Where the manual work hides in property management

Property management software is good at holding records and less good at helping anyone produce them. A lease sits in a system. An inspection is scheduled, then marked complete. Between those recorded events, a property manager spends the day on work the platform never sees: walking a property with a phone, typing up findings that were typed up in much the same words last time, ringing a tenant who has not replied, chasing a tradie who has not confirmed, and answering an owner who wants an update.

The work that never appears in the system

Where the week goes is rarely the part the software reports on. Portfolio size is recorded. Arrears are recorded. What is not measured anywhere is the time it takes to turn a set of photographs into a report, or the number of calls it takes to agree one access arrangement. That work is invisible because it is not a record. It is the labour of producing records.

It is also the work that scales worst. Add properties and the recorded events grow in a predictable way. The chasing, the transcribing and the re-explaining grow faster, because each one is a small negotiation with a busy human being.

Inspections are a writing job disguised as a walking job

A routine inspection looks like a physical task. Turn up, walk through, photograph what matters, leave. The walking is the short part. The long part happens afterwards at a desk, where photos are sorted, compared against the entry condition report, described in words, graded and assembled into something an owner will read and a tenant might contest.

Exit inspections are harder again, because the writing carries consequences. Every observation is potentially a bond deduction, and every bond deduction is potentially an argument. The same problems tend to repeat.

• The same defect described differently across reports, because different people wrote them.
• Photos captured in a burst with no note attached, then interpreted from memory a day later.
• Findings that are obvious standing in the room and ambiguous once they are a thumbnail in a PDF.
• Reports rebuilt from a blank page each cycle, when most of the property has not changed.

Maintenance arrives as unstructured messages

Almost nothing about a maintenance issue arrives in a form software can use. It arrives as a text message late at night, a voicemail, a photo with no context, or a sentence buried in an email about something else. Someone has to read it, work out which property it belongs to, decide whether it is urgent, decide whose cost it is, find a tradie who covers that trade and that suburb, get an approval from the owner, then keep the thread alive until the job is done.

Each step is small. None is difficult. Together they take up much of the day, which is why a property manager's inbox becomes a work queue nobody designed.

Disputes are argued from a photo nobody labelled

When a claim is contested, the case is only as good as the evidence trail. What usually exists is a folder of images with camera filenames, a report written in general terms, and someone's recollection of the property at handover. What is needed is a specific finding, tied to a room and an item, tied to the photo it came from, with a date and a name against it.

The difference between evidence and a photo album is whether somebody wrote down what they were looking at.

Where AI helps, and where it must not decide

There is a real use for computer vision here, and it is narrower than most property tech marketing suggests. A model looking at inspection photos is good at surfacing candidates: a mark on a wall, staining under a window, a cracked tile, a missing smoke alarm cover, something that looks like mould. It can point at an area, say this may be worth a finding, and hold the same attention at the end of a photo set as at the start.

What it must not do is decide. Whether a mark is fair wear and tear or damage is a judgement with money and a tenancy attached to it. That call belongs to the property manager, who accepts the suggestion, edits it, or overrides it. The workable design has the model propose and a person dispose, with the record showing which is which.

• Vision surfaces likely defects across a photo set. A person confirms each one before it becomes a finding.
• Every accepted finding carries the photo, the room, the item and the person who accepted it.
• Rejected suggestions stay in the history, so nothing quietly disappears.
• The report is assembled from confirmed findings instead of from a blank page.

It is the same principle behind the studio's estimating venture, where vision reads a plan and an operator confirms what it read before any deterministic calculation runs. Use the model for perception, and keep the decision with the person who will have to defend it.

Why the studio thinks this is worth building

There is no shortage of property software, and most of it holds records competently. The gap sits in the production of those records, which is still typing, chasing and remembering. That is where the hours go, and it gets harder rather than easier as a rent roll grows.

The Concept Realty is the studio's answer to that gap, and it is in development. The product work covers routine and exit inspections, AI-assisted defect detection with a human validating each finding, maintenance, reports and disputes. It is being built the way LoopSmiths builds everything it owns: run the operation, feel the manual work directly, then decide what the software should take off someone's plate.