The Hidden 90% of AI Costs Every CRE Firm Ignores

CRE firms have now moved through three technology cycles that each asked the same question at the wrong altitude: business intelligence in the 2000s, RPA in the 2010s, and generative AI now. Each time, the industry asks what the license costs. Two decades of watching these cycles play out says the license was never the number that mattered. The number that matters is what happens to cost per additional use case at the fifth deployment, the 20th, the 100th. Most CRE firms still can’t answer that question, and it’s costing them more than the license line ever will.

The wrong approach is easy to spot. Most firms evaluate AI economics the way they’d evaluate traditional enterprise software: seat cost, contract length, implementation fee. That framing breaks once AI is deployed across a portfolio, because the real cost structure isn’t the license. It’s everything wrapped around it: integration, governance, maintenance, and the recurring labor of keeping a fast-moving technology from going stale. License cost is the visible 10 percent. The other 90 percent is where the budget actually goes, and almost nobody prices for it up front. It’s not that traditional CRE platforms are the wrong investment; they remain the operational backbone for many firms. The challenge is that AI introduces new economics that extend beyond the software purchase itself.

Three cost models have played out at scale across this industry. All three break the same way, just on different timelines.

The first is buying point solutions — a lease abstraction tool here, a market intelligence tool there, a separate product for tour books. Each tool stands on its own. But at scale, none of them share infrastructure. Each requires its own integration, governance and maintenance. The result is what the industry calls AI islands: real capability, zero compounding. This is a large part of why roughly 95 percent of generative AI pilots fail to produce measurable business impact. The models work, but each deployment starts from zero. Hence, the economics never improve.

The second is building in-house. Nearly every CRE tech team underestimates this one the same way: coding is maybe 10 percent of the real work. Solutioning (keeping systems current, auditable, and accurate as data changes) is the other 90 percent. That’s not a one-time cost. It requires an ongoing operating capability. The firms that get this right build that capability once and reuse it across every new AI workflow, rather than rebuilding the same foundation with each deployment.

The third is hiring a consultancy or systems integrator to run the transformation. This model scales with time and headcount by design — billed hourly or by FDE, so the incentive runs toward more hours, not faster outcomes. The pattern is consistent: strategy decks and forward-deployed engineers, a multi-quarter runway before production. Across the industry, large enterprises average nine or more months to move a single AI use case from pilot to production, against roughly 90 days for mid-market firms doing the same thing. That gap isn’t a capability gap. It’s what happens when a delivery model is priced by the hour instead of the outcome.

None of these three models solve what actually determines AI economics at CRE scale: whether cost per additional use case falls, stays flat, or climbs from governance sprawl. In a portfolio business, that answer is everything. A firm running lease abstraction, tour book generation, bid management, and market intelligence across dozens of business lines isn’t deploying four projects. It’s deploying one capability four times, and the economics work only if the second deployment is cheaper and faster than the first.

Scale also changes what a small accuracy gap is worth. Small workflows tolerate rough edges because errors are caught manually. A workflow running against tens of thousands of leases a year cannot; a few points of accuracy translate directly into hours of rework and downstream reporting errors. Economics at scale isn’t just the platform cost. It’s what a model’s error rate costs at real portfolio volume, week after week, with no one checking each output by hand.

The shift that’s happening: pricing per solution, per year, against a measurable outcome — not per token, not per seat, not by the hour. That sounds like a commercial detail. It isn’t. Token and seat pricing punish exactly the workflows CRE runs hardest — lease volume that spikes around renewal cycles, tour book demand that surges before a big pitch, underwriting volume that swings with deal flow. A pricing model built around steady per-seat usage will always look expensive against real estate’s lumpy, cyclical volume. Outcome-based pricing, tied to a shared platform rather than metered consumption, aligns with the portfolio-driven nature of commercial real estate.

The compounding piece is easy to underestimate until it’s observed directly. Global commercial real estate leader Cushman & Wakefield started with a single lease abstraction workflow, processing a portion of the roughly 40,000 leases it handles annually, and now runs 15-plus AI solutions — on one shared data and governance layer that works with (not against, not instead of) the firm’s existing technology stack. The economic argument is simple: The cost of the next use case should go down, not reset, and its accuracy should not start from scratch.

Don’t think of it as how much AI costs this quarter. The real question is how much it’ll cost for the next 10 use cases on top of it, or the cost of errors when use cases run at real portfolio scale. If the answer is “about the same as the first one, and errors stay rare,” that’s a model built to scale. If the answer is “unclear — it’ll depend on integration and maintenance as it grows” — that’s a failure waiting to happen.

The firms that win with AI won’t be the ones that buy the most tools. They’ll be the ones that build economics that compound across every AI deployment. 

Explore the Top 50 AI Agent Use Cases in Real Estate by downloading the guide here.