The constraint on AI in real estate is not capability. It is that nobody wants to adopt anything — and the firms winning right now are the ones who stopped asking.
Two conversations, two weeks apart, told me the same thing.
The first was with a founder in London running an AI platform for real estate. Good product. Real customers. His problem was not signups. It was that clients who had asked for the software, bought it, and been trained on it were quietly canceling a subscription of roughly $180 a month. Not because it was too expensive. Because they never got around to using it.
The second conversation was with my own team, about something we did not plan. After we relaunched our platform in March, a subset of clients stopped using it the way we designed. They started sending spreadsheets and property lists with a single instruction: handle it. Nobody offered them that. We were not selling it. They asked anyway, and they were willing to pay considerably more for it than they were paying for access to the software.
Two companies, two continents, two different corners of real estate, same finding. The constraint on adoption is not price, and it is not capability. It is that busy professionals will not adopt anything, and the firms getting traction are the ones who stopped requiring it.
The real price of software is not the subscription
Every software purchase has two prices. The one on the invoice, and the one nobody puts in the proposal: the hours it takes to learn the thing, the onboarding sessions, the data migration, the habit change, the internal champion who has to nag everyone for six weeks until it sticks.
Call it the adoption tax. For most of the last 25 years, we all agreed to pay it, because the alternative was worse and there was no third option.
Do the arithmetic on a working professional. An agent’s time runs around $182 an hour. A coordinator at an engineering firm is cheaper, but that person is already underwater. Now ask either of them to spend six hours learning a platform, plus the slower weeks while the new way is still slower than the old way. The tax on that $180 subscription is not $180. It is several thousand dollars of billable attention, paid up front, for a benefit that arrives later.
That is why the cancellation isn’t a pricing problem, and dropping the price won’t fix it. The client is not comparing $180 to the value of the software. They are comparing the cost of changing how they work to the cost of continuing exactly as they are. Continuing is free today and expensive later, which is a trade almost everyone takes.
And it explains something that looks irrational on a spreadsheet: the same firm that cancels a $180 tool will happily pay four figures to have somebody else just do the work. They are not being inconsistent. They are pricing their own attention correctly.
Conversational interfaces were a half-step, not the fix
I have spent much of the last year arguing that the fix was conversation. Stop making people learn a dashboard. Let them type or say what they need in plain language, the way they would tell a coordinator. No training, no tutorial, no login screen followed by a tour nobody finishes.
I still believe that is a real improvement. It is not the whole answer, and our own clients showed us why.
Given a chat window that required nothing but a sentence, a meaningful share of them skipped it. They did not want to type the sentence. They wanted to forward the email they already had, or attach the spreadsheet already sitting on their desktop, and go back to work.
An interface you do not have to learn is still an interface you have to open. That is the part I had wrong. Removing the learning curve reduces the adoption tax. It does not eliminate it, because the residual cost is not comprehension. It is attention — the act of stopping what you are doing, going somewhere else, and thinking about a task you would rather hand off.
In April I wrote The Work Disappears, arguing that real estate was heading toward a model where firms pay for completed outcomes instead of software seats. The direction was right. What I did not anticipate was how the shift would show up: not as clients adopting a better interface, but as clients declining to use any interface at all and handing the work over instead.
What it looks like when the client stops using your software
Here is the old sequence for scheduling and confirming a single inspection. Fourteen emails. Six phone calls. Three days. Most of the industry still runs this way — by our count, about 94% of due diligence coordination still happens over email, phone, and spreadsheets.
Here is the new one. The client sends a list of forty sites. They specify what they need at each one. Then they do nothing, and completed work comes back.
Notice what is missing. No login. No dashboard. No training. No status-check meeting. The client’s total time investment is the time it took to send the file they already had.
What they are buying in that transaction is not software and not labor. It is a completed outcome with a name on it. The infrastructure doing the work is invisible to them, and that is the feature, not a compromise. The industry has started calling this service-as-software: the buyer pays for the result rather than for the right to log in. Deloitte’s 2026 technology predictions make the same call at the category level, arguing that seat-based licensing gives way to usage and outcome pricing as agents absorb the underlying work.
The language is new. The buyer behavior is not. Professionals have always preferred to hand work to someone competent. What changed is that the economics of doing that at scale finally work.
This only works if the middle is built before the front door opens
Now the part most people writing about this leave out.
Telling clients to send you anything is easy. Being able to absorb whatever arrives is not. The gap between those two is where this model kills firms that rush it.
When you open the door that wide, you get requests you should not accept. An asbestos survey priced below what any qualified vendor will take. A property in a market where you have no reliable coverage. A turnaround the client assumes is possible but isn’t. If the system says yes to all of it, each one becomes a loss, a broken promise, or both.
So the order of operations matters, and it is counterintuitive: build the middle before you make the front easy. Before you can safely tell a client “just send it over,” you need the ability to price incoming work automatically, reject what is unprofitable or unfeasible, route what remains to vendors who will actually perform, and hold accountability for the outcome when something goes sideways.
Most firms attempting a managed model today have not built that. They have built a promise, and they are absorbing the gap with quiet human effort. That works at ten projects a month. It breaks at a hundred, and it breaks in the worst possible way, by failing the clients who trusted the promise first.
If you take one operational point from this piece, take that one. The interface is the cheap part. The judgment layer behind it is the business.
This is not a services business wearing a software costume
A fair objection, and the one a sharp reader raises immediately: what you are describing is a service company. Service companies add a person for every increment of volume. That is a worse business than software, not a better one.
It would be, if the volume worked that way. The distinction is what happens to headcount when demand doubles. In a traditional coordination business, you hire. In this one, you don’t, because the platform lets a single coordinator handle roughly ten times the volume a person could manage before. The client experiences a service. The economics behave like infrastructure.
A second effect matters more over time. Every project a human handles inside that system is a recorded decision — which vendor got selected and why, how an unusual scope was priced, what the sequence was when something fell through. That is training data you cannot buy, produced by revenue-generating work rather than by a research budget. Manual work today makes automated work possible tomorrow, which means the model gets lighter as it scales rather than heavier.
That is the test for anyone evaluating a firm making this pitch, including the one you might be building: does the tenth client cost less to serve than the first, or more? If more, it is a staffing company with a chatbot. If less, something else is happening.
Three questions to ask about your own firm this week
Skip the strategy offsite. Ask these.
Which clients bought your software and stopped logging in? Pull the usage data. The accounts that went quiet aren’t unhappy, and they aren’t gone. They are telling you the adoption tax exceeded the value, and they are your first candidates for a managed handoff. Call three of them and ask what they would pay to never open the tool again.
How much of your client-facing time moves information rather than applies judgment? Walk one workflow end to end and mark each step: is a human here because expertise is required, or because someone has to carry a message from one place to another? The second category is the entire coordination middle of this industry, and it is the part that is going away.
What would you need to guarantee before you could tell a client “just send it to us”? Write the list. Pricing rules, coverage, vendor accountability, the authority to decline bad work. That list is your actual roadmap, and it is a more useful document than any product plan built around features.
The shift underneath the shift
The industry conversation about all this is mostly about pricing — seats versus usage versus outcomes. That is the surface, and it is the least interesting part.
The bigger change is that the entire software model of the last 25 years rested on an assumption: that the customer would meet the product halfway. That they would learn it, log into it, and change how they work to accommodate it. For a long time that assumption held, because there was no alternative worth choosing.
There is now. And once a professional has sent a spreadsheet and gotten finished work back, no interface improvement wins that client back. You are not competing against other software anymore. You are competing against the client doing nothing at all, which is the strongest competitor there is.
The firms that understand this early will look, from the outside, like they got lucky with a service line nobody planned. They did not. They stopped asking clients to adopt anything.
· · ·
Blue222 is an AI-driven orchestration platform coordinating inspections, appraisals, vendors, and property services across the real estate transaction lifecycle. If your team is still running due diligence through email and spreadsheets, I am happy to talk through what the handoff model looks like for your operation. alang@blue222.com, Alan Grosheider, Founder and Chairman, Blue222.
· · ·
Frequently asked questions
What is service-as-software? A model where the buyer pays for completed work rather than for access to a tool. Instead of a subscription that lets your team do a task faster, you send the requirement and receive the finished outcome. The software still exists — it is just on the provider’s side of the transaction, not yours.
Does this mean real estate software is dead? No. Dashboards, reporting, and systems of record matter as much as ever, and they are how a serious operation tracks a portfolio. What is changing is which side of the relationship operates them. The back office still runs on software. The front door is becoming a handoff.
How is this different from just outsourcing? Outsourcing moves the labor to someone cheaper. Headcount still scales with volume, and quality varies depending on who’s assigned. In a platform-backed model, the volume scales against infrastructure instead of people, and the process is consistent because it is systematized rather than delegated.
What margins does a managed model actually hold? Early projects often run higher than the long-run number because direct costs are minimal and pricing reflects the value of the outcome rather than the hours. As volume grows and labor requirements become clearer, expect compression. Model the conservative case and treat anything above it as upside, not the plan.
How do you price completed work instead of seats? Per project, per outcome, or as a blended managed-service rate against a defined scope. The requirement is that you can predict your own delivery cost before you quote. Firms that cannot do that should not offer outcome pricing yet.
What has to exist before a firm can offer this? Automated scope pricing, the ability to decline unprofitable or unfeasible requests, verified vendor coverage in the markets you accept, and a clear accountability path when something fails. Without those four, you are selling a promise your operations cannot keep.
