Zero missed calls and a quarter of all new leads: how a Toronto property operator put AI on the phone.

The challenge

The phone was the business, and the phone was leaking.

A property management company and its brokerage arm ran on inbound calls. Tenants with emergencies, owners with questions, cooperating agents booking showings, renters asking whether a unit was still available. Every one of those calls was either a lead, a maintenance ticket, or a relationship.

The calls did not arrive on a schedule. They arrived at 7:38 on a Friday evening from an agent standing at a door with a lockbox that would not open, and at 9:01 on a Wednesday night from someone submitting an offer. Outside office hours the whole operation answered with a voicemail box, and a voicemail box does not book a showing, dispatch a plumber, or write a lead into the CRM.

Nobody could say how big the leak was, because the calls that went unanswered left no record. That is the shape of this problem in most operations. The cost is invisible precisely because the thing that would have measured it is the thing that never happened.

1 in 4 of the calls the system now answers arrive outside office hours or on a weekend. Before it was live, those hit voicemail.

What we built

Two agents, one for each side of the business, on the line 24 hours a day.

• Answer and understand. Every inbound call picked up on the first ring, at any hour, with the caller's intent understood in conversation rather than pushed through a phone tree.
• Capture and route. Caller, callback number, property, and intent captured, then routed by the operator's own rules: handled in the call, flagged urgent to staff, or written into the CRM as a lead.
• Escalate on judgment. Emergencies, disputes, and anything needing a decision go to a person with the full context attached, rather than being resolved by a machine that should not be deciding.
• Log everything. A structured summary, transcript, and recording of every call posted to the team in seconds, so the record exists whether or not anyone was free to take it.

The agents sit in front of the phone system rather than replacing it, and write into the CRM the team already used. Nothing was migrated, no staff workflow was rebuilt, and the operator can read every decision the system made and reverse it. The client owns the source code, the prompts, and the integrations.

Measured over a June call sample and a separate two-week lead export in July.

The results

• 0 calls missed. Across the whole sample, every inbound call was answered on the first ring. Not every issue was resolved by AI, and it was never meant to be. Nothing rang out.
• 26% of all new leads. Of 130 new leads recorded in the CRM between 8 and 22 July, 34 arrived through the voice agent. That made it the second largest source in the business, ahead of the next two listing portals combined.
• 27% after hours or weekend. 74 of 277 calls in the sample came in outside office hours. This is the volume a voicemail box was absorbing, now captured, summarized, and routed.
• 7 in 10 brokerage calls leave a routable lead. 120 of 167 brokerage calls ended with a callback number and enough context for someone to act on it the next morning.

Where these numbers come from: Call figures come from the live call logs, deduplicated by call ID: 277 calls, 167 on the brokerage line over 17 days and 110 on the property line over 10 days, which is a run rate of roughly 600 a month at current volume rather than an audited annual total. Lead figures come from a CRM export covering 8 to 22 July 2026 and 130 new leads. The export shows where leads came from, not what lead volume was before the agents went live, so this is a share of source and not a lift claim.

Key insights

• The after hours share is the number nobody has. Operators can usually tell you their call volume during business hours. Almost none can tell you what arrives at night, because the system that would have counted it is the voicemail box. Measuring that share first is usually what makes the case, and it is bigger than people expect.
• Answering is not the same as resolving. Two thirds of property calls were closed inside the call. The rest were handed to staff with the context already written down. Both outcomes are wins, and conflating them is how vendors end up overpromising a resolution rate they cannot hold.
• Automate the intake, never the judgment. A quarter of the property calls in the sample came from upset callers: urgent repairs, a towing charge dispute, a maintenance request that had gone unanswered for weeks. Every one was captured in full and put in front of a person. A system that tried to settle those itself would have made the relationship worse, not cheaper.
• Share of source is provable. Lift is not, without a baseline. The export proves a quarter of new leads now arrive through the agent. It cannot prove those leads did not exist before, because nobody measured the equivalent window beforehand. If you are deploying something like this, take the baseline first. It costs one export and it is the difference between a claim you can defend and one you cannot.