Nobody sells the thing that makes you different

Realcomm 2026 produced a consensus worth taking seriously. Two of the five takeaways out of the conference were these. AI is only as good as the data behind it, and the organizations getting results "aren't chasing the newest AI models, they're investing in clean, connected, and trustworthy data." Then, on build versus buy: "commercially proven technology often delivers lower long-term costs than custom-built applications." (PredictAP)

Deloitte's 2026 Commercial Real Estate Outlook, fielded to more than 850 C-suite executives and their direct reports, arrives at the same intersection from a different road. Seventy-three percent of CRE firms now see AI as crucial for advanced analytics or market signal detection. Twenty-seven percent still report challenges rolling it out: technical complexity, no in-house expertise, or people who will not change how they work. (Deloitte)

Read separately, those are two pieces of sensible advice. Read together, they do not fit.

The constraint is named as your data. The remedy offered is somebody else's software.

The part nobody says out loud

The buy-versus-build debate in CRE is conducted almost entirely in the language of cost. Total cost of ownership, maintenance burden, vendor roadmap, who fixes it at 2am. Those are real concerns and the conference answer is usually correct. Commercially proven software is cheaper over five years than the equivalent thing you write yourself.

It is also, by construction, the thing your competitor can buy on Tuesday.

That is not an argument against buying. It is an argument that cost is the wrong axis for one specific slice of what a firm runs on. There are two different questions hiding inside build versus buy, and the industry keeps answering the second one with the metrics from the first.

The first question is: what is the cheapest way to get a capability every firm in my asset class already has? Buy it. Every time. Lease administration, general ledger, document storage, e-signature. Nobody has ever won a deal because their rent roll lived in better software.

The second question is: what is the cheapest way to get a capability no competitor has? That question has no answer, because cheapest is not a property the thing possesses yet. Nobody sells it. That is what makes it worth having.

What the money is actually rewarding

Four proptech companies have crossed a billion dollars in valuation since mid-2024, and all four are AI-native. EliseAI at $2.2B. Bedrock Robotics at $1.75B. Vantaca at $1.25B. Juniper Square at $1.1B. (Bisnow)

Look at what they do rather than what they are worth. EliseAI answers the resident, schedules the tour, and audits the lease. Bedrock Robotics runs the equipment on the site. Juniper Square does the fund administration. Vantaca handles the HOA's documents and its residents. Not one of them sells a panel that summarizes something for you.

A CRETI employee was quoted saying it without decoration: "The market no longer rewards technology that 'helps.'"

Yardi's Rob Teel, at Realcomm, described the goal as software that gives employees capacity back by automating repetitive work. Same idea from the incumbent side. Dealpath shipped Dealpath AI in May, moving from chat into the daily work of investment teams. The direction of travel is not subtle and it is not in dispute.

Software that does the work is winning. Software that assists with the work is being repriced.

But isn't custom software usually a money pit?

Often, yes. That reputation was earned honestly, and most of the firms that earned it made the same mistake: they custom-built the commodity. A bespoke general ledger. An in-house CRM. A homegrown lease abstraction tool that one person understood and then left.

Sometimes the money pit is just the cost of a bad decision. But it is more often the cost of building on the wrong side of the line, pouring eighteen months into something a vendor already sells for four figures a year, while the actual differentiator stays in a spreadsheet on a principal's laptop.

The failure was not that it was custom. It was that it was custom and generic at the same time.

What we would do if we ran a CRE firm right now

Draw the line first, on paper. Two columns. Left: capabilities every firm in your asset class has. Right: how you actually underwrite, what you know about your submarket that CoStar does not, your comp history, the quirk in your asset class that you have been handling manually for nine years. The left column is a purchasing decision. The right column is the firm.

Buy the left column without sentiment. All of it. Fast.

On the right column, do not run a pilot. Pilots are how the 27% got to be the 27%. A pilot has no owner, no deadline that matters, and no consequence for producing nothing. Pick one workflow that costs real hours and ends in a decision somebody signs, and build that one thing until it is in daily use.

Fix the data inside the project, not before it. This is where "AI is only as good as your data" gets misread. Firms hear it as a prerequisite and commission an eighteen-month warehouse program, and the AI work never starts. Data quality is real, but it is a per-workflow problem, not a per-company one. The data behind one underwriting decision can be made clean in weeks. The data behind everything cannot be made clean at all.

Where we stand

We build custom software for CRE, so read all of the above knowing which side of the question we sit on.

The position is not build everything. Most of what a CRE firm runs on should be bought, and the Realcomm advice is correct for that majority. The position is narrower than that, and it is this: the cost question and the differentiation question are different questions, and a firm that answers both with a procurement process ends up with the same capability stack as every competitor in its market, at a good price.

That is a fine outcome. It is also the definition of interchangeable.

Buy the plumbing. Build the part with your name on it.

Got something you want made? We're around.