Commercial real estate has long resisted automated valuation. Unlike residential properties, commercial buildings are fundamentally more complex. Valuing a house is largely an exercise in finding comparable sales. The large deal volume and asset conformity make automating valuation relatively straightforward. CRE valuations depend on factors that vary widely by asset, market, and moment. Context is key.

A commercial valuation depends on things that do not appear in a transaction record. The creditworthiness of the tenants and the long term prospects of their industry. What is happening in the submarket, including buildings that have not traded and tenants who have not yet moved. Whether a new employer is coming, whether an anchor is leaving, whether the zoning is about to change. “The information needed for good commercial real estate analyses comes from a lot of different places,” said Travis Valentine, Chief Technology Officer at Green Street. “It’s not just data but also research.” Two identical buildings on the same block can be worth meaningfully different amounts based on facts that live in news coverage and analyst reports rather than in any database.

Assembling that context has traditionally been the job of teams of people. Analysts read, synthesize, and apply judgment about which of a hundred relevant facts actually matter for a particular asset. It is slow work, it requires expertise, and it does not scale easily. Which is precisely why it has become interesting as an AI problem.

“AI now allows you to ask really complex questions, things like ‘what has COVID done to the NYC office sector,’ and can provide a ton of context about all of the macro and micro forces that are influencing the answer to the question,” Valentine said. Questions like that used to require a research team and a week. The answer draws on hundreds of sources and requires understanding which forces are causal and which are coincidental, which is exactly the kind of synthesis that language models have gotten good at when they have quality material to work with.

To help AI agents with this hard work Green Street has created an MPC Server provides accesses its real estate data alongside more than 150,000 news articles and over 25,000 research reports. “All of our research and news is broken up into labeled sections so AI agents can quickly understand what part of each article or report is pertinent to its question,” Valentine said. An agent that has to read an entire report to find one relevant paragraph is slower and more prone to pulling the wrong thing. Structuring the information so that it can be queried at the section level is the unglamorous work that determines whether the output is useful.

The applications get more interesting as the question gets narrower. Broad market analysis is well served by existing tools but specialized property types and focused investment theses are where analysts have traditionally been most necessary and most expensive. “If a company is looking at something like self storage they can pull stats about number of units in a market, rental rates, and vacancy,” Valentine said. “AI can add a lot of value by looking at research around the particular players in the market and the trends happening in that area.” Knowing the vacancy rate in a self storage submarket is useful. Knowing which operators are expanding there, who is under pressure, and what supply is planned is what actually informs an underwriting decision.

That context is also making automated valuation models work for commercial property in a way they mostly have not. An AVM that runs on transaction data alone will produce a number, but it will miss the tenant situation, the submarket trajectory, and the local factors that a human analyst would weigh heavily. AI can only add value when the data is from trusted sources, which usually excludes most data from the open web.

Firms can also bring their own information into the process. “Our AVM pulls in our data but also allows users to add their own data,” Valentine said. “This helps not only add context to the valuation but also see how proprietary data stacks up against our analysis.” That second function is arguably more valuable than the first. A firm that has been underwriting a market for years has views embedded in its own numbers, and the ability to test those views against an independent analysis is a form of error checking that most companies have never had access to.

The competitive implications are worth thinking through. Large property companies have long held an advantage that had little to do with real estate skill and everything to do with headcount. They could afford analysts. Smaller firms could not, which meant they operated with less information on every deal. “Historically big property companies had a huge advantage over smaller competitors because they could pay for analysts,” Valentine said. “This levels the playing field and helps smaller shops have the same kind of research without hiring an entire research team.” Analysis that took days now takes minutes, and the firms that benefit most are the ones that never had the days to spare.

None of this works without the underlying material being sound. An AI system reasoning over thin or unreliable research will produce confident conclusions that are wrong. Commercial real estate investment is not a place where that failure is acceptable.

What we are seeing now is an early version of what this will eventually look like. The models will improve, the research will get deeper, and the connections between qualitative context and quantitative output will get tighter. Commercial valuation has always been a research discipline that has to boil a huge amount of factors into one number. Now the industry can produce that number, in a contextualized way, considerably faster.

The post AI Is Adding Context to Complex CRE Valuations appeared first on Propmodo.

​