The question most operators are asking about AI is "how much will it change property operations," and the answers come back as percentages that mean nothing from the COO's chair. Thirty percent of hours. Half of all tasks. A number that large is not actionable, because "property operations" is not one kind of work. It is three, and AI does something entirely different to each of them. The useful question is not how much. It is which. Which of the things your people do all day does AI actually touch, and which does it only appear to.
Sort the work into three layers and the picture stops being a headline and starts being a plan.
The physical layer, where AI does the least
Start with the work that is actually the point of the business: the property itself. The leak at 11pm. The unit turn between tenants. The fire inspection. The snow. The tenant standing at the leasing office door. This is physical, local, and specific, and it is the layer where AI changes the least, no matter what the demo implied.
This is worth saying plainly, because the honest version builds more trust than the hype. Sensors and predictive maintenance tools are real and useful, but notice what they actually do: they flag and they inform. A model can tell you the compressor is trending toward failure. It cannot replace the compressor. The wrench work of keeping buildings running stays human, stays on site, and stays roughly as labor-intensive as it was.
McKinsey's Global Institute put a number to this in late 2025. It estimated that currently demonstrated technologies could, in theory, automate activities accounting for about 57 percent of US work hours. The split inside that number is the part a COO should sit with: software agents account for 44 of those points, physical robots for only 13. The automatable share of work is overwhelmingly non-physical. If a vendor's pitch implies AI is about to transform your field operations, that is the claim to be most skeptical of, because it runs against where the technology is actually strong.
The coordination layer, where the real change lives
Now the layer nobody puts on a brochure, because it is unglamorous: the connective tissue. The work of moving information between people and systems. Assembling the owner statement from four places. Drafting the late notice. Triaging the maintenance queue by urgency. Chasing the missing W-9. Reconciling the deposit. Summarizing the month for an owner. Answering, for the four-hundredth time, the tenant question you have already answered in writing.
This is where AI is genuinely good, and it is a large share of where your administrative and back-office hours actually go. An earlier McKinsey estimate put the work activities that generative AI and related tools could automate at 60 to 70 percent of the time employees spend today, with the effect concentrated on knowledge and language work rather than manual work. Property operations runs on exactly that kind of work: reading, drafting, routing, summarizing, and reconciling text and numbers. The connective tissue is made of language, and language is what these tools handle.
The honest framing of the change matters here. This is not "AI decides." It is closer to "AI drafts, assembles, triages, and routes, and a person approves." The leverage is real, and it shows up as time given back to people who currently spend their days as human middleware between systems that do not talk to each other. That is the actual near-term story of AI in property operations, and it is quieter and more valuable than the version being sold.
The judgment layer, where a person still has to own it
The third layer is the set of calls that carry consequences. Whether to renew a tenant who pays late but always pays. How to handle the owner who is quietly shopping for a new manager. When to escalate a dispute before it becomes a lawsuit. What a budget variance actually means for this specific property, given what you know that is not in any system.
AI can inform every one of these. It can surface the payment history, draft the options, model the outcomes. What it should not do is own the decision, and a COO who lets it own the decision is not saving time, they are acquiring a liability with no one accountable for it. McKinsey is careful to note that its automation figures describe technical potential, not a forecast, and that AI is more likely to change how skills are used than to make them obsolete. The judgment layer is where that distinction bites. This is the work your operation is actually paid for, and it is the layer to protect from automation rather than hand to it.
Why most operators do not get the gains that are real
Notice that the one layer where AI clearly helps, coordination, is also the one that depends entirely on something AI does not provide: connected data. An assistant cannot assemble an owner statement from four disconnected systems any faster than a person can, because it faces the same problem the person does, which is that the four systems disagree and no one record is authoritative. The leverage in the coordination layer is real, but it is gated behind whether the operational and financial data live in one place or in a dozen. Where they are scattered, the AI has nothing coherent to coordinate, and the pilot quietly underdelivers.
The gains are also not free, which is worth conceding rather than glossing. Gartner expects the cost of governing AI to rise far enough that by 2028 it offsets roughly 60 percent of the savings that agentic AI generates. The middle-layer automation is worth pursuing, but it comes with an oversight cost that belongs in the business case from the start, not as a surprise later.
Put those together and the COO's real move is not "adopt AI." It is to consolidate the coordination layer onto a single record first, so that when AI is applied to it, there is something solid underneath. This is the case for running operations and financials on the same system rather than syncing between separate ones. RIOO is built directly on NetSuite for that reason: when the work order, the ledger entry, and the owner record are one record, the coordination layer AI is good at finally has coherent material to work with.
The sorting is the strategy
So stop asking how much AI will change property operations, and start asking, of any specific claim a vendor makes, which layer it touches. Physical work: very little, and be wary of anyone who says otherwise. Coordination work: a great deal, but only if your data is connected enough to give it something to do. Judgment work: nothing you should be willing to hand over. That sorting, applied claim by claim, is the entire strategy. It turns an unanswerable question about the future into a set of answerable questions about your own operation, which is the only version a COO can actually act on.
FAQs
Q1. Will AI reduce my headcount in property operations?
It is more likely to change what your existing people spend time on than to cut the count outright, at least in the near term. The hours it frees are concentrated in administrative and coordination work, so the realistic outcome is people spending less time as middleware between systems and more time on judgment and resident-facing work. Treating it purely as a headcount cut usually means missing where the actual value is.
Q2. Which property operations tasks should I automate first?
Start where the work is repetitive, language-based, and low-stakes if a human checks the output: drafting routine notices, summarizing owner activity, triaging and routing maintenance requests, first-pass reconciliation. These sit squarely in the coordination layer and keep a person in the approval seat, which is where early wins are both real and safe.
Q3. Can AI handle maintenance and field operations?
Only at the edges. It can predict likely failures, prioritize a queue, and schedule more intelligently, but the physical work stays human and on site. Expect help with deciding what to do and when, not with doing the physical work itself.
Q4. Why do our AI results depend so much on our systems?
Because the layer AI is best at, coordination, is only as good as the data it coordinates. If the operational and financial records live in separate, disagreeing systems, an AI assistant inherits that disagreement and cannot resolve it. Connected data is the precondition, not a nice-to-have.
Q5. Is it safe to let AI make decisions about tenants or owners?
Let it inform those decisions, not own them. Renewals, disputes, and owner relationships carry consequences that require an accountable human. Using AI to assemble the facts and draft the options is sound. Using it to make the call removes accountability without removing the risk.
Q6. How do I evaluate an AI vendor's claims?
Sort each claim into the three layers. If a vendor promises transformation of physical operations, be skeptical. If they promise help with coordination and administrative work, ask what data it needs and whether your systems can supply it. If they imply the tool will make judgment calls on its own, treat that as a risk to manage, not a feature to buy.
Q7. What is the real cost of adopting AI in operations?
More than the license. Oversight and governance carry a genuine ongoing cost, enough that Gartner expects it to offset a large share of the savings from agentic AI by 2028. A sound business case includes the cost of supervising the tool, not just the price of the tool.
Q8. Does this mean AI is overhyped for property management?
No, it means the hype points at the wrong layer. The value is real but concentrated in coordination and administrative work, which is less exciting than "AI runs your buildings" and considerably more useful. The overhyped part is the physical and judgment layers. The underappreciated part is how much back-office time the middle layer quietly consumes today.
Q9. Where should a COO start if the systems are not connected yet?
With the data before the AI. Consolidating the operational and financial record onto one system is the step that makes everything downstream possible, and it delivers value on its own even before any AI is applied. Adopting AI on top of fragmented data tends to produce impressive demos and disappointing pilots.
.svg)
.svg)
.svg)
