Morgan Stanley Says 37% of Real Estate Work Can Be Automated. Start With the Maintenance Queue.

Morgan Stanley analyzed 162 real estate investment trusts and commercial real estate firms, companies with a combined $92 billion in labor costs and 525,000 employees, and concluded that 37% of the work they do is automatable. According to Morgan Stanley Research, that translates to as much as $34 billion in efficiency gains over the next five years.

Most operators read 37% as a someday number, a horizon figure to revisit when budgets and tooling catch up. That is the wrong way to read it. The number is not a prediction about the industry. It is a question about your own sequence. Out of everything your team touches, which tasks move first, and why those instead of the others. The firms pulling ahead have already answered it, and they did not start with the glamorous parts of the business. They started with the queue nobody wants to talk about in a strategy deck.

The 37% Is Not Spread Evenly

Morgan Stanley did not find that 37% of every job is automatable. The automation potential clusters in four task categories: management, sales and related activities, office and administrative support, and installation, maintenance and repairs. Those are the areas where the work is repetitive, the decision rules are already written down, and the volume is high enough that automation pays back fast.

Maintenance and repair intake sits at the intersection of all three conditions. A tenant reports a leak. Someone has to read the message, work out how urgent it is, figure out which unit and which system, check whether a photo or an access code is missing, decide who to dispatch, and follow up after the job. Every one of those steps follows rules a property manager could write on a single page. That is exactly the profile of work that automates first, and it is why sequencing matters more than ambition. You do not start where AI is most impressive. You start where the decision rules already exist and the ticket volume is highest.

Adoption Is Already Splitting the Field

This is not a theoretical exercise that operators can defer. The gap between firms that have adopted AI and firms that have not is already showing up in growth numbers. The AppFolio 2026 Property Manager Benchmark Survey found that firms with broad AI adoption expect average portfolio growth of 31% in 2026, nearly triple the 12% anticipated by firms that have not implemented the technology. In the prior year's benchmark, AI usage among property managers jumped from 21% to 34% in a single year.

The instinct when you see those numbers is to adopt broadly and fast, to switch on AI everywhere at once. That instinct is what produces stalled pilots. The firms reporting real portfolio growth did not spread a thin layer of AI across every function. They took one high-volume flow, gave it depth, and made it production grade before touching the next one. Maintenance intake is the natural first flow because it is the one where a slow response costs you a tenant.

Why the Maintenance Queue, Specifically

Two things make maintenance the right place to start. The first is that the cost of doing it slowly is measurable and direct. Properties that automated maintenance intake report response times dropping from days to hours, and tenant satisfaction climbing as a result, according to industry data compiled by property maintenance researchers. A turnover triggered by a frustrated tenant costs well over a thousand dollars per unit. Slow maintenance response is one of the cheapest ways to lose a renewal, and one of the easiest to fix.

The second is that the payback on the operational side is just as concrete. McKinsey research cited across the maintenance sector shows that predictive and well-managed maintenance cuts overall maintenance costs by 18 to 25% and reduces unplanned downtime by up to half. Those gains do not come from a chatbot that answers tenants politely. They come from intake that captures the right information the first time, triage that scores urgency correctly, and dispatch that matches the job to capacity. The work is in the structure behind the conversation, not the conversation itself.

This is the distinction most property-management AI misses. Answering the tenant is roughly 10% of the ticket. The other 90% is creating the work order, scheduling it against contractor capacity, chasing the vendor, and pushing status back to the tenant. An agent that only answers has automated the easy tenth and left the expensive nine tenths untouched. Resolution, not response, is where the cost actually lives.

What Production Grade Actually Requires

The reason most maintenance automation stalls is that a demo handles the clean ticket and reality sends messy ones. A real maintenance queue arrives across WhatsApp, email, a web form, and a phone call transcribed after hours. It arrives in more than one language when your tenants do. It arrives without the unit number, without a photo of the damage, and without the access instructions the contractor will need. A production-grade intake layer has to absorb all of that, normalize it into one structured ticket, and notice what is missing before a human ever opens it.

That is where the operational building blocks earn their place. Multi-channel intake pulls the request in regardless of how the tenant sent it. Classification and urgency scoring separate the burst pipe from the squeaky hinge. Automatic elicitation asks the tenant for the missing photo, the location, or the access code, in their own language, before the ticket reaches dispatch. Then the agent acts inside the property manager's own system, creating the work order, scheduling it against contractor capacity, pushing status updates, and chasing the follow-up, with a human deciding on the genuine exceptions and an audit trail on every step. We have built exactly this for operators running 200 properties with multilingual maintenance flows across Dutch, English, Polish and Romanian, and the lesson is consistent: the value is in the integration depth, not the chat window on top of it.

Sequencing Beats Ambition

The reason maintenance goes first is not that it is the most valuable flow in real estate. Leasing and renewals carry more revenue. The reason is that maintenance is where the decision rules are most complete and the volume is highest, which makes it the flow most likely to reach production rather than die as a demo. Once intake, triage, dispatch and follow-up run reliably on maintenance tickets, the same machinery extends outward. The classification logic, the missing-information elicitation, the integration into your domain system, and the audit trail all transfer to leasing inquiries and renewal workflows with far less effort than building them from scratch.

That is the practical reading of Morgan Stanley's 37%. It is not an instruction to automate a third of your operation this year. It is a map of where the automatable work concentrates, and maintenance intake is the densest, most rule-bound corner of it. Start there, get it genuinely production grade with a human deciding on the exceptions, and you build the foundation that makes the next flow cheaper. Start with the most impressive use case instead, and you are likely to join the pilots that never ship.

The 37% will not automate itself, and it will not automate in the order that looks best in a board deck. It automates in the order the decision rules allow. For a property operator drowning in maintenance tickets, that order starts with the queue already sitting in the inbox.