Property Management Chatbot: Maintenance

Triage before anything else

In most sectors the bot's first job is to understand the question. Here it is to establish urgency, because the cost of a delayed response varies by orders of magnitude between message types.

A water leak, a gas smell, a lift with someone inside, a failed lock on a ground-floor door and no heating in winter are all emergencies where minutes matter and where a bot asking three qualifying questions is actively harmful. A dripping tap, a broken cupboard door and a question about the parking rules can wait until morning.

So the design starts with a classifier that is deliberately biased towards caution. In this sector a false emergency costs someone a phone call; a missed emergency costs a building. Tune it accordingly and accept the false positives.

The emergency path

• Detect on keywords and route immediately, without qualification. Water, gas, smoke, fire, electrical, lift, lock, no heat, no water, flooding, a person trapped.
• Give the safety instruction first, quoted from the managing agent's own written procedure — where the stopcock is, not to use switches if there is a gas smell, the emergency number to call.
• State the official emergency services number where the situation warrants it, before anything about the building's own process.
• Route to a person or the out-of-hours line with a flag, not into the general maintenance queue.
• Never ask for a photograph before routing an emergency. Photographs are useful for a dripping tap and an obstruction when someone is standing in water.
• Have this path written in every language the tenants speak. A tenant in an emergency reverts to their first language without exception.

What a maintenance ticket must contain

Most maintenance delays are not caused by the contractor. They are caused by a ticket that does not contain enough to dispatch, which produces a phone call, a wait and a second visit.

1. 1Location, preciselyBuilding, unit, and where in the unit. 'The bathroom tap in flat 14B, the basin not the bath' prevents a wasted visit, and tenants supply this readily when asked directly.
2. 2What is happening, in the tenant's wordsNot a category. 'Water comes out brown for the first minute' is a diagnosis aid; 'plumbing issue' is not.
3. 3Since when, and whether it is getting worseThis is what separates a routine job from one that should be brought forward, and it is almost never on a standard form.
4. 4A photograph, where it helpsFor visible damage, leaks and appliance error codes. Ask for it after the ticket is captured, not as a condition of raising one.
5. 5Access arrangementsWhen the tenant is available, whether a key is held, whether there is a pet, whether anyone else needs to be present. Missing access information is the single most common cause of a failed appointment.
6. 6Contact preferenceHow the contractor should reach them and on what number. An assigned job nobody can arrange is not progress.

Tenant questions that automate well

• Rent dates, payment methods and what a reference should contain.
• What the tenant is responsible for versus the landlord or agent, quoted from the tenancy documentation rather than summarised.
• Bin collection days, recycling rules and where the bins are.
• Parking rules, visitor parking and how permits work.
• Building rules: noise hours, pets, alterations, subletting, what needs written permission.
• How to report a problem, what the response times are and what happens next.
• Move-in and move-out procedures, notice periods and the deposit process as published.
• Communal facilities: access, opening hours, booking rules.

Where a person must take over

• Anything about the deposit, deductions or its return, which is a decision with a legal framework around it.
• Rent arrears, payment plans and hardship, which need judgement and carry obligations.
• Notice, eviction and lease termination in either direction.
• Disputes between neighbours, which are rarely about the facts they are presented as.
• Any complaint about the agent or the landlord, immediately and without a qualifying question.
• Anything touching legal rights or obligations under a tenancy, where the bot may quote the document and must not interpret it.
• Safety certificates, insurance and compliance questions, which should quote the record or route to whoever holds it.

What to measure

1. 1Emergency routing accuracyBoth directions: emergencies that were not routed as such, and routine issues that were. The first number is the one that matters and it can only be found by reading.
2. 2Complete-ticket rateTickets containing location, description, duration and access details. This is the number that determines how many jobs need a second visit.
3. 3First-visit fix rateJobs resolved on the first attendance. It is the clearest downstream measure of ticket quality and it is a cost the agent already tracks.
4. 4Out-of-hours volumeTenant issues do not respect office hours, and evenings and weekends are when the expensive ones occur.
5. 5Repeat contact rateTenants chasing an existing ticket. A high figure usually means the acknowledgement did not say what happens next or when.

How Vexvon fits property management

The bot answers from your own material — your site crawled to as many as 4,500 pages, PDFs up to 25 MB for tenancy documents and building rules, and manual or bulk entries at up to 200 per request. Entries are typed into thirteen categories including branch, policy and opening hours, so building-specific rules can be held separately from agency-wide ones rather than blurred together.

Extraction runs on the conversation rather than a form, which suits a tenant describing a problem in their own words: phone numbers are detected through a five-stage process, free-form fields such as unit, location and access are captured alongside, and a one-sentence summary is written to the record. Photographs sent by the tenant can be understood on the channels that support image input.

The escalation mechanisms are explicit and logged, which is what the emergency path needs: a request for an operator raises a notification, an agent's stop character pauses the AI on that conversation for thirty minutes, and the AI can be switched off for a conversation entirely. A ticket reaches the maintenance group as a Telegram notification with a Start button, and whoever presses it is assigned.

The record then carries the job through a nine-stage pipeline with a priority marker, a close reason and a contact-attempt count, with reminders checked every minute and every change written to an activity log covering ten action types. Because all channels share one customer record, a tenant who reported on WhatsApp and then called is one thread rather than two tickets.

Frequently asked questions

1. 1What is the main job of a property management chatbot?Triage. A tenant message can be an emergency, a routine repair or an information question, and those need completely different speeds. Establishing urgency matters more than answering well.
2. 2How should emergencies be handled?Detect on keywords and route immediately without qualification, give the safety instruction quoted from the written procedure, state the emergency services number where warranted, and flag it to a person or the out-of-hours line.
3. 3What must a maintenance ticket contain?Precise location, what is happening in the tenant's own words, how long it has been going on, a photograph where it helps, access arrangements and a contact preference. Missing access details are the most common cause of a failed visit.
4. 4Which tenant questions automate well?Rent dates and payment methods, responsibility splits quoted from the tenancy, bins, parking, building rules, reporting procedures and move-in or move-out processes.
5. 5What must always go to a person?Deposits and deductions, arrears and payment plans, notice and termination, neighbour disputes, complaints about the agent, and anything requiring interpretation of a tenancy agreement.
6. 6Which metric matters most?Emergency routing accuracy, measured in both directions. After that, first-visit fix rate, because it is the clearest downstream measure of whether tickets are actionable.

Read last month's out-of-hours messages

Take every tenant message received outside office hours last month and classify each as emergency, routine or information. Then check what actually happened to the emergencies. In most agencies that single exercise finds both the gap in the process and the business case for closing it, and it takes an afternoon.