Every real estate portfolio operator is being pitched the same thing right now: let one of the big AI "model companies" come in, embed a team of engineers in your operation, and transform your business. It sounds impressive. It is also, for most operators, the wrong deal — and it's worth understanding why before you sign one.

Tokens are a cost, not a result

When a model company charges you, it charges by the token. That's not a business outcome. It's a meter, no different in spirit from paying for CPU, storage, and bandwidth. The vendor gets paid when the meter spins, whether or not your NOI moved an inch.

Here's the uncomfortable implication: the vendor's incentive is to make the meter spin faster. The most reliable way to do that is to make the architecture more complicated. More retrieval steps, more model calls, more "agentic" layers, more infrastructure — every added layer of complexity is also an added layer of consumption. Complexity isn't an accident of their approach. It's the business model.

That's why the model companies have spent the last year building large forward-deployed engineering and consulting teams and parachuting them into enterprises. Framed as "help you find use cases," it often functions as high-end sales: find the workflow that burns the most tokens, wire the model deep into it, and lock in the switching cost. A doctor who sells the drugs he prescribes has a conflict of interest. So does an AI vendor who designs your system and profits from how much it runs.

The market has already figured this out

This isn't a contrarian hunch anymore — it's the story of the moment. In a July 2026 piece titled "Corporate America Has Suddenly Decided to Stop Blowing Money on AI," The Wall Street Journal reported that companies "big and small" are done overpaying, mixing in cheaper and open models and shopping "a la carte" for AI. The mindset flipped almost overnight: what the Journal calls "tokenmaxxing" — bragging about how much you spend — has given way to "thrift-maxxing." A field CTO at Cursor put the premium-model reflex bluntly: "It's like driving a Lamborghini to go to the grocery store to pick up milk."

The savings aren't rounding errors. The same reporting cites one task that cost more than $10,000 run entirely on a single top-tier model, but $1,339 when the work was mixed intelligently — a cheaper model doing the bulk, a premium one reserved for the hard parts — and one startup that faced a $100K daily bill under a pay-per-use pricing change before switching approaches. The premium-everything, meter-always-running architecture is precisely the cost structure operators are now racing to escape.

What a portfolio operator actually needs

You don't need a token meter attached to your rent roll. You need answers to the questions that move a portfolio: Where should rents sit across these assets this quarter? Which of these deals clears our criteria? What's actually happening across the portfolio's KPIs this month? Those are value-creation questions, and the whole point is to answer them for less than they cost.

An AI deployment only makes financial sense if the benefit — a better rent decision, a faster underwrite, a deal you didn't miss — clearly exceeds what you pay to run it. A general-purpose model, wrapped in general-purpose complexity, sold by the token, is structurally the hardest way to make that math work. Most enterprise AI stalls in exactly this trap: impressive demos on the edges, ballooning bills, and no line you can draw to the P&L.

Why a specialist, and why smaller

This is where a focused partner beats a model company, and it has nothing to do with who has the better model. In real estate services, the value doesn't live in the model — it lives in the data, the domain, and the workflow. It lives in knowing what an OM actually contains, how underwriting really works, where proprietary market data changes the answer, and how to fit AI to the way your team operates rather than the other way around.

That's what a smaller, specialized shop is built to do. At Locate Alpha we custom-deploy systems into real estate operations — pricing rents automatically, letting you converse with your own data warehouse, structuring deal data, and monitoring portfolio KPIs — using proprietary real estate data, on an architecture we deliberately keep as lean as the job allows. We're model-agnostic on purpose: we match the right model to each task instead of routing everything through the most expensive one, so you're never driving the Lamborghini to buy milk. We win when your outcomes improve, not when your usage bill climbs. We can even run it on a local, secure model so your consumption isn't a monthly surprise.

The model companies should keep doing what they're genuinely great at: building models. Turning those models into value for your portfolio is a different job — and it's not one you want done by the party that profits from complexity.

Want to see what a lean, outcome-first deployment looks like on your portfolio? Request a demo.