What a Machine Was Never Going to Do
In part one, I made the case that a meaningful share of this region's luxury buyers are structurally unsearchable. 32% of luxury sales in the DC metro closed all cash last quarter, and inside that group sit the people whose names never enter a public record before closing, and in some cases not after it either. This means the signal any AI model would need to see them coming doesn't exist in a form it can reach. Which leaves the question. Where does the machine stop?
Here's where we draw the line at FORWARD, and it's a staffing decision rather than a philosophy. We automate everything that touches process. Anything we find ourselves doing the same way each time: deal status notifications, comparative analysis, listing operations and checklists, transaction management, database discipline, market monitoring across three jurisdictions with three sets of rules and three working definitions of what counts as disclosure. Every one of those runs faster and cleaner with a machine on it, and any team or brokerage still doing them by hand is charging clients for the privilege of being slower.
The day-to-day judgment stays with people. Pricing strategy sits at the top of that list, followed by when to hold and when to reduce, which is almost never the same call twice. Which of two offers will actually close, as opposed to which one looks better on paper. And who to call before a property is ever listed. In a market where the best inventory frequently never reaches a public search field, that's the entire game.
The honest version is that the boundary between those two columns moves, and I'd be selling something if I claimed otherwise. Pricing strategy sits downstream of our comparative analysis. If the machine assembles the wrong comp set, the person exercising judgment is working from a corrupted foundation but will feel completely certain the entire time. What matters is where that lands, because it doesn't stay with us. A listing priced off a bad set of comps doesn't fail on day one. It sits. The showings thin out, the adjustment comes three weeks after it should have, and the property collects a days-on-market number and a price history that every buyer's agent in the region can read before they ever walk through the door. We can make that mistake in an afternoon, and the client carries it for the length of the listing. Policing that line is a human job. The day we stop is the day the automation starts making decisions nobody consciously handed it.
What automation buys is the calendar. A team our size can act on market intelligence the same week it arrives instead of the same quarter. That compression only produces value if a person is deciding what to do with the head start. Speed applied to a bad read just gets you to the wrong place first.
Most of what gets sold as an AI strategy in residential real estate is a chatbot pointed at public records, which is a faster route to information that was already free. The teams that hold this market through the next cycle will be the ones that gave up every task a machine does better, then defended the four or five decisions a machine was never going to make, and understands the difference well enough to keep re-drawing the line as the tools improve.
Ryan Serhant's split between revenue-supporting and revenue-producing is the right frame, and he's working the same problem from New York. The difference here is one of degree. In a market this opaque, those relationships aren't only how a deal gets closed. They're how you learn there's a deal to be made at all. In Washington, the deal that defines this year is already being discussed by the people who will sign it, months before it appears anywhere an AI model can reach.
Marc Cashin is the Founder and CEO of FORWARD at Corcoran McEnearney, representing buyers and sellers in Washington DC, Maryland, and Northern Virginia. This article was first published on LinkedIn.
