AI for Real Estate Agencies: Three Systems We've Built and What Each Solved

Real estate is the sector where we have built the widest variety of systems, because agencies differ enormously in where their time goes. Three engagements, three completely different problems.

Client details are omitted throughout — the work belongs to them.

One: the content pipeline

The problem. An agency was producing listing content by hand. Photos came back from the photographer, someone assembled them into a video, wrote the caption, and posted it. Days of delay per listing, and by the weekend the listing had missed the window when buyers actually browse.

What we built. A pipeline that takes uploaded photos and produces short-form video automatically — sequencing, transitions, text overlays pulled from the listing data — then posts to their social channels on a schedule.

Why it works here. Listing content is genuinely templated. The same shape works for almost every property: exterior, key rooms, a feature, a call to action. That is the profile of work that automates well — high frequency, explicit rules, and a quality bar that is consistency rather than artistry.

What it does not do. It does not write the listing copy that matters, and it does not replace a photographer. The photos still have to be good. Automation multiplies the input quality in both directions.

Two: the agentic email and call handler

The problem. A firm was drowning in first-contact volume. Enquiries arriving by email and phone, most asking the same fifteen questions, each one needing a response fast enough to compete with the other agents who received the same enquiry.

What we built. An agentic system handling both channels — reading incoming email, answering calls, and responding with the firm's actual answers to their actual questions.

The important design decision was where the answers came from. The firm supplied their own responses to the common questions. The agent retrieves and delivers those; it does not compose new ones. That constraint is what makes it safe to run unsupervised on customer contact.

What it escalates. Anything about price negotiation, anything from an existing vendor, anything where the caller is unhappy, and anything outside the defined question set. Those reach a person with the context attached.

Why it works here. First contact in real estate is speed-sensitive and largely predictable. The firm's advantage was never their ability to answer "when is the next inspection" — it was what happened after. Automating the first exchange bought them the second one.

Three: the document and reporting system

The problem. Property reports. Financial documents and property information arriving in every format imaginable, needing to be checked, organised, labelled correctly, and assembled into a report. The work was mechanical, high-volume, and consumed a full working day per report.

What we built. An agent that processes incoming documents, extracts the relevant figures and details, files them against a consistent labelling and documentation standard, and generates the property report from the assembled data.

The result: roughly six hours saved per report. Not the whole day — a person still reviews the output and handles the exceptions — but the assembly work, which was the bulk of it.

The privacy dimension. This system handles financial and personal information, so the constraints were set before the architecture. Data collected only where the report genuinely needed it, access restricted, retention defined rather than indefinite, and encryption throughout. That was a design requirement rather than a compliance review afterwards, which is the only way it works without a rebuild.

The pattern across all three

Notice what these have in common. In each case the agent handles the mechanical majority and a person handles the exceptions and the judgement.

None of them replaced anyone. What each did was remove the work that was being done at 9pm because there was no time for it during the day.

Where we have advised against automation in real estate

Appraisal conversations. The appraisal wins the listing and it is relationship work. We have talked agencies out of automating this more than once.

Vendor communication during a difficult sale. When a campaign is not going well, the vendor needs a person. An automated update at that moment reads as avoidance.

Anything touching trust accounting. Regulated, audited, and not a sensible place to experiment.

Working out which of the three applies to you

Losing enquiries to slow first response — the email and call handler addresses that directly, and it is usually the fastest payback.

Listings going up slowly — the content pipeline, though only if your photography is already good.

Someone spending a day a week assembling documents — the reporting system, and this is the one with the clearest hours-saved arithmetic.

Most agencies have one of these problems considerably worse than the other two. Start there rather than attempting all three.