Generative AI: PropTech

Generative AI in PropTech and Real Estate

AI that answers the portal enquiry at 10pm, drafts the listing from facts you hold, abstracts the lease and triages the leaking ceiling, without inventing features or making decisions about people that a human should make. We build agents grounded in your CRM, listings and property management data, with audit trails and human approval where it matters. Senior engineers own the design, AI coding agents carry the volume, and every line is reviewed by a human.

Get a free assessment PropTech software

Key highlights

The use cases that pay back first in property are instant lead response, listing content, lease abstraction and maintenance triage.
Agents answer from your CRM, listing data and property management system, so a reply never promises a feature or availability you do not have.
Tenancy, lending and screening decisions stay with people. AI drafts and organises; it does not decide who gets a home.
Listing copy is checked against material information and labelled where imagery is virtually staged.
A proof of concept usually takes about 4 weeks; a production agent wired into your CRM usually takes 10 to 16 weeks.

Why property AI projects stall, and what is worth solving

Property is document heavy and relationship driven. AI stalls when it cannot see the CRM or the lease, or when legal review finds it making decisions about people.

Enquiries that arrive after hours. Rightmove, Zoopla, Bayut and website leads land in the evening and at weekends. By the time a negotiator replies on Monday, the buyer has booked viewings with two other agents.

Listing copy that oversells. A generic generator adds a south facing garden or a recently fitted kitchen that the property does not have. Under UK material information rules that is a misleading listing, not a typo.

Leases nobody has time to read. Commercial leases, rent review clauses, break options and service charge schedules sit in PDFs. Abstraction is manual, so key dates are missed and portfolio reporting relies on memory.

Maintenance requests by WhatsApp photo. Tenants send a photo and a sentence. Someone has to decide whether it is damp, a leak or an emergency, find the right contractor and keep the tenant informed, often for hundreds of units.

Fairness risk that stops the project. Lead scoring or tenant screening that leans on postcode, name or language can act as a proxy for protected characteristics. Legal review rightly blocks it, and the pilot dies.

Data spread across CRM, PMS and portals. Contacts in the CRM, tenancies in the property management system, certificates in a shared drive. A chatbot trained on the website cannot answer anything that matters.

Use cases

Generative AI use cases we build for property

Each use case is grounded in your own systems, tested on a golden dataset of your listings, leases and messages, and keeps decisions about people with people.

Lead response and viewing booking agent

A tool using agent replies to portal and website enquiries in minutes on email, WhatsApp or chat, answers from listing data, qualifies the applicant, books viewings into the negotiator's diary and hands over warm leads with a summary in the CRM.

Listing content copilot

Drafts descriptions and portal copy from structured listing data, floorplans, EPC and room measurements. It flags missing material information such as tenure, council tax band or service charge, and the agent approves before publishing. Virtually staged images are labelled.

Lease abstraction

Document extraction turns commercial and residential leases into structured data: parties, term, rent reviews, break options, repair obligations and service charge terms, each with a citation to the clause. A surveyor or property lawyer confirms the abstract.

Maintenance triage agent

Classifies tenant requests from text and photos, sets urgency, asks follow up questions, raises the job in your property management system and dispatches the right contractor. Damp, mould and safety hazards are escalated to a person against the timescales your landlord obligations set.

Semantic property matching

Applicants describe what they want in plain language, and hybrid retrieval over listings, commute data and school catchments finds matches with an explanation. It works alongside your existing AVM rather than replacing it.

Portfolio and compliance copilot

Property managers ask about arrears, voids or certificates due to expire and get a text to SQL answer over the property management system with the query shown. Read only by default, with alerts for EICR, gas safety and lease events.

How we deliver

From one branch to the whole portfolio

01
Use case discovery and data readiness
We review your lead flow, listing data, lease library and maintenance tickets, and check API access to the CRM, portals and property management system. Output: a ranked use case list with the first one scoped and measured.
02
Evaluation harness and prototype
A golden dataset built from your own enquiries, listings and leases with acceptance criteria, including fairness checks, then a prototype tested against two or three candidate models.
03
Guarded pilot with human in the loop
Live on one branch or building. Listing copy and lease abstracts are approved before use, urgent maintenance goes straight to a person, and every agent message is logged against the contact or unit.
04
Production rollout with monitoring
Roll out by branch or portfolio behind a feature flag, with response time, accuracy and cost dashboards, drift alerts and quarterly model reviews.

AI governance for property

Property AI meets consumer protection, anti discrimination and data protection law the moment it talks to a buyer or tenant. We design for all three.

EU AI Act GDPR and UK GDPR UK Equality Act 2010 US Fair Housing Act FCRA (tenant screening) UK DMCC Act 2024 consumer protection AML and CDD (MLR 2017) UAE PDPL awareness

People decide on people. Lead scoring, applicant matching and screening are built without protected characteristics or obvious proxies for them, tested for disparate outcomes, and never make the final tenancy or lending decision. In the EU, AI used to assess the creditworthiness of individuals is a high risk use under the AI Act, and in the US the Fair Housing Act and FCRA apply to screening whatever tool is used.

A listing is still advertising. Generated descriptions must not mislead buyers or tenants under the UK unfair commercial practices rules in the DMCC Act 2024 and the material information guidance agents already follow. Copy is grounded in listing data, gaps are flagged instead of filled, and staged imagery is labelled.

Personal data stays controlled. Applicant and tenant data is redacted before prompts where the task allows, models run under enterprise terms that exclude training on your data, regional or self hosted models are available, and every agent action is logged with the contact or unit it touched.

Models and tooling we use for property AI

Portable language models, retrieval over your listings and documents, and integrations with the portals and systems agents already use.

Claude, GPT-5 and Gemini Llama 4 and Qwen (self hosted) LangGraph pgvector and Qdrant AWS Textract Rightmove and Zoopla feeds RESO Web API (US MLS) Bayut and Property Finder Twilio and WhatsApp Business API PostgreSQL React and React Native Langfuse

Related case study

Multi Tenant CRM SaaS for UK Real Estate Agencies

How we built a white label multi tenant CRM for UK agencies with AI lead scoring and Rightmove and Zoopla intake. Lead to viewing conversion rose 40%.

Read the full case study

FAQ

Common questions

How much does a generative AI project for property cost?

It depends on scope, so we quote from a written estimate rather than a price list. Most clients start with a fixed price proof of concept for one use case, such as lead response or lease abstraction, then move to a production build or a dedicated engineer. Model API costs are passed through at cost. Our pricing page lists current entry points.

Is applicant and tenant data safe with an LLM?

Yes, when it is designed for. We use enterprise model APIs that do not train on your data, redact personal data before prompts where possible, offer UK, EU or UAE hosting or self hosted models, and log every access. Agents get the minimum CRM permissions they need.

How do you stop the AI inventing property details?

By grounding it in your data and checking the output. Listing copy can only state facts present in the listing record, answers to enquiries come from the CRM and listing data, and lease abstracts cite the clause they came from. A golden dataset runs on every change, and agents approve copy before it is published.

How long does a first property AI project take?

Usually about 4 weeks for a proof of concept and 10 to 16 weeks for a production agent. Lead response and listing copy tend to be fastest; maintenance triage takes longer because it touches contractors and landlord obligations.

Should we buy an AI tool for agents or build our own?

Buy where the task is generic and build where your data and process are the advantage. CRM vendors increasingly offer AI features that suit standard workflows. When you need your own lettings rules, multiple portals, a property management system and an audit trail working together, a custom build on portable models usually fits better.

Can AI decide which tenant gets a property?

It should not. We use AI to organise applications, check completeness and summarise documents, and leave the decision with a person. That keeps you on the right side of equality and fair housing law and makes every decision explainable.

Keep exploring

Related services and industries

Free consultation, no commitment

Ready to answer every enquiry in minutes?

Tell us about your lead flow, listings or lease library. Written scope, timeline and cost estimate within 48 hours.

Book a free consultation All industries