Product · The numbers

How the numbers are measured

Who counts as a customer, what each figure on the home page counts, what the counts cannot show, and how to get the same measurement on your own portfolio.

Who counts as a customer

The numbers on the home page are measured over thirty days across every organization that received at least twenty guest messages in that window. On 2 September 2026 that was ten operators and 696 properties. Test workspaces, partner sandboxes and our own internal organizations are excluded, and no single customer is weighted more than its own traffic.

A guest message is a message a guest wrote in a guest inbox: the PMS inbox, WhatsApp, email or SMS. Conversation history imported from the PMS when a customer connects is excluded, and so are messages sent by the PMS's own automations, so an auto-sent check-in email does not count as anything the AI or a person did.

What each number counts

Each figure is a case-level count, not a message count. A case is one guest conversation.
80% prepared
Conversations in which the AI wrote the reply, either as a draft the customer read in Proptonomy or as a message it sent itself, divided by all conversations with a guest message. Counting drafts is deliberate: most customers run in observe mode, where the AI drafts and a person sends from the PMS, and their work is real work.
73% closed by the AI
Measured only at customers running auto-reply, since it is the only place the AI can close a case on its own. Conversations where the AI replied and no person wrote to the guest, divided by conversations where the AI replied at all. It says nothing about drafts.
63% outside office hours
Guest messages that arrived on a weekend or outside 08:00 to 17:00 on a weekday, in the customer's local time, divided by all guest messages.
6,420 stops
The times the AI raised something to a person rather than acting: a message it posted into the customer's operations channel about a guest thread, plus an attention flag it raised on a case, task or property. Approval requests on tasks are counted separately and are not in this figure.
First reply times
The median time from a guest message to the next reply in the same conversation, measured separately for AI replies and for replies a person typed in Proptonomy. Replies a person sent from the PMS are not visible to us and are not in the staff figure.

What we do not claim

Some things the counts cannot show, so the page does not say them.
  • "Closed by the AI" means no person wrote to the guest. It does not mean the guest was satisfied; that needs grading, and the page does not pretend otherwise.
  • Money saved. There is no customer cost data in the system, so any figure would be an assumption dressed as a measurement.
  • A share of all replies sent by the AI. That figure only holds where auto-reply is on. At customers in observe mode the person sends from the PMS, which we cannot see, so a cross-customer percentage of that kind would flatter us.

One window for every number

All three tiles use the same thirty days. A page that quotes its best week for one number and its full month for another is choosing its figures, and you should not have to check.

Reproducing the figures

The queries are plain SQL against a read-only replica, kept in the repository beside the data file for the date they were run. If you are evaluating Proptonomy and want the queries or the same measurement on your own portfolio after a month, write to hello@proptonomy.ai and a person will send them.