Everyone Adopted AI. Almost Nobody Got the Payoff.

Douglas Elliman just showed one clear way to close the gap between adoption and results.

Real estate brokerages just posted some of the highest AI adoption numbers of any industry. Ninety-seven percent of brokerage leaders say their agents use AI, up from eighty percent in 2024. In commercial real estate, JLL (formerly Jones Lang LaSalle, a global commercial real estate services firm) surveyed over a thousand industry professionals and found AI pilot programs jumped from five percent of firms to ninety-two percent in just three years.

But adoption was never the hard part. The same JLL research found only five percent of commercial real estate firms say they have actually achieved most of their program goals. A separate industry report on residential brokerages found the productivity gains concentrate in a small group of power users, while most agents report little to no meaningful impact on their actual output. CBRE (formerly CB Richard Ellis, a global commercial real estate services firm) alone spent nearly one point seven billion dollars on computer hardware and software last year, up three hundred million from the year before, yet the return on that spending is still described internally as uneven.

A May 2026 industry survey found something worth sitting with: agents are getting real productivity gains from AI, but often not from the tools their own brokerage rolled out. Many reach instead for outside tools like Claude, Gemini, or Perplexity, tools nobody procured, licensed, or trained them on. The company bought the platform. The employees quietly went around it.

The Douglas Elliman Example

On July 8, 2026, Douglas Elliman, one of the largest luxury real estate brokerages in the country, announced a new AI subsidiary called Elius, built with Google Cloud. Chief Executive Officer Michael Liebowitz said the goal is to turn the brokerage's own data into a real intelligence layer, looking at the work agents actually do task by task, instead of treating "real estate agent" as one job title with one blanket automation answer.

That is the approach we think more companies should take, whether or not they build their own AI subsidiary. Look at each role task by task, the way Elliman is doing with Elius. Then talk directly to the employees doing that work and ask what would actually make them faster and less frustrated. Combine the two. The task by task data tells you where the opportunity is. Your employees tell you whether the fix will actually get adopted. The verdict on Elliman's approach is still out, since the platform only launched this month, but they are one of the first major brokerages to commit to this model publicly rather than simply buying a chatbot add-on.

This is also, plainly, the model our own scoring approach at ActivateAI and tobywins.ai is built on: task level analysis of a role using government occupational data, not a blanket score for a job title. Companies do not need to build their own Elius to start this way. The same task by task lens can be applied to any workforce.

What You Can Actually Do This Quarter

1. Pick one department and map its roles task by task, not title by title. A loan officer or an operations manager is not one job, it is a bundle of a dozen or more distinct tasks, some easy to automate, some not.
2. Ask the people in that department directly what slows them down day to day, before introducing any new tool. Their answer will not always match what the task data suggests, and that gap is useful information.
3. Combine both lists. Automate the tasks where the data and the employees agree. Pilot, do not mandate, the tasks where they disagree.
4. Measure adoption and output separately. Ninety-seven percent of people using a tool means nothing if the same small group is carrying all the productivity gains.