How AI in commercial property is actually used

A pitch document is due Friday. The data exists across three spreadsheets, a planning portal, and last quarter’s report. Someone has to pull it together, format it consistently, write the commentary, export to PDF, and send it with the right cover note. The analysis takes an hour. The rest takes until Thursday afternoon. That shape repeats across every specialist business I work with: surveyors, developers, planning consultants, contractors. The constraint isn’t insight, it’s the mechanics around it. This is where AI in commercial property businesses actually gets used, and it’s not the bit that makes headlines.

What does AI in commercial property actually automate?

It automates the repetitive assembly work that sits before and after the judgement. Pulling data into a standard format. Drafting the commentary that frames the numbers. Generating the comparison table. Updating the document when a figure changes. Sending the follow-up email three days later if no one replies. The analysis itself still needs a person who knows what matters.

A McKinsey study on generative AI adoption in professional services found that document preparation and data formatting account for more automation hours than analytical tasks. The insight work stays human. The scaffolding around it doesn’t need to be.

Where does the human judgement stay?

Anywhere a decision has consequences that a system can’t evaluate. Which comparables matter for this particular site. Whether the tone of a planning objection response should soften or sharpen. What gets included in a board report and what gets held back. Whether a tender needs another day or goes out now. The systems I build for AI in commercial property clients stop at the boundary of judgement and hand back control.

This isn’t a principle, it’s practical. A system that drafts a valuation commentary can use last quarter’s structure and pull the updated figures. It can’t decide whether the narrative emphasis should shift because the market mood has changed. That still wants a surveyor.

What kind of tasks get automated first?

The ones that follow a template but require manual effort every time. Weekly reports that pull the same data fields into the same structure. Pitch documents where the format is fixed but the content changes. Follow-up emails that go out if a response hasn’t arrived within a set window. Marketing updates that draw from a project list and publish to a schedule. Anything that makes someone think, “I’ve done this exact task a dozen times, why am I still doing it by hand?”

A developer I work with had an associate spending half a day each week updating a project status document for the board. Same template, same sources, same export routine. The system now does it overnight. The associate reviews it in ten minutes and sends it. The board gets the same document, on time, and the associate has the afternoon back.

Does AI in commercial property replace the team?

No. It changes what the team spends time on. A surveyor who used to spend Tuesday compiling a rent roll can now spend Tuesday on the site visit that actually needs their experience. A planning consultant who used to spend Thursday formatting a committee report can spend Thursday refining the argument. The output still carries their name and their judgement. The mechanical assembly work doesn’t.

This framing matters because the fear is usually replacement. The reality is reallocation. Teams stay the same size but get more done, or stay the same size when they would otherwise have needed to grow. According to research from Stanford HAI on AI in professional work, the productivity gain shows up as capacity, not headcount reduction.

What does AI automation actually cost?

Less than hiring someone to do the task manually, more than ignoring the problem. A working system that automates a weekly report might cost the same as two months of an associate’s time. If the report takes half a day each week, the system pays for itself in a year and runs indefinitely after that. The cost sits in building it properly, not licensing software month to month. I build systems, hand them over, and they belong to the client.

The savings aren’t always financial. Sometimes it’s hitting a deadline that would otherwise slip. Sometimes it’s freeing up the one person who knows how to do the task so they can do something else. Sometimes it’s consistency: the document looks the same every time because the same system produces it.

Where does AI automation go wrong?

When it’s built around the technology instead of the task. A system that requires someone to learn a new interface, remember to trigger it, or reformat their data to fit it won’t get used. The best systems sit inside the tools the business already uses and run without anyone needing to remember they exist. A planning consultant shouldn’t have to open a separate dashboard to publish a project update. It should draft itself from the project tracker they already maintain and publish when it’s due.

The other failure mode is overreach. A system that tries to make decisions instead of preparing options will either get ignored or cause problems. The line between automating preparation and automating judgement is where most implementations go wrong. I keep systems on the preparation side and hand control back before the decision point.

How long does it take to build a working system?

For a single repeating task, a few weeks from brief to handover. For something more involved, like a content engine that writes and publishes updates across multiple projects, a few months. The timeline depends on how much existing structure there is to build around. A business that already uses a project tracker and a standard document template has a faster route than one starting from scratch.

Speed matters because the task is still happening manually while the system gets built. I prefer to ship a simple version that handles eighty percent of the work, hand it over, and refine it in use rather than spending months building something perfect that launches all at once.

What does a handed-over system actually mean?

It means the business owns it and can adjust it without calling me every time something changes. The system runs on their accounts, using their data, publishing to their channels. If a template needs updating or a new field needs adding, someone internal can do it. I’m not licensing access, I’m building and handing over the tools.

This matters for AI in commercial property businesses because the structure changes. A developer switches CRM. A surveyor adds a new service line. A planning consultant picks up a project type they haven’t handled before. A system that requires my involvement for every adjustment becomes a dependency. A system that’s theirs doesn’t.

You can read more about how I build AI automation systems or see other posts on AI and design in the journal.

If you’re spending half a week on tasks that follow the same pattern every time, send a paragraph describing what happens and where the time goes. I’ll write back with what a working system might look like and what it would cost to build.