Most letting agents I've sat with in Ireland are running their business out of three places: a CRM that was built for sales, an inbox that's drowning, and a shared drive nobody fully trusts. The work — the actual letting work — happens in the gaps. A tenant emails about a leak on Sunday night. A BER cert is somewhere, probably. An RTB registration was done, but by whom and when. The property-management side of the Intelligence Brain is built for that mess. Not to replace the agent, but to give the agency a memory that doesn't quit at 6pm and doesn't forget what was said on a phone call three months ago.
Why generic AI doesn't fit a letting agency
If you put a public LLM in front of a property manager, two things go wrong fast. First, the model has no idea what your portfolio looks like, who your landlords are, which units are HAP, which are short-term, or what the standing instructions are for unit 4B at a particular address. It will guess, and it will guess plausibly, which is worse than guessing badly. Second, the data you'd need to feed it — tenancy agreements, landlord correspondence, deposit records, RTB registrations, contractor invoices — is exactly the data you cannot put into a public chat window without breaking GDPR and your own client confidentiality terms.
So the question isn't "should letting agents use AI". It's "where does the model run, what does it see, and who controls the audit trail". That's an architecture question, not a marketing one. The property-management brain is designed to run on hardware in your office, see only what you give it, and keep a record of every question asked and every document touched. That's the floor. Anything below that floor is a compliance problem dressed up as productivity.
What the property brain actually indexes
An agency's data is more structured than people realise — it's just spread across formats. The brain ingests and indexes the things that matter for daily letting work:
- Tenancy agreements — fixed-term, Part 4, further Part 4, licence agreements. Parsed for parties, dates, rent, deposit, break clauses, and any non-standard terms a previous agent slipped in.
- Landlord agreements and instructions — including the standing rules ("never give out my mobile", "approve repairs under €250 without asking", "don't accept pets").
- Compliance documents — BER certificates with expiry, gas safety where applicable, electrical certs, smoke and CO alarm records, RTB registration confirmations.
- Maintenance history — every contractor visit, every invoice, every "we already fixed that boiler twice" that you'd otherwise only remember if you were the agent who took the call.
- Communications — email threads, file notes from calls, WhatsApp exports if the agency uses them, all linked back to the unit and the tenancy.
- Deposit records — amounts, who's holding them, what was deducted last time, photos from check-in and check-out inspections.
The point of indexing is not search. The point is that when an agent asks "what's the deal with 17 Mill Road", the answer is a synthesised brief: current tenancy, rent, last inspection, outstanding maintenance, landlord preferences, last three communications, and any flags. That brief should land in under five seconds. If it doesn't, the agent goes back to the inbox and the system has failed.
The tenancy lifecycle, end to end
I think about the property-management brain as something that follows a tenancy from before it exists to after it ends. Each stage has its own work, and most of it is repetitive enough that an AI layer earns its keep without doing anything clever.
Pre-letting
When a landlord onboards a unit, the brain extracts the essentials from whatever paperwork they hand over — title docs, BER, previous tenancy if any, management company correspondence — and produces a structured profile. It also flags what's missing. You'd be surprised how often a landlord doesn't have the BER they swore they had.
Marketing and viewings
Drafting listings is a job a model is genuinely good at, provided it's working from your unit data and not the open internet. The brain produces a Daft-ready description, a shortlist of likely questions tenants will ask, and a viewing brief for whichever agent is showing the property. Nothing fancy — just the details the agent would have wanted ten minutes before the appointment instead of scrambling for them in the car.
Application and reference checks
This is where I'm careful. AI in tenancy AI contexts touches on equality law and the Residential Tenancies Acts. The brain doesn't make decisions on tenants. It assembles a reference pack — employer letter, previous landlord reference, ID verification, affordability calculation against stated rent — and presents it to the human agent. Decisions stay with the agent and the landlord. Anything else is a lawsuit waiting to happen.
In-tenancy
This is where the volume is. Maintenance triage, rent queries, neighbour complaints, end-of-fixed-term notices. The brain drafts responses in the agency's voice, pulls in the relevant tenancy clause when needed, and routes anything urgent — gas leak, water ingress, anti-social behaviour — to a human immediately. Drafted, not sent. The agent reviews and presses send. That distinction matters.
Renewal and end of tenancy
Notice periods under the Residential Tenancies Acts are unforgiving and have changed several times in recent years. The brain tracks the dates that matter, generates compliant notices for review, and handles deposit reconciliation against the inspection record. When a tenant moves out, it produces the closing pack: final statement, deposit return calculation with deductions referenced to specific inspection findings, and an updated unit profile ready for the next letting.
Compliance, RTB, and why on-prem matters
Irish letting is regulated, and the regulation is getting tighter, not looser. RTB registration, rent pressure zones, notice-of-termination rules, deposit handling under the forthcoming scheme, equality requirements, GDPR for tenant data, anti-money-laundering for certain transactions. None of this goes away because you bought software.
What on-prem AI does is shrink the compliance surface. Your tenant data — which includes PPS numbers in some workflows, financial information, sometimes medical information where reasonable accommodation is being discussed — never leaves your premises. There's no third-party processor agreement to renegotiate every time a US AI vendor changes its terms. There's no "we trained on your data by accident" headline. The model runs on a box you own, in an office you control, and the logs are yours.
For PSRA-licensed firms this also matters at audit time. When the regulator asks how a decision was reached, you can show the actual prompt, the actual documents the model saw, and the actual draft the agent then edited and sent. That's a defensible audit trail. A screenshot of ChatGPT is not.
Where letting AI helps and where it doesn't
I'm wary of the "AI does everything" pitch. In property management AI there are jobs the model is good at and jobs it shouldn't touch.
It's good at: summarising long email threads, drafting routine correspondence, extracting structured data from messy documents, finding the one clause in a tenancy agreement that answers the question, generating compliant notices for human review, producing landlord reports that previously took an evening per landlord, and flagging anomalies — a unit whose rent hasn't been reviewed in years, a BER about to expire, a deposit held by the wrong party.
It shouldn't touch: tenant selection decisions, dispute outcomes, anything that ends up in front of the RTB, valuations, or any communication where tone matters more than content. A bereaved tenant's family doesn't need an AI-drafted reply about a fixed-term tenancy. They need an agent on the phone.
The honest framing for irish property management firms is this: the brain takes the boring 70% off the agent's plate so the agent can do the 30% that actually requires judgement. If a vendor tells you the AI can do the 30%, walk out.
How this fits the wider Intelligence Brain
The property module isn't a standalone product in the way a CRM is. It's one face of a single on-premise system — the broader Intelligence Brain platform — that I've built to handle regulated information work across verticals. The same engine that reads a tenancy agreement reads a contract or a clinical letter in the other modules. The advantage for an agency is twofold: you're not buying into a niche tool that one vendor might quietly retire, and if your business grows into block management, sales, or commercial, the underlying system extends rather than gets replaced.
It also means the property brain benefits from work done elsewhere. Document extraction improvements made for legal work flow into tenancy parsing. Email-handling improvements made for accountants flow into agent inboxes. That's the architecture. One brain, several faces.
Where to start this week
If you run a letting agency and you're reading this, don't start with software. Start with one folder of one landlord's properties — pick the messiest one — and write down every question you had to answer about those units in the last month. How many of those questions could have been answered if a system had read every document and every email and remembered everything? That's your business case. Once you have it on paper, then talk to me, or talk to anyone serious about on-prem AI for regulated work. But have the case first. The technology is the easy part; knowing what you actually want it to do is the part most agencies skip and then regret.