do not count the calendar's completed mark as held in our own reporting. The two numbers answer different questions.

How do we count our own meetings?

We publish three numbers about our own outbound, and each one has its own definition. They are not meant to be added, compared one for one, or read as a funnel.

The 1,500 is our headline because it counts every call booked. It is a booked count, and this post does not claim how many of those calls were held. The 351 is the part of it that we can reproduce from our data, so it is a floor on bookings, not a second opinion on the total. The 33 is a floor on held calls under a strict rule: no recording and no host mark, no count.

We also say which base a rate uses. The 351 bookings were 3.2% of the 10,978 people who replied and about 1.3% of the 27,737 people messaged (our own arithmetic from the published counts). Both numbers are true, and they answer different questions.

What do the counts look like in other studies?

These studies were rebuilt by reading the conversations. The labels in a CRM or a calendar did not give the answer.

Which other numbers mislead?

• Stage labels. A lead moved to "meeting" or "customer" by a bulk update is not a meeting. In the foundry study, CRM stage labels overstated outcomes, and every one of the 293 replied threads was read by hand (Lamothermic study).
• Meetings in a calendar you cannot see. If the booking goes to the client's own calendar, your system does not learn about it. Ask for a weekly list.
• Reply rate when the offer changes. If you change the pitch, watch meetings per reply, not replies. In one study a new option in the pitch kept replies steady and produced no meetings in 181 conversations (Uberwood).
• Different bases in one sentence. "3.2%" and "1.3%" for the same meetings differ only by the base. Say which base you used.
• A snapshot taken too early. Bookings arrive weeks after the first message. In the ShortVibe study the median from first touch to booking was 22 days. Recent cohorts look weak for that reason.

What does a weekly meeting report look like?

Five columns are enough. Fill them from evidence, not from a status field.

Unknown is the most honest column. In the wealth network study, 115 of 117 bookings had no recorded outcome, so almost the whole table would have been unknown. A report with a large unknown column is telling you where the process is weak.

Who owns the meeting after the agent books it?

A person does. The agent can send the link, answer questions, and nudge after the link: on our own account, 56% of bookings confirmed in the chat needed at least one more nudge. After that, the human side has to be as reliable as the agent. That means someone is told about every booking, someone is assigned to host it, and someone records what happened.

In the hotel study the same point shows up for replies: a hand off to staff needs an alert and a clock (Hotel Pirin case study, a related party). The lesson holds for meetings. A booked slot with no host is a lost conversation with a delay in it.

A checklist to count meetings honestly

• [ ] Define booked, held, no show and cancelled before you launch, and write the definitions down.
• [ ] Do not count the calendar's "completed" as held.
• [ ] Count one booking per person.
• [ ] Keep a source of evidence for held: a recording, a transcript or a host mark.
• [ ] Report per person messaged and per reply, and say which one a number is.
• [ ] Read the threads behind every meeting you report.
• [ ] Date every snapshot and flag cohorts too recent to have matured.
• [ ] Make sure a person is ready to host. The meeting is where the human side has to be as reliable as the agent.