AI or Human Answering Service: Which Suits a Property Manager?

Most property managers do best with a hybrid: AI for routine calls (rent, viewings, repair logging), with anything that might be an emergency transferred to a person, either on-call staff or a human answering service. Pure AI suits portfolios with few emergencies; pure human suits older stock, where one mishandled burst pipe costs more than a year of the service.

The deciding question isn't which sounds more natural. It's what happens on the worst call of the month: a tenant at 2am with water coming through a ceiling light. Whichever option you choose has to recognise that call, reach a person who can act within minutes, and leave a record. Price that scenario first and the choice usually becomes obvious.

What property calls actually look like

Before comparing services, count your calls for two weeks. The mix drives everything. Here's an illustrative month for a manager of 300 homes, mostly flats with some older houses:

Two things stand out. Around 80% of calls don't need a person if they're captured accurately, which is AI's strength. And the 18 possible emergencies, most of them out of hours, are where any service earns or loses its keep.

How the two kinds of service are priced

Pricing models differ more than headline prices, so compare on the same unit. From the vendors' own pages in September 2026:

• Per call: Smith.ai charges $300 a month for 30 calls, $810 for 90 and $2,100 for 300, with overage from $11.50 down to $8.50 a call. The same plans apply whether you choose AI-first or human-first answering, month to month.
• Per minute, human: Ruby's plans run from $250 for 50 minutes to $1,725 for 500 minutes a month, with trained receptionists and 24/7 coverage.
• Per minute, AI: Rosie charges $49 a month for 250 minutes, $149 for 1,000 minutes (adding calendar booking and warm transfers, where the AI briefs your staff member before connecting) and $299 for 2,000 minutes.
• Per unit, human, maintenance only: Buildium's Maintenance Contact Center uses professional call-centre agents 24/7, including holidays, to take maintenance calls and decide whether they're emergencies. It costs $1.10 per unit a month for all maintenance calls or $0.70 for missed ones only, with a $140 monthly minimum and a $300 set-up fee.
• Built into your software: AppFolio's AI maintenance tools talk to residents by text and chat, review photos, walk them through simple fixes, create work orders, classify emergencies and dispatch vendors under rules you set. If you're already on a platform like this, check what's included before buying a phone service.

The same 300 homes, priced four ways

Using the call mix above: 420 calls a month at an average of three minutes is about 1,260 minutes, of which roughly 130 calls (about 390 minutes) come after hours.

The spread is tenfold, and the cheapest option isn't automatically wrong. What the table can't show is the cost of a missed emergency. If a single mishandled leak each year leads to a $6,000 repair and an unhappy owner, the human element in the hybrid (about $4,000 a year) looks cheap. Get your own quotes, because these vendors change plans and overage rates without much notice.

To redo the sums for your own portfolio, you need three numbers from your two-week count: calls per month, average call length in minutes, and the share that arrive after hours. Per-minute services cost roughly calls × minutes against the plan's allowance; per-call services cost calls against the plan's allowance plus overage; per-unit services cost units × the unit rate, subject to any minimum. Ask every vendor what counts as a call or a minute (spam, hang-ups, transfers, time on hold) because those definitions can move a monthly bill noticeably.

The same method gives a very different answer for a small portfolio. An owner-manager with 40 homes, at the same 1.4 calls per home a month, gets about 56 calls, or roughly 170 minutes at three minutes each. That fits inside Rosie's 250-minute plan at $49. On Smith.ai, 56 calls is the 30-call plan at $300 plus 26 overage calls at up to $11.50 each, around $600, or the 90-call plan at $810. At that size, a per-call human-first service costs more than ten times the AI plan, which is why the owner usually becomes the on-call person and the AI's job is to answer, log and put the real emergencies straight through.

Head to head on what matters to a property manager

The emergency rules that decide it

Write your emergency rules down before you talk to any vendor, and make each vendor show you how their service applies them. A filled-in routing card might look like this (illustrative):

EMERGENCY: transfer to on-call manager immediately, then text the
contractor on the rota.
- Water coming through a ceiling, light fitting or electrics
- Burst pipe or water that can't be stopped at the stopcock
- Smell of gas (tell the caller to leave and ring the gas emergency
  line first)
- No heating or hot water in cold weather, for vulnerable residents
- Resident locked out at night, or a door or window that won't secure
- Roof damage letting rain into a living space
- Fire, flood or break-in in progress (tell the caller to ring the
  emergency services first)

URGENT: log and alert on-call by text, contractor next morning.
- No heating in mild weather; one appliance down; leak contained

ROUTINE: log as a work order, confirm reference to the caller.
- Everything else

A realistic failure shows why the wording matters. A tenant rings at night and says, "There's a drip coming from the light in the bathroom, it's only small." An AI answering service classifies it as a routine leak, because the caller called it small and nothing in its rules mentioned light fittings. By morning the circuit has tripped and the ceiling is stained. Adding "water near a light fitting or electrics is always an emergency, whatever the caller says about size" fixes it. Test your own rules against descriptions like this, not against tidy textbook phrases. There's a fuller method in setting up an AI answering service that never misses an emergency.

The second failure to design out happens after the emergency is recognised. At 2am the AI correctly spots a burst pipe and warm-transfers the call, but the on-call manager's phone is on silent and the transfer goes to voicemail. The AI ends the call with "someone will call you back shortly", and nobody does until 6am. The emergency card needs a fallback chain, written in the same place: if the on-call manager doesn't answer within 30 seconds, try the second on-call number; if that fails, text the rota plumber with the address directly; and tell the caller plainly what has happened and how to shut off the water at the stopcock while they wait. Test the chain by making one emergency test call with the on-call phone switched off.

A storm night tests the opposite problem. Forty tenants in the same block ring within an hour about water coming through the top-floor ceilings after the same roof failure. An AI service answers every call, which a human queue couldn't, but forty warm transfers to one on-call manager make the phone useless for the job itself. Agree an incident mode with the vendor: once the manager confirms a known incident at an address, the AI tells callers from that building what's happening and when the roofer is due, takes any new details (a different flat, someone vulnerable, electrics affected) and only transfers calls that add something new.

Before and after: the out-of-hours process

Before (illustrative): after 6pm, calls go to a voicemail that says "for emergencies ring the on-call mobile". Half of callers leave a voicemail anyway. The on-call manager checks messages at 10pm and 7am. Repair reports are typed up from voicemails the next morning, often with no flat number.

After: every call is answered. The AI asks for name, address and flat number, what's wrong and whether water, gas, electrics or security are involved. Routine reports become work orders with photos requested by text. Anything matching the emergency card is warm-transferred to the on-call manager, who hears a one-line summary before connecting, and the rota contractor (the locksmith, the roofer, the heating engineer) gets a text with the address. The manager wakes up to a list, not a pile of voicemails.

The difference shows most in the repair reports themselves. Before, a typical voicemail typed up the next morning read: "Hi it's about the boiler again, it's making the noise, can someone come, thanks." No name that matched a tenancy, no address, no callback number that worked. After, the same call becomes a work order like this (illustrative):

• Property: Flat 4, [building name], [street]
• Caller: named tenant, number confirmed
• Issue: boiler making a loud banging noise when heating comes on; heating and hot water still working
• History: caller says it's the second time this month
• Priority: routine (heat and hot water working, no leak, no smell of gas)
• Access: tenant home weekdays after 3pm; no pets
• Photos/video: requested by text

The "history" line is worth checking against your own records rather than trusting. If the software shows a previous job on the same boiler, the contractor goes with the earlier notes; if it doesn't, the tenant may be remembering a call that was never logged, which is its own finding about the old process.

Ten test calls to make before you sign

Every vendor offers a trial or a demo number. Make these calls yourself, and score each as pass, partial or fail:

1. A vague leak: "Something's dripping from the ceiling near the light."
2. A gas smell, described hesitantly.
3. A lockout at midnight from someone who's not the named tenant.
4. No heating, from an older resident who mentions they're unwell.
5. A rent question that needs the tenant's account balance.
6. A viewing request for a property that's already let.
7. An angry neighbour complaint that starts with swearing.
8. A caller who speaks limited English and switches language.
9. A contractor asking for the key-safe code.
10. A caller asking whether the landlord accepts applicants who receive housing support.

A filled-in scorecard from one AI trial (illustrative): passes on 1, 2, 4, 6 and 8; partial on 5 (took details but couldn't see balances, which is correct but needs a clear promise of a call-back) and 7 (handed over, but only after two minutes); fails on 3 (it offered to send a locksmith without checking the caller's identity) and 9 (it read out a code from the property notes, which should never have been in its knowledge). Call 10 was a pass only because the manager had written the answer into the script; applicant questions are where consistent, fair treatment matters, and your script should give one approved answer. Two fails in ten is normal at first; the question is whether the vendor can fix them within a week.

The approved answer can be short. One manager's version, before and after (illustrative). Before, the AI improvised: "I'm not sure, some landlords do and some don't, you could try applying and see." After, the script said: "Every application is assessed on the same published criteria. I can text you those criteria now, and a member of the lettings team can answer anything they don't cover during office hours." The second version doesn't guess at a landlord's policy, treats every caller the same way and creates a record that the criteria were sent. Check the criteria themselves with whoever advises you on lettings law, because what a landlord may lawfully consider varies.

Which suits which portfolio

Whatever you choose, review a sample of call transcripts every week for the first month. The service will only be as good as your emergency card and your script, and both improve fastest in the first four weeks. Make the review a count, not a skim. A first-week check of 20 transcripts from the 300-home portfolio might read: 3 of 3 possible emergencies transferred, one after 70 seconds of questions that should have come later; 14 of 15 repair reports with a flat number (the miss was a caller who gave only the street); 2 viewing requests booked into the wrong branch diary. Each line points to one change in the script or the settings, and next week's count shows whether it worked. For triaging the repair reports the calls produce, see how property managers use AI to triage maintenance requests, and for the wider decision between AI receptionists and traditional services, AI receptionist or answering service.

Further reads

• How Much of a Property Manager's Week Can AI Take Over? — Which other parts of a property manager's week AI can take.
• AI for Letting Agents: Handle Tenant Queries Without Extra Staff — Handling tenant queries by message as well as by phone.
• How to Write Call Scripts and Escalation Rules for an AI Receptionist — Writing the scripts and escalation rules in detail.
• How to Measure an AI Receptionist's Return in the First 90 Days — Checking whether the service is paying back after launch.
• AI vs Human Answering Service for HVAC Firms — How your HVAC contractors are making the same choice.
• Can an AI Chatbot Book Valuations for Estate Agents Out of Hours? — What an out-of-hours valuation bot needs: live diary access, eight qualifying questions, written diary rules and a list of things it must never say.
• AI Inventory and Check-Out Reports: A Letting Agent's Guide — How AI drafts inventory and check-out reports from photos or video, where its descriptions go wrong, and how to keep the report defensible in a deposit dispute.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai receptionist pricing page; Ruby pricing page; Rosie (heyrosie.com) pricing page; Buildium Maintenance Contact Center page; AppFolio product pages on AI maintenance handling. Prices checked September 2026.