Routine Maintenance Call or Emergency? How an AI Phone Agent Should Tell Them Apart

If you look through a month of maintenance calls, most of them are things that belong in tomorrow's queue. A leaky faucet. A dishwasher that won't drain. A bedroom outlet that stopped working. AC that runs but doesn't cool.

A small number are different. A gas smell. Water coming through a ceiling. Smoke. No heat during a freeze. A resident locked out with a child inside.

Your team already knows the difference. The problem is that the line has to know it too, at 2am on a holiday.

The list isn't the same for every company, either. No heat during a freeze is on almost everyone's list. An AC failure might be one too if an elderly resident is in a top floor unit during a heat wave. If your policy treats it that way, it goes on the list. So the first rule is simple: the emergency list belongs to your team, and the AI follows it.

Every call would start the same way. The agent asks three things:

1. What's wrong?
2. Which property and which unit?
3. Is anyone in danger, or is there water, gas, smoke or sparks involved?

Then it reads the answer back. "So the kitchen faucet in unit 4B at Maple Court is dripping, and nobody's at risk. Is that right?" The read back matters more than it sounds. It catches the wrong unit number before a technician drives to the wrong building, and it gives the resident a chance to add the detail they forgot, like "actually, the water's going under the cabinet."

When the answers point to a routine problem, the agent's job is paperwork. It writes the work order in your property software with the resident's own words, the unit, the best time to reach them, permission to enter, and anything a technician should know before walking in, like a dog or a broken buzzer.

Then it sends the resident a text confirming what was logged. That text does two jobs. The resident knows they were heard, and they have something to reply to if the description is wrong.

Nothing about this is a judgment call. It's the same intake form your coordinators fill in every day, done at 11pm without anyone being woken up.

Here's the part I feel most strongly about. Emergency calls shouldn't depend on the AI weighing up how serious something sounds.

The emergency path should be a fixed rule your team writes. If the caller mentions gas, flooding, smoke, sparks, no heat below a temperature you set, or anything else on your list, the call goes straight to whoever is on call. If they don't pick up, it goes to the backup, then the manager. The chain is yours to define.

While that transfer happens, the agent gives the resident your safety instructions, word for word as your team wrote them. For a gas smell, that usually means leaving the unit right away and calling the gas company's emergency line or 911 from outside. The AI never tells a resident that a gas smell can wait until morning, because it never gets to make that call in the first place.

Phone calls are messy. The line is bad, the caller is upset, or the problem is ambiguous. "There's water on the floor" could be an overflowing toilet or a burst pipe behind the wall.

So the third path matters as much as the other two: when the agent isn't confident about what it heard, it hands the call to a person. It doesn't guess. A transfer costs a minute of someone's time. A wrong guess on a water leak can cost a ceiling, a unit below it, and a resident's trust.

Every call would also leave a transcript and a short note on why it went the way it did: "logged as routine, resident confirmed no water spreading" or "sent to on call, caller mentioned gas." When a resident later says "I called and nobody did anything," you can see exactly what happened.

The AI's job here is narrow. It answers, asks the three questions, logs routine requests as work orders, sends confirmations, and routes emergencies and unclear calls to people. That's it.

People stay in charge of everything that matters:

• The emergency list and the on call chain. Your team writes them and changes them.
• Dispatch. A person decides who goes out, when, and whether a vendor is needed.
• Entry and cost decisions. The agent records permission to enter. It never approves spending or schedules a vendor on its own.
• Anything a resident disputes. Complaints, charge backs and "this has happened five times" calls go to a person.
• Fair treatment. The agent sorts calls only by what's wrong. Every resident gets the same questions and the same rules, and your team can check the transcripts to confirm it.
• Review. Someone on your team reads the previous night's calls each morning, at least during the first months.

You don't need to put every property on an AI line to find out if this works. One property's after hours calls is enough.

Week 1: Write down your emergency list and on call chain.

Week 2: Have your own staff call the line and play residents. Include the tricky ones: the mumbled unit number, the resident who downplays a leak, the call that starts routine and turns urgent. Fix whatever goes wrong.

Week 3: Turn it on for one property's after hours calls. A person reviews every call the next morning.

Week 4: Look at what happened. Were emergencies routed fast? Were work orders accurate enough that technicians didn't have to call residents back for details? Did any resident complain about talking to an AI? Then decide whether to stop, adjust, or extend it to a second property.