AI can only read the records you keep

Landlords and property managers have taken to AI fast. They use ChatGPT and its rivals to draft notices, polish listings and ask what the law says, and most of them would tell you it saves time.

So here is a fair question: if AI is that good, why can't it tell you which tenants owe rent this month, or produce the owner statement that is due on Friday?

The answer has very little to do with the AI. It has everything to do with what you hand it. An AI can only read the records you keep. If the records live in a receipt book, a WhatsApp group and your memory, there is nothing for it to read.

Everyone is trying AI. Almost nobody lets it run anything.

The numbers from the US, where property managers have had AI built into their software for a few years now, tell the story. In a September 2026 survey of 3,662 property management professionals, 58% said they use AI in some form. Only 8% had a written policy for how to use it, and Buildium's research finds the same 8% have fully automated even one workflow.

What are all those managers using it for? Mostly writing. Property descriptions, replies to tenants, first drafts of notices. Among small independent landlords, the top uses are asking about tenancy rules, drafting notices and writing listings.

Writing is a natural first job for AI because it needs no records at all. You give it a few facts in the prompt and it gives you a paragraph. The jobs that would actually save a property manager hours are different. They need records:

• Who owes what this month needs every payment matched to the tenant and the month it pays.
• The owner statement needs every expense recorded against the right property, with a receipt behind it.
• Settling a water bill dispute needs last month's reading, this month's reading and a photo of the meter.
• Reconciling the bank or mobile money statement needs a list of units and tenants to match each payment against.

None of these can be done from a chat window, however clever the model. Gartner expects organisations to abandon 60% of AI projects that lack AI-ready data. In commercial real estate, Deloitte finds 92% of firms still researching or piloting AI and only 8% with it built into how they work. Large companies with data teams are stuck on the same thing a ten-unit landlord is: the records.

What happens when AI meets messy records

Picture Jane, the tenant in unit A3, paying her rent over a few months. In July she types "A3" as the payment reference. In August it is "a 3", in September "apt3". In October she pays by mobile money with no reference at all, only her phone number. The same week, another 1,200 arrives from a number you have never seen. Whether your tenants pay by bank transfer, M-Pesa or a payment app, the reference is whatever they decide to type.

A person who knows the building sorts this out in a minute. An AI with no unit list has three choices: guess, ask someone, or hand the job back to a person. That is how one consultancy describes AI agents working on split records, and it is the honest description. Guessing on money is how a tenant gets an arrears letter for rent they paid. Asking someone every time saves nobody any time.

Give the same AI a fixed unit list (Unit A3, Jane W., rent 1,200, phone ending 418) and the problem changes completely. Every spelling of A3 lands in the same place, and her mobile money payment matches on her phone number. The payment from the unknown number, perhaps a relative paying for another tenant, still goes to a person, which is exactly where it should go.

There is a second, quieter risk. AI models are good at reading a single figure and much worse at combining many. One 2025 test on company annual reports found top models answering simple look-ups correctly almost every time, then falling to near zero on questions that needed several figures combined. An arrears balance or an owner statement is exactly that kind of question. The totals have to come from your ledger, not from the model.

A photo is a check, not a record

Meter readings are a good example of where AI genuinely helps and where it doesn't. Reading a meter from a photo sounds like a solved problem. Look at the numbers, though. In a published study of automated dial-meter reading, the system read individual dials correctly 93.6% of the time. Because every dial has to be right for the reading to be right, it read whole meters correctly only 75% of the time.

That does not make the photo useless. It makes it a check. When your caretaker records a reading, attaches the photo, and the system already knows last month's figure, anything that looks wrong can be flagged before a bill goes out. Without the recorded reading, the photo is just a picture, and the tenant disputing their water bill has as much evidence as you do.

The six records to put in place first

You do not need a data team. You need six things, and each one unlocks a specific job you could later hand to AI. Just as important is what stays with you.

1. A fixed unit list

One name for every unit, used everywhere: on the lease, on the invoice, in the payment reference you ask tenants to use. Each unit linked to its current tenant, their phone number and their lease dates. This is what turns "a 3" and "apt3" into a match.

2. Every payment linked to what it pays

Each payment recorded with its date, method and transaction reference, and linked to the invoice or month it settles. Totals are not enough. When a tenant says they paid in March, you need to see the March payment, not a running balance that happens to look right.

3. Every expense against a property, with the receipt

Plumber, security, garbage collection: each one recorded against the property or unit it was for, with a photo of the receipt. This is what an owner statement is made of. AI can read the receipt and suggest the entry. You approve it.

4. Meter readings with the date, the previous reading and a photo

Recorded by whoever reads the meter, reviewed before anyone is billed. The previous reading is what makes the photo useful.

5. A history of changes

When a figure changes, you should be able to see who changed it and when. That is what answers an owner who asks why last month's statement looks different, and it is what lets you trust any summary built on top.

6. A short written rule for using AI

Fewer than one in ten US property managers have one, so you would be ahead of most. It can fit on one page: use a business version of the AI tool rather than a free personal account, never paste tenant ID numbers into a chatbot, and keep decisions about arrears, notices, deposits and emergencies with a person.

Start with this month, not with ten years

The most common reason people never fix their records is the size of the backlog. Years of receipt books, spreadsheets with three versions and WhatsApp threads nobody wants to scroll through. Don't start there.

Start with the tenants you have today. Set up the unit list, record this month's payments against this month's invoices, and photograph this month's receipts and meter readings. Within a couple of months you have clean, current records, which is what any report or any AI actually needs. History can be added later, or left where it is. If your current system is a spreadsheet, our earlier post on why spreadsheets feel fine until they don't explains why now is a better time than after the next dispute, and how to track rent payments covers the payment side step by step.

The rule worth remembering

When AI does reach the parts of your business that involve money, the safe arrangement is simple. It is also the pattern behind the AI that works in property management today: invoice reading and bank-statement matching that check against an existing ledger and hand the exceptions to a person.

AI will keep getting better at reading, drafting and spotting what looks wrong. It will not keep your records for you. The landlords who get real value from it will be the ones who can point at any figure in a report and show where it came from. That is worth doing now, whether or not you ever ask an AI to read it.

Sources

• Florida Realtors: AI use outpaces office policies and training (IREM and AppFolio study, September 2026)
• Buildium: tech shifts property managers can't ignore
• Avail: 2026 independent landlord survey
• Gartner: lack of AI-ready data puts AI projects at risk (February 2025)
• IREI: 92% of commercial real estate firms still haven't integrated AI (Deloitte 2027 outlook)
• Enterprise DNA: prepare property data before AI agents
• DEV Community: automating mobile money rent collection (the A3 / a3 / apt3 reference problem)
• arXiv 2005.03106: image-based automatic dial meter reading
• arXiv 2508.05201: hallucination on financial tables (citing the FAITH benchmark)