Questions to Ask Before Vendor Promises About AI Sweep You Off Your Feet

JLL recently released its look at artificial intelligence as a tool for “real estate transformation.”

“The potential for artificial intelligence (AI) to transform businesses, industries and society has been mounting for decades,” the company wrote. “But recent advancements, have moved the science from niche to mainstream. The technology’s proficiency in writing, drawing, coding and composing has compelled corporate leaders to consider both the opportunities and threats that AI presents for their future.”

In the near term, the firm pointed proptech as having “laid a solid foundation” for using AI in CRE applications and that there were more than 500 companies “providing AI-powered services to real estate and already delivering value in terms of improved efficiency and cost-savings.” Some of the applications include document sorting and data standardization, scheduling, price modeling and prediction, satellite image processing for asset valuation and risk management, and recommendations and matchmaking for leasing and investment transactions.

All true to some degree, but for investors in tech and users of it, it’s important to remember two things. One, how often over the last 40 years vendors have claimed capabilities that they didn’t have. Two, the range of what is considered “AI” and how limited many types of the technology can be.

Since at least the 1980s with a hypes wave of “paperless office” technology (and probably before then), many vendors have jumped on bandwagons whether or not they significantly implemented the concepts. Similar things have happened with predictive analytics, supply chain management, ERP, and other areas.

Also, AI started as a technology type in the 1950s. There have been many ways of approaching the concept of offloading types of cognitive work to computers, all with potential benefits and significant limitations. For example, machine learning, a concept close to 70 years old and which began to be widely used in the early 2000s, can go wrong when training data has problems, like incompleteness or bias that will throw off decision quality. Even with so-called deep learning systems, having to train and retrain software, if anyone realizes such additional work is needed, can take far more time and resources than users expect.

Generative AI like ChatGPT holds a lot of promise, but has already seen significant hiccoughs, like fabricating sources of data or needing much more direction and shepherding than the casual user realizes.

This isn’t to suggest that no one profits from using the technology. As JLL notes, some companies have seen significant savings in energy costs as well as carbon reductions. But buyers of technology should always ask questions. What exactly is the “AI” part of a product or service? What are the demonstrable benefits? Are there existing users (particularly when found outside of the reference accounts vendors offer) that can validate such claims? What is the roadmap for future development? And are vendors forthcoming about the potential problems with the technologies they use and how they mitigate them?