AI for property management pays back on the routine messages from tenants, applicants and landlords, and only if it answers from content your team has approved and hands everything else to a person. On the levelpartner.ai platform that is a Conversation AI bot trained on your lettings knowledge base, with handover rules for anything about money, repair liability or tenancy decisions. This article covers what that bot should answer, what it must pass on, how to build and train it on the platform, and how a 30-person lettings business would scope it.
Where the hours go in a lettings office
A property manager's week is dominated by small inbound messages. Is the flat on Mill Road still available? When is the inspection? Has the landlord approved the boiler quote? Has my deposit been registered? Each takes two minutes. Together they take most of the day.
Count them before you build anything. Ask the team to tag every inbound message for two weeks: availability and viewings, maintenance reports, rent and payment, documents, landlord updates, complaints, other. The tags tell you exactly what the bot should cover and what it should never touch.
Suppose the team tags 2,400 inbound messages over a fortnight: 820 on availability and viewings, 610 maintenance reports, 340 on rent and payment, 220 document requests, 190 landlord updates, 60 complaints and 160 other. In that split, availability, viewing and document questions add up to 1,040 messages, roughly two in five, that could be answered from listings and approved policy text. Maintenance reports could be collected but not judged, and everything else would stay with people. The figures are illustrative.
What the bot answers and what it hands over
Draw the line before anyone opens the builder. The rule is simple: the bot answers anything your approved content already answers, and hands over anything that commits the business to money, liability or a decision about a tenancy.
Safe to answer from approved content
- Availability, rent, move-in dates and what is included, taken from the current listing.
- How the application process works, which documents are needed and what happens next.
- Viewing slots, booked into the right calendar.
- How to report a repair, collecting what, where, since when and a photo.
- Office hours, how rent is paid, how to give notice in principle.
Always handed to a person
- Anything about money that is not on a published list: arrears, deposit deductions, fee disputes, rent changes.
- Repair liability: who pays, whether a contractor attends, when. The bot collects the report and stops.
- Tenancy decisions: whether an applicant is accepted, renewals, notices to end a tenancy, permission for pets or subletting.
- Complaints, and any message suggesting a tenant is vulnerable.
- Emergencies such as gas, flooding, fire or no heating in winter, which go to the on-call person by phone with only your approved safety text from the bot.
Tenant selection stays with people. In the United States fair housing rules govern how applicants are treated; in the United Kingdom equality law and the rules on rental discrimination do. A bot that hints an applicant "should be fine" has made a decision in the applicant's eyes.
Before you start
- The list of safe topics and handover topics above, signed off by a director.
- Approved content for the safe topics: listing pages, application guide, repairs reporting guide and an FAQ.
- Calendars for viewings, assigned to the right people.
- A named person who receives handovers in working hours, and the out-of-hours route for emergencies.
How to set it up on the platform
The prompt based route suits lettings, because most of the work is in the instructions: what to answer, what to refuse and when to hand over.
- Go to AI Agents, then Conversation AI, then + Create Bot, and select Prompt Based Bot.
- Pick a starting option: General Q&A, Appointment Booking, Marketplace Template or Start from Scratch.
- Continue into the bot editor.
- Configure Bot Settings and Bot Training.
- Configure Bot Goals with the prompts, model settings and actions the bot should use.
Write the handover list into the goals in plain words: "If the message is about arrears, deposits, repair costs, a tenancy decision or a complaint, do not answer. Say a named member of the team will reply and by when." If you build with the flow builder instead, Human handover is one of the actions you can place in the flow. Either way, turn on email notifications so the right people hear when the bot cannot answer.
How to train it on approved lettings content
The bot is only as good as what it reads. Train it on the documents you would hand a new property manager on their first day, and nothing older.
- Go to AI Agents, then Conversation AI, and open Agent List.
- Select your lettings bot and open the Bot Training tab.
- In the web crawler, choose Exact URL and enter the full address of an approved page, including https://.
- Click Get Data and review the pages it finds.
- Select only the pages that answer tenant, applicant or landlord questions, then click Train Bot.
- Wait for each selected address to show Trained.
- Click + Add Q & A, enter a question as a tenant would ask it and a short complete answer, and save it.
- In Test your Bot, try a common question, a reworded version, an FAQ and an out-of-scope question.
Use Exact URL rather than whole-domain crawling. A whole-domain crawl picks up old blog posts and last year's fee page, and the bot will quote them. For listing and policy pages that change, turn on Auto Refresh with a daily or weekly schedule so the bot does not offer a flat that let last Tuesday.
Property management AI for each audience
Applicants, tenants and landlords ask different questions and need different tone. One bot can serve all three if its instructions say who is who, using a contact tag for role. Larger offices often split them into separate bots so a landlord never receives applicant chatter.
Landlords want three facts: is it let, is the rent paid, what is happening with repairs. The bot can tell them where to find their statement and confirm a repair has been logged. It does not discuss costs, contractor choice or arrears action. Those are handed over every time.
Training the bot on the whole website and the full tenancy agreement. It will paraphrase clauses about deposits and break clauses to tenants who ask in passing, and a paraphrase of a contract is advice. Keep legal documents out of training. Link to them and hand over the questions.
Testing before it goes live
Run it in Suggestive mode first, where the bot drafts each reply and a property manager sends it. Two weeks of drafts will show you every gap in the content and every handover topic it misreads. For a fuller test plan, including the scripted questions to run before switching on, read testing property management AI before launch.
After launch, Response Info beside AI messages in Conversations shows which knowledge base sources produced a reply. When an answer is wrong, that tells you which page to fix.
What most articles on AI in property management get wrong
They lead with tenant screening and rent pricing, because those sound advanced. They are also the uses with the highest legal risk and the least time saved. Screening happens once per tenancy. Messages happen every hour.
The other error is treating handover as a failure. A bot that hands over a deposit question in ten seconds, with the conversation already on the record, has done its job. The aim is not the fewest handovers. It is the right ones, fast.
The trade-off
A bot on the inbound queue makes the office faster and more consistent, and a little less personal at first contact. Some long-standing landlords prefer to speak to the person they know. Let property managers flag those contacts so the bot hands them straight over, and state in the first message that it is an automated assistant with a person available.
What you give up is the feeling that every message is handled by someone familiar. What you get is a team that can look after more units without more staff, and replies that do not depend on who is on leave. For how the sales side of a real estate business uses AI differently, see conversational AI for real estate sales enquiries.
Measuring it
Each month track the time to first reply by message type, handovers by reason and the share of maintenance reports that arrive complete. If handovers marked "wrong answer" rise, read those conversations and fix the content before widening the scope. Keep role tags and listing status current on each record, as covered in organising real estate contacts by role.
Frequently asked questions
Can AI screen tenants?
It should not make or influence screening decisions. It can explain the process, list the documents required and book a viewing. The assessment belongs to a person working within the fair housing or equality rules that apply.
What should the bot do with a repair report?
Collect a complete report, log it on the tenant's record and tell the tenant a named person will be in touch. It never promises a date, a contractor or who pays.
How often should we retrain it?
Whenever a policy, fee or listing changes. Auto Refresh handles web pages on a schedule; FAQs and uploaded documents need a person to update them.
Will tenants mind talking to a bot?
Most care more about speed and accuracy. Say clearly it is an assistant, make reaching a person easy and never use it for complaints or arrears.