Nezam AI proof: every number we publish, and where it comes from.
Every figure on this site, with the line it was measured on, the window it covers, and the page that shows the method. The numbers come from two phone lines we answer for a Toronto property management and brokerage operation, and from that operation's own CRM export. None of them is a lift claim, because nobody measured the same window before the agents took the line.
The voice agents, June 2026 sample
Two agents answer every inbound call on the operation's property management line and its brokerage line. The sample is the first stretch we published: the brokerage line from 4 to 20 June 2026, the property management line from 11 to 20 June, deduplicated by call id.
| Figure | What it measures | Window |
|---|---|---|
| 0 of 277 | calls missed in the sample | Both lines, 4 to 20 June 2026 |
| 74 of 277 | arrived after hours or on a weekend | Both lines, same sample |
| 120 of 167 | brokerage calls left a callback number | Brokerage line, 17 days |
| 73 of 110 | property management calls closed inside the call, the rest handed to staff with the context written down | Property management line, 10 days |
| 34 of 130 | new leads in the CRM export came through the voice agent, the second largest source in the business | CRM export, 8 to 22 July 2026 |
Source: The full impact study.
The call study, one quarter of both lines
Once an agent answers everything, the arrival curve stops being a guess. From 1 June to 10 September 2026, 102 days, every call on both lines produced a timestamped record, so the count is a count. The post carries the tables by hour and by day, and the method.
| Figure | What it measures | Window |
|---|---|---|
| 3,332 | calls across both lines: 2,142 on the property management line, 1,190 on the brokerage line | 1 June to 10 September 2026 |
| 19.3% | arrived outside Monday to Friday, nine to six, across both lines | Same quarter |
| 12.7% | of the property management line's calls arrived after hours; the brokerage line ran 31.3% | Same quarter |
| 21 | calls arrived between 10pm and 7am across both lines. Nights barely exist; evenings do | 102 days |
| 158 of 2,142 | property management calls carried the system's own emergency tag, applied at the moment of the call | Property management line |
| 41% | of those tagged emergencies arrived in the after-hours window that held 12.7% of the calls. An after-hours call was tagged an emergency 24.0% of the time, against 5.0% in business hours | Property management line |
| 70.7% | of after-hours brokerage callers left a callback number, against 72.5% during business hours. We expected a gap and there is none | Brokerage line |
| 12.2% | of brokerage calls were a wrong number, a robocall or silence; after hours the share rose to 14% | Brokerage line |
Source: The study, with the method and the tables.
What these numbers are not
- They are share of source, not lift. The export proves a quarter of new leads arrive through the agent; it cannot prove those leads did not exist before, because nobody measured the equivalent window beforehand.
- They are one operation, one market, one quarter. A Toronto property management and brokerage operation, anonymized by agreement. We do not name the client.
- They cover the property management line and the brokerage line, where the voice agents run in production. We publish no production figure for trades, law firms or agencies, because we have none.
- The reason categories on the brokerage line (asking for a specific person at 28.7%, and so on) are per-call judgments and the study says so. The timestamps, the callback fields and the emergency tag are logged by the system, not judged afterwards.
How the numbers are made
Every call the agents take produces a structured record: a timestamp, who called and why, what was decided, and a recording, posted to the operation's team within seconds. The published figures are computed from those records by script, deduplicated by call id, and never counted by hand or by a model.
Personal data stays inside the operation. Caller names, numbers, addresses and staff names never leave its systems; what we publish is the aggregate, and the anonymized dataset and method note behind the study are available to a reporter or a prospect on request.
When a number changes, this page changes with it, and the old figure is not quietly replaced. The study is refreshed annually rather than quarterly, because a stale claim is worse than a smaller one.
What a client owns
The source code, the credentials and the documentation, from day one. The agent's routing rules are written in plain words and kept in the client's repository. A person approves anything that moves money, and anything near a legal step stays with a person. The rules every build follows are written down on the trust page.
The first conversation is free and takes thirty minutes. You get a number on that call, before any proposal, and the audit that decides the build order is a fixed fee agreed before it starts.
About the name
Nezam AI is the trading name of 1001592704 Ontario Inc., operating as Nezam AI Consulting, an AI deployment firm headquartered in Toronto with people in Boston, Cairo and Dubai. Nezam is Arabic for the system, and we named the company for it. Other companies use the Nezam name for unrelated products; none of them is this firm, and nothing on this page describes them.