What AI Actually Changed at Our Firm

What Our Prior Process Looked Like

Before getting into what’s changed, it’s worth being clear about what we’re comparing it to. Our prior workflow was, frankly, the workflow most middle-market sponsors still use today. Property managers and leasing brokers produced monthly reports. Asset managers reviewed them, added context, and passed summaries up to Excelsior management. Management then synthesized the information into investor communications. Notes from calls and meetings lived in individual notebooks and email threads.

That process worked. It was — and still is — the industry standard. We’re not telling a story about how broken our old way was, because it wasn’t. But every handoff in a chain introduces the possibility of interpretation, compression, or loss. A nuance noted by a leasing broker might not survive translation into an asset manager’s summary. A point made on a quarterly call might be remembered differently by three different people a year later. None of this is anyone’s fault. It’s just how information moves through a firm when the only tools available are human attention and document-passing.

What changed isn’t that we discovered our prior process was inefficient. What changed is that new tools became available that genuinely enhanced what was possible. Once that happened, the question wasn’t whether the old way was broken. It was whether we were willing to adopt better ones.

The Dashboard

The center of gravity for our AI work is an asset management dashboard we built collaboratively with Claude and which lives in Excel. It holds the static information you’d expect — square footage, lender, maturity dates, extension options — alongside dynamic information that updates as the portfolio moves:

• Lease expirations across all assets
• Projects flagged by property managers
• pProperty damage incidents
• A/R trends, budget variances

Everything in one place, organized to be queried.

It’s hard to overstate how much that single shift has changed the rhythm of how we work. The dashboard is up on management’s screens every day. When we need to pull every lease expiring in a particular window across the portfolio, we have it in seconds. Our bi-weekly asset management meetings are now driven by the dashboard — it generates the agenda, surfaces the issues that need attention, and ensures we’re focused on what’s actually moving rather than what someone happened to remember to flag.

Layered on top of the dashboard is a monthly financial review agent. It doesn’t replace our normal monthly review process — that still happens, and human judgment still drives the analysis. What it does is produce a written report that surfaces portfolio-level trends and flags items worth digging into:

• A/R direction over the trailing three months
• Variances against budget
• Near-term lease rolls.

It’s a parallel capability, not a replacement. It means the next time someone asks how A/R is trending at a specific asset, we don’t have to pull spreadsheets and reconstruct the picture. The picture is already there.

The harder thing to convey is what this does to information flow. The dashboard standardizes the questions we ask across different property types and different third-party vendors. Standard questions yield standard inputs. Standard inputs flow cleanly to investor reporting that preserves the underlying detail rather than smoothing it away. The chain that used to move from broker to asset manager to management to investor — with potential loss at every step — now moves through a system that holds the original signal intact. That, more than any specific tool, is the operational change that matters.

The Other Places AI Shows Up

The dashboard is the flagship, but it isn’t the whole story.

On deal sourcing, we built a screening agent that processes the inbound flow from brokers — hundreds of opportunities per week — against our investment criteria. It auto-categorizes each one as a probable fit, something worth a closer look, or not a fit. Early on we reviewed everything to validate the prompts were working correctly. Now the categorization runs largely on its own, and we spend our time on the fringe cases where judgment actually matters. The shift sounds small but the cumulative effect is real: we no longer pay an attention tax every time an OM hits the inbox.

For underwriting, Argus still does the modeling. We haven’t tried to replace that. But we do use AI to pressure-test our assumptions, organize market research, and accelerate the work that surrounds the model rather than the model itself.

For document review, we use AI for lease abstraction and contract review — as do our attorneys, which has meaningfully reduced legal bills on routine work without sacrificing the human judgment that matters on the consequential pieces.

For asset-specific document interaction, we use Google’s NotebookLM to query loan documents, leases, and other materials for a given property. This was once the centerpiece of our AI work — for a while, it was the thing we were building everything else around. It’s still useful and still in active use, but it’s moved to a supporting role as the dashboard ecosystem has grown. That evolution wasn’t a failure of NotebookLM. It was a recognition that the dashboard had become a higher-leverage place to put our building effort.

What Didn’t Work

Honesty matters here, because anyone serious about AI adoption is going to have a list of things that didn’t pan out.

The most significant one for us was an attempt to build an investor relations portal that would automate parts of investor communication and access. We started down this path early and pivoted away quickly. Two reasons.

• First Reason: The security requirements for handling investor information are real, and our internal expertise in building and maintaining that kind of infrastructure wasn’t where it needed to be to take on the responsibility.
• Second Reason: Even if we had built it, the cost of delivering the level of security our investors deserve would have been comparable to what we already pay for the existing infrastructure — with significantly more time input on our end. The math didn’t work, and the risk profile was wrong. We moved on.

The other honest thing worth saying is that the spreadsheet-building process — which has been one of the genuine surprises of this work — is also one of the more humbling. AI can build a working spreadsheet, including reasonably complex logic, on the first pass. That’s still impressive every time it happens. But the first pass is rarely the final pass. Getting from 95% correct to actually correct requires real iterative conversation: finding errors, identifying incorrect assumptions, walking through the logic out loud. It’s productive work, and the end product is usually better than what we would have built unassisted. But anyone expecting magic on the first try is going to be frustrated.

The broader lesson from both of these is the same: the test is value, not novelty. The IR portal didn’t add value once we understood the actual cost and risk. NotebookLM still adds value but earned a smaller role as something better came along. The spreadsheet work adds enormous value, but only if you commit to the iteration. None of these are failures of the technology. They’re judgments about where the technology earns its place.

Where We Don’t Use AI

Lines matter. Here’s where we’ve drawn them.

Anything touching investor information is off-limits. That’s the lesson of the IR portal pivot, and it now operates as a firm policy. Investor data flows through the channels we already trust.

Investment committee decisions, distribution decisions, lease decisions, and any judgment call about the direction of an asset or the portfolio still rest with the team. AI organizes the information that informs those decisions. It doesn’t make them.

Tax and legal questions are areas where AI can help us frame conversations and ask better questions, but our partners — our tax advisors, our attorneys — still lead. The technology doesn’t have skin in the game. They do.

The throughline across all of these is the same one that has shaped the rest of the work. AI is useful where it organizes, surfaces, standardizes, and accelerates. It isn’t useful where the judgment, the relationship, or the risk lives somewhere that AI can’t actually go.

What This Adds Up To

A year and a half in, the way we’d describe what AI has done for our firm is this: we are not making different decisions than we would have been making otherwise. We are making the same decisions with better, faster, more complete information — and with documentation that lives in an auditable place rather than in the heads of whoever was on the call. A year from now, we can pull up the notes from a tenant’s one-year extension conversation. Two years from now, we’ll still have them. That kind of institutional memory used to depend on whose notebook a note happened to land in. Now it doesn’t.

The firms that will be at a structural disadvantage in three to five years aren’t the ones that fail to use AI for grand transformation. They’re the ones that fail to use it for the unglamorous work of organizing information well, communicating it cleanly, and remembering what was said. The compounding effect of doing those three things better than you did them before is significant — even if no single use case feels revolutionary on its own.

The other thing worth saying as we close: this is not finished, and it never will be. The pace at which these tools are evolving means that staying on top of what’s available, evaluating new capabilities against the same value lens, and being willing to pivot away from tools that no longer earn their place is now a permanent part of running a firm.

Six months from now there will be capabilities that don’t exist today. Some of them will add real value for our properties and investors. Some won’t. Distinguishing between the two — and being honest about which is which — is the discipline. It’s the same discipline that’s shaped this work from the beginning, and it’s the one we’ll keep coming back to.

Excelsior Capital is a private real estate investment firm focused on value-add acquisitions of industrial, medical office, and retail assets in growth markets across the Southeast and Midwest United States. Nothing in this commentary constitutes investment advice or a recommendation to buy or sell any security. Past performance is not indicative of future results.