A fire, an unconscious tenant, and an AI on the phone: what actually happened

Early one evening in January, out of hours, a tenant in a London block called their managing agent's emergency line to report smoke coming from the flat below. The call hit our AI receptionist. Within seconds, it had routed the call through to the most senior property manager, who picked up the phone himself, told the tenant to dial 999, and said he would be on site within fifteen minutes.

By the time he arrived, police, fire brigade and ambulance were already on the scene. The tenant in the flat below had been found unconscious. He had left a pan on the hob and it had started a fire. He was alive because the emergency services had been called in time.

In the days that followed, the insurer got involved. They wanted a timeline. Who called what, when, who was notified, what was said, what was decided. Our software had every minute of it. The original call, the routing decision, the property manager's response, the case that opened automatically the moment the call ended, every follow-up email between the agency, the freeholder and the loss adjuster, all attached to the same record, all timestamped, all retrievable in seconds.

We had sold this firm on a pitch about call deflection, response times, and operational efficiency. None of that was wrong. But at no point in the sales process had we described OdjoAI as fire safety infrastructure, or as the source of truth in a serious insurance claim, because we did not yet think of it that way.

This is a blog about why out-of-hours and emergency call handling in UK property management is broken, what it costs when it fails, and what AI call handling actually does about it.

Why the out-of-hours problem has never been solved properly

Property emergencies do not respect office hours. Pipes burst at eleven at night, boilers fail on a Sunday morning, lifts trap people on bank holidays, and on rare but consequential occasions a tenant rings their managing agent's emergency line to report smoke coming from a stairwell. The industry has known this forever. The response, until very recently, has been to outsource the problem to human call-handling services like PropCall or Message Direct, which take the call, work through a triage script, send an email to the office, and bill somewhere between four and eight pounds per call plus a retainer.

These services are competent and they fill a real gap. But anyone who has spent time inside a property management firm knows the limitations. A human call handler reading from a script, asking for the spelling of a road name they have never heard before, working through a decision tree at the speed of speech, takes between four and seven minutes per call on average. That is fine for a maintenance request about a dripping tap. It is much less fine when the caller is reporting smoke in a stairwell.

The deeper problem is that none of this scales. A block management firm of any size will spend between thirty and sixty thousand pounds a year on out-of-hours call handling that, in the overwhelming majority of cases, is just taking messages for the team to read the next morning. The cost is bad. The variance is worse. The agent who picks up at 2am on a Tuesday in February is not the same agent who picks up at 11am on a Saturday in August, and the experience the tenant gets, and the handover note the property manager wakes up to, varies accordingly.

And then there is the fourth option, which is what most UK property firms actually do, which is route the emergency line to a property manager's personal mobile and hope they hear it ring. This is the version of the system that fails most often, fails most expensively, and fails the most quietly, because nobody writes a case file when the phone went to voicemail and the tenant gave up and went back to bed.

What an AI receptionist actually does that a human service cannot

The voice AI category had a long and well-deserved reputation for being terrible. The phone trees that asked you to "please state the reason for your call" and then misheard you four times in a row are the same lineage as the early chatbots that escalated to a human the second anything went off-script. Nobody trusted them with anything that mattered, and nobody was wrong to feel that way.

That has changed in the last eighteen months. The current generation of large language model voice agents can hold a multi-turn conversation with a stressed tenant in the middle of the night, understand context, follow a property-specific triage protocol, and make a real decision about who needs to be woken up and who can wait until morning. They do not get tired. They do not get flustered. They answer on the first ring. They do this for every call, in every language the firm supports, at every hour of every day, without variance.

For UK property management, this changes the economics and the duty-of-care calculation simultaneously. It is the reason an AI receptionist on the fire call from January did what no human-staffed out-of-hours service in the country would have done at the same speed, which is recognise the call for what it was inside the first few seconds and put the senior property manager on the phone before the tenant had finished describing what they could see.

The boring volume is where the business case lives

We want to be careful not to lean too hard on the fire call. It is the dramatic example. It is not the everyday one. Most of the value we deliver is in the unglamorous middle of the call distribution, the calls that never make it into anyone's case study deck.

A mid-sized letting agency or block management firm takes somewhere between fifty and two hundred calls a day, and roughly eighty per cent of them are about the same handful of things. Rent queries. Repair status updates. Viewing bookings. Contractor coordination. Tenancy renewals. The slow constant stream of leaseholder questions about service charges, major works, and Section 20 consultations that block management teams field every day of the week.

Our AI handles the great majority of these end-to-end, without ever escalating to a human. It looks up the tenancy in the firm's property management system in real time, opens cases directly into the case management workflow, books contractor visits, sends confirmations by email and SMS, and logs everything against the relevant property and tenant record. The numbers from the firm we ran the fire call for, six months in:

Around four thousand calls and two thousand emails handled a month. Average call routing time down from two minutes to under a minute. Zero calls missed since deployment. More than seventy individual staff inboxes consolidated into one triaged feed. A receptionist role on the team that, when it falls vacant, is not being recruited for.

These are not modelled projections. They are operating metrics from a live portfolio.

The Renters' Rights Act made this a compliance question, not just an efficiency one

The legal context shifted underneath every UK property management firm on 1 May 2026, and we do not think the industry has fully absorbed what it means for call handling.

Under the Renters' Rights Act, every tenant interaction now sits inside a tighter regulatory frame. Pet requests have a twenty-eight day response window with automatic approval if missed. Disrepair complaints have escalating timelines tied to potential rent repayment orders. Data subject access requests can be raised by any tenant at any time, and the firm has thirty days to produce a complete audit trail of every piece of personal data it holds on that tenant, which now explicitly includes call recordings and transcripts.

Property management firms that are still running on voicemail, on human call handlers who email a summary to the office Monday morning, or on property managers who answer their personal mobile when they remember to take it off silent, are going to find themselves on the wrong side of the new compliance regime within months. There is no version of the new statutory framework where "we will get back to you on Monday" survives a tribunal challenge from a tenant whose pet request expired in the meantime, or whose disrepair complaint sat in a voicemail box for three days.

This is the part the fire call brought into focus for us. The audit trail that turned out to be the source of truth for the insurer is the same audit trail that, in a different scenario, would be the source of truth for a disrepair tribunal, a DSAR response, a safeguarding investigation, or a coroner's request for evidence. The system was not designed for any one of those situations specifically. It was designed to be the place where everything that happens between a property firm and a tenant is recorded, structured, and retrievable. That is what compliance under the new Act actually requires. Most firms are not set up to do it. The ones who are will not have to think about it.

What to look for in an AI call handling system for property management

Not every AI voice agent is built for property. The generic AI receptionists, the ones designed for dentists, gyms, and law firms, will not survive contact with a leaseholder asking about a Section 20 consultation at seven o'clock on a Friday evening. They do not know what a managing agent is, what a freeholder is, what an out-of-hours emergency looks like in a block of flats, or what the difference is between a tenant query and a landlord query.

The five things that actually matter in a property-specific system, in our experience:

1. Triage by caller type. A leaseholder, a tenant, a landlord, and a contractor all need different routing and different information. The system needs to work out which is which inside the first thirty seconds of the call.
2. Live integration with the property management system. Without real-time access to the tenancy database, the property record, the open cases, and the contractor list, the AI is just a glorified answering machine. It needs to be able to look things up, raise things, and log things, in the moment, against the right property.
3. Native handling of the seven or eight emergency categories that genuinely require it. Fire, flood, gas, structural risk, injury, lift entrapment, no heat in winter, no water. Each has its own protocol. The system needs to know what they are and execute them without hesitation.
4. Intelligent escalation. Not every call needs the senior property manager. Most do not. The AI needs to distinguish between the call that can wait until Monday and the call that requires the most senior person on the line right now, and it needs to be right every time.
5. A complete audit trail by default. Every call, every transcript, every routing decision, every escalation, every case, every email attached to that case, all logged against the property and the tenant, all timestamped, all exportable. This is not a feature. It is the architecture. The compliance value is a consequence of having built the system this way from the beginning.

The window is closing faster than the industry realises

UK property management has been operating with broken out-of-hours infrastructure for thirty years. The reason it has survived this long is that there has been no credible alternative. Voicemail was bad, human call handling was expensive and slow, and AI was not good enough. That equation flipped in the last eighteen months. AI voice agents are now good enough, cheap enough, and reliable enough to handle the entire front line of UK property management call traffic, and the regulatory pressure from the Renters' Rights Act is going to force the issue for every firm that has been putting the decision off.

The firms that move first are going to lock in operational advantages, cost, compliance, tenant satisfaction, landlord retention, and audit-readiness, that the rest of the industry will spend years trying to catch up with. The firms that wait are going to find out, one statutory deadline or one insurance claim or one tribunal hearing at a time, that the system they have been running on is no longer fit for the job.

The fire call in January was the moment we understood what we had actually built. If a fire alarm goes off in one of your buildings tonight, and a tenant calls your emergency line, what actually happens? If the answer is voicemail, a script-reading call handler six minutes deep into a triage tree, or a property manager whose phone is in another room, the answer is not good enough anymore. It is going to stop being defensible faster than most of the industry is prepared for.

OdjoAI builds AI communication and case management for UK letting agents and block management firms. Our AI receptionist handles inbound calls 24/7, triages by caller type, escalates emergencies to the right human within seconds, and builds a complete audit trail of every tenant interaction. Book a walkthrough at odjoai.com/property.