How to check whether your AI SDR's meetings actually happened
Take the list of meetings your AI sales agent counted as booked, find each one in your own calendar and CRM, and check four things: that the AI booked it rather than one of your reps, that the person was a new prospect, that it was not cancelled or moved indefinitely, and that someone from the prospect's side attended. Then read a sample of the conversations that led to the meetings to confirm each was a sales meeting with a potential buyer.
The checks are those of the Attify Audited Meeting Standard (v0.6), all of them counted, as for a vendor that claims held meetings. A vendor that claims meetings booked is audited only on who booked it, a new prospect, a rep stepping in and the conversation read, with its cancellations and no-shows shown beside the claim (the claimed unit).
1. Get the vendor's list, with ids#
Ask the vendor for every meeting it counts as booked in the period, with the meeting's id in your calendar, CRM or scheduling tool. AI SDR vendors count at booking or earlier (a booking event, a status set by hand, a reply label, any meeting within a window after a touch; see how vendors define a meeting), so the list is the claim, not the result.
Use a period long enough to settle: 90 days of first scheduled times, ending at least 7 days before you check.
2. Check each meeting in your own systems#
HubSpot's meeting outcome has default values Scheduled, Completed, Rescheduled, No show and Canceled (HubSpot). Google Meet's conference records list each participant's join and leave times (Google).
Count each meeting at the first check it fails.
3. Read a sample of the conversations#
For a random sample of the meetings that passed, read the conversation that led to the booking. Fail meetings with job seekers, recruiters, vendors pitching back, students, investors, journalists or competitors, existing customers asking for support, and anyone who declined or was paid to attend.
4. Compute and compare#
- Audited share = meetings passing every check ÷ meetings counted as booked.
- Your own team's baseline: put your reps' meetings with outside people in the same period through the cancellation and no-show checks. Some of any gap is the prospects, not the AI.
- Qualified: of audited meetings, how many led to a new opportunity within 30 days of the meeting. Report it separately; it depends on your sellers too. Human SDR teams set quota on held meetings and then converted opportunities (The Bridge Group, 2025).
Worked example#
Synthetic example. 120 meetings counted as booked over 90 days.
Audited share 63 ÷ 120 = 52.5% of the meetings counted as booked. If the vendor bills $250 per booked meeting, spend is $30,000 and each audited meeting cost $476.19.
Do it with Attify#
Connect your CRM (HubSpot, Salesforce), calendar (Google Calendar, Microsoft 365) and scheduling tool (Calendly, Cal.com) on the buyer dashboard to get the audited share every month, your own team's baseline beside it, and the cost per audited meeting.
Questions
Our pilot booked 47 meetings and 4 became opportunities. Is that a failure?
- First check how many of the 47 were held, with new prospects, without a rep stepping in. If 30 were held, 4 opportunities is 13.3% of held meetings. Compare that with your own reps' meetings with similar prospects in the same period, not with a vendor's benchmark.
How do I report AI-booked meetings separately in HubSpot?
- Use the record source or the creating user on each meeting: an AI agent writes as its own user, app or integration. Where the AI books through a rep's calendar, the vendor's list of booked meetings, with ids, is the only way to tell.
The vendor counts reschedules as new meetings. How do I dedupe?
- Treat a meeting moved to a new time as one meeting, and a cancelled meeting rebooked with the same person within 14 days as the same meeting. Count one per person per 60 days.
What if nobody logs meeting outcomes?
- Video-call participant records (Google Meet, Microsoft Teams) show who joined. Without them, and without a team that logs outcomes, attendance cannot be tested, and any held rate is a guess.