When evaluating AI, it ultimately comes down to two questions: can I trust the output, and can I explain it? There is always a risk that you can't really explain the output when AI is done badly.

What "done badly" looks like is more common than most firms realize. A generic AI can produce output that looks polished regardless of what's underneath it. Because outputs can vary — whether because of how a question is phrased, how a model is configured, or changes outside the user’s view — the result isn’t always reproducible. For any AI operating in an audit-sensitive environment, that’s a disqualifier. 

Despite this, most vendors in this space have no external accountability for how their AI handles data, produces outputs, or manages risk, and the governance claims on their websites haven't been tested by anyone outside the company. That grey area between what's claimed and what's verified is where the risk lives. 

Can the output be audited? If the auditor followed the same exact process, would they get the same exact answer? Can a valuer defend that output to a lender? Most AI produces probabilistic outputs, meaning the result can shift depending on how you phrase the question. Because Altus's AI is anchored to a deterministic calculation engine, the same inputs produce the same output every time. Every output is traceable to its assumptions, every assumption is visible and challengeable, and the results are reproducible.  

2. Data handling

CRE deals routinely involve confidential financial information, personally identifying information, and material non-public information. The questions that should come before features or pricing: is client data training the model? Who has access? How long is it retained? At Altus, client proprietary data is not used to train the model. Customer data ownership stays with the customer. 

3. Validation

Who built the AI, who tested it, and against what? ARGUS Assist, our first AI agentic layer in the ARGUS Intelligence Platform, was developed in partnership with our own Valuation Advisory Services team, practitioners who manage thousands of property valuations every quarter. They trained the model against thousands of real-world test cases, interrogated its outputs, and corrected its reasoning. If a vendor can't tell you what data the model was trained on and who validated it, that's your answer. 

The CRE industry doesn't have a long track record with AI. Most firms are evaluating these tools for the first time, and the vendors selling them know the buyer's governance fluency is still developing. That asymmetry is where bad decisions are made. 

ISO 42001 certification exists to close that gap. Altus already holds ISO 27001 for information security, and ISO 42001 extends that same rigor to AI. In an industry built on trust, independent verification is the essential baseline. 

As our CEO Mike Gordon puts it: "Our customers rely on Altus to inform some of the most consequential real estate decisions they make. When it comes to AI, they need confidence not only in the data, but also in the technology behind it."

The technology will keep evolving. The firms that adopt it well will be the ones that asked the hard governance questions before they signed, not after something went wrong.