1. Every portfolio-level question requires manual assembly

A portfolio manager building a portfolio strategy wants to know which markets justify new acquisitions, and for which property types. That means comparing occupancy assumptions against market averages, looking at which industries are growing in each submarket, weighing asset-level factors (like accessibility for industrial or walkability for multifamily), and narrowing by subtype, such as grocer-anchored retail.

In most firms, answering that question means someone opens several systems, pulls each metric, lines up definitions that don't quite match, and builds the comparison by hand. The analysis itself might take an afternoon. Getting the data to a point where analysis is possible can take days.

Adding AI to that process speeds up the assembly, but the tool is still stitching together data that was never connected. Where definitions don't line up, it may reconcile them in ways nobody can see, and the comparison it hands back will look polished and plausible, even if it’s not. When those same metrics are tied to the asset with consistent definitions, the question becomes something you can ask directly, and the time goes into the decision instead of the preparation.

The check: think of the last cross-portfolio question your investment committee asked. How many systems did someone have to open to answer it?

2. You can't trace an assumption from two years ago

Two people are working on an acquisition model. One adjusts a vacancy assumption and sends the file back. Nothing flags what changed. Six steps into the acquisition workflow, someone asks why vacancy is at 30% in this submarket, and finding the answer means retracing every document, version, and handoff to see where the number came from.

The information exists, but it lives in disconnected files with no consistent record of what changed, when, or why, so reconstructing the chain becomes a manual project.

An AI tool working from those files inherits the same blind spot. It can tell you vacancy is 30%, but it can't tell you who set it or why, and that's the question a lender or auditor will ask. This can be solved by anchoring your data to the asset itself, so every assumption and model version sits on a single record anyone can trace back through.

The check: pick a model from two years ago and ask who changed the key assumptions and why. Can you answer without tracking down a person?

3. You don't know what's in your data until AI finds it for you

Even when files are easy to find, they are usually built for a specific purpose. Spreadsheets carry formulas, subtotals, and formatting designed for one report or to answer a specific question. When they become inputs for a different analysis, the AI can't tell a data point from a calculation artifact. The same thing can happen with rent rolls and valuation models, where a total rent figure gets read as a single tenant paying well above market. The output looks clean, but the error is buried in inputs that were never designed for the question being asked.

The check: would the files you plan to feed an AI tool make sense to someone who has never seen the report they were built for?

4. Your institutional knowledge walks out the door every Friday at 5:00

Most firms have someone who knows where everything lives: which file is current, which version of the assumptions to trust, which naming convention a former colleague used. The system works as long as that person is available.

Christina Gratrix, Senior Director, Product Management at Altus Group, remembers asking colleagues early in her career how certain decisions were made and getting an answer that amounted to “you just know.” That works when knowledge passes from one experienced person to another. “If the logic only exists in someone’s head, an AI agent can’t reliably use it,” she says.

Commercial real estate has a particular challenge here, because terminology isn’t standardized. A valuation model can make perfect sense to the person who built it without being consistent with anyone else’s. Small inconsistencies become hidden hurdles, like a suite entered as 300 in one model and 0300 in another, or different analysts labeling vacant space as “Vacant,” “VAC,” or “Speculative.” An AI tool can search and surface files quickly, but without standards underneath, it returns noise.

The check: if the person who knows where everything lives took two weeks off, could your team still find the current version of a model?

5. You're evaluating AI tools before asking where the answers come from

The first four signs are about your data. The last one is about the tool, and it's the easiest to miss because evaluating vendors feels like progress. Demos look impressive, but most firms skip a basic question: when you prompt this tool, where is it getting its answers?

Not all AI platforms offer the same transparency into where an answer comes from. Some show their sources, while others hand back a confident answer with no clear trail to the data behind it. The response may read well, but if you can’t see where it came from, it won’t hold up in front of a board, an auditor, or an investment committee.

Consistency is the other problem. Ask the same question about the same asset five times and you may get five different answers, because the tool isn't grounded in your data. In an industry where a valuation opinion has to withstand scrutiny, an answer that changes depending on how you ask is a liability. Valuation work calls for deterministic AI: grounded in your data, and giving the same answer to the same question every time.

"If you don't know where to find the data to answer the question, you're not ready to get AI to do it," says Gratrix. "You should use AI to help you flag trends and analyze data, so you can make an informed decision efficiently, and draw your own conclusions faster. Not to generate the information itself."

The check: ask a vendor where a specific answer came from, then paraphrase the same question twice and compare.

What these signs have in common

Recognizing a few of these is common, and it doesn’t mean AI is off the table. It means the order matters. Every sign above traces back to the same foundation: data tied to the asset, a record of what changed and why, standards your team actually follows, and clarity on where an answer comes from.

On their own, each of these gaps is manageable. Once AI is in the workflow, every one of them gets amplified. Firms that do the foundation work first will get more out of AI, and they’ll be able to stand behind what it tells them.

Want to be notified of our new and relevant CRE content, articles and events?

Disclaimer

This publication has been prepared for general guidance on matters of interest only and does not constitute professional advice or services of Altus Group, its affiliates and its related entities (collectively “Altus Group”). You should not act upon the information contained in this publication without obtaining specific professional advice.

A number of factors may influence the performance of the commercial real estate market, including regulatory conditions and economic factors such as interest rate fluctuations, inflation, changing investor sentiment, and shifts in tenant demand or occupancy trends. We strongly recommend that you consult with a qualified professional to assess how these and other market dynamics may impact your investment strategy, underwriting assumptions, asset valuations, and overall portfolio performance.

No representation or warranty (express or implied) is given as to the accuracy, completeness or reliability of the information contained in this publication, or the suitability of the information for a particular purpose. To the extent permitted by law, Altus Group does not accept or assume any liability, responsibility or duty of care for any consequences of you or anyone else acting, or refraining to act, in reliance on the information contained in this publication or for any decision based on it. The distribution of this publication to you does not create, extend or revive a client relationship between Altus Group and you or any other person or entity. This publication, or any part thereof, may not be reproduced or distributed in any form for any purpose without the express written consent of Altus Group.