.png)
Commercial real estate is adopting AI faster than it has adopted any technology before — faster than it adopted the internet. But most of the industry is still asking where it fits. In this conversation, two early adopters — Dave Welk, Managing Director of Acquisitions at Origin Investments, and Graham Russell, Senior Vice President, Americas at QuadReal — sat down with Altrio's Raj Singh to talk about what's actually working.
Key takeaways
Embed AI in the workflow, don't bolt it on. The tools that get used are the ones inside an existing process. Anything that requires a separate login or a new habit tends to stall.
The model is the easy 80%. Your data is the 20% that matters. Everyone has access to the same frontier models. The differentiator is well-organized proprietary data — and how you use and interpret it.
AI won't give you discipline. It gets you to an answer faster and helps you scale, but vetting the output and questioning the assumptions is still the human's job. Outsourcing judgment to a model that hallucinates is the real risk.
Smaller firms can close the data gap — and move faster. With third-party data and lighter compliance overhead, smaller firms can access information and adopt new tools quickly, narrowing scale's traditional advantage.
The analyst's job changes, not disappears. Less time building models from scratch, more time on the economics that move a deal — location, supply, submarket risk — and on managing and verifying an "army of agents."
AI magnifies the differences between firms. It doesn't level the playing field. Firms that already invested in good data, systems, and processes pull further ahead when they add AI.
Transcript
This transcript has been edited for clarity.
Why AI adoption in CRE is different this time
Raj Singh: Thank you all for listening in on this discussion around the adoption of AI within the commercial real estate investment industry. There's been no shortage of discussion on this topic — it's top of mind for folks at every level, at every organization, in real estate and in every other industry. Much of that discussion has revolved around what these tools can do and the exciting use cases people have discovered. There's also been a lot of debate about the relative benefits and risks for the industry and for society as a whole: the impact on jobs, the economy, the markets.
While there are many fascinating aspects of this technology, the one that stands out to me is the rate at which it's being adopted. It makes the adoption of the internet seem glacial. Real estate as an industry is famously not at the forefront of new tech — real estate companies are not usually early adopters. But this time around, it's different. We speak to dozens of firms every week, and almost without exception, everyone we talk to is well beyond the initial experimentation phase with these tools. Adoption is moving quickly, which doesn't leave a lot of time for learning and increases the likelihood of mistakes. In this environment, learning from the experience of others is very important.
So to dive into where we are, what we've learned so far, and how our thinking is evolving across a whole array of issues, I'm joined by two thoughtful and knowledgeable members of the real estate investment community. We have Graham Russell, Senior Vice President, Americas, on the investment team at QuadReal, and Dave Welk, Managing Director of Acquisitions at Origin Investments. Graham, Dave — thank you both for being here.
I'll start with the same question for both of you, sticking with this theme of learning: what's the biggest lesson each of you has learned recently about how best to leverage AI effectively, at your firms or even in your personal lives? Dave, do you want to kick us off?
The biggest lessons so far
Dave Welk: Thanks for having me, Raj — good to be on with you today. The biggest lesson we've learned over the past several months and years on our AI journey is that you get the most value from it when it's embedded into your actual workflow, not sitting alongside it. We've adopted six or seven distinct AI tools over the past couple of years, and the ones that moved the needle the most are the ones where the teams don't have to leave their existing process to use them — no separate login, no separate add-in. The ones with friction are the ones that stall. The ones seamlessly integrated into your daily work are the ones that stick.
Raj Singh: Makes sense. Before I ask a follow-up — Graham, what's been your biggest learning recently?
Graham Russell: Great to be here, Raj, thank you. My biggest learning lately, personally, has been the need to stay continuously curious and to question even what the status quo is. Throughout 2026, the pace at which these tools have been updated and changed really requires organizations and individuals to continually ask: is this the right way to do it? Can I be more efficient? Is there a new and better way to get this task done? I completely agree with Dave — integrating AI into your overall workflow is critical, and point solutions are becoming a thing of the past. But even with these new solutions, after the time you spend integrating and deploying them, you're constantly looking at what's next, and asking whether the architecture you've built is sustainable, or whether something better will come along next month.
Systems vs. agents — and why data is the foundation
Raj Singh: You both alluded to this idea of designing systems — Graham, you used the word "architecture"; Dave, you talked about embedding solutions into workflows. Traditionally, thinking through an architecture or a workflow was something you had time to do: you'd come up with a future state, build a plan, and put the pieces in place. But things are changing so quickly now that it's tough to commit to a direction without staying vigilant about your assumptions and revisiting them frequently. So let me take this toward systems versus agents. Dave, you mentioned meeting people where they are; Graham, you mentioned point solutions are dying. How do you think about agents versus systems, and how both relate to friction and adoption within your teams? Dave, maybe I'll have you start again.
Dave Welk: I think the two go hand in hand. Agents help you do very specific work — if they're prompted well and fully integrated. But another part of the overall takeaway is that if you don't have well-structured, well-organized data, even an agent can only do the task effectively because of what it can leverage. It can leverage publicly available data — there's a lot of that out there — and an agent can get you maybe 80% of the way there in terms of what it can access from public systems. But the last 20% is where the sausage is really made. The companies that have well-organized proprietary data can use the agent to hone in on the specific tasks they need it to complete.
We've built a system, in partnership with a third-party provider, that integrates our entire OneSite and OneDrive capabilities. All of our documentation is pulled into a centralized system with an LLM layer on top of it. But we're not even done organizing the data to make that system fully functional and efficient. We're having to piecemeal it up by division: legal has its own initiative with its own guardrails, accounting has its own, then investor relations, and then the deal teams I oversee. We're at the tip of the iceberg on cleaning and organizing data to make the system work efficiently. Then we're layering agents on top for very specific tasks — underwriting, deal evaluation, forecasting. You need both, but both will fail if you don't have the right foundation and infrastructure built.
Raj Singh: What I'm hearing is that the agents do the work, but what lets them do it well — potentially better than others — is the systems and data you have in place. Graham, I know that's been a big topic at QuadReal. At one of our recent catch-ups you mentioned looking at a lot of new data sources and feeds to use with these tools. Carrying on from where Dave started, how are you at QuadReal thinking about data, systems, and agents, and how they interdepend or support one another?
Graham Russell: What Dave alluded to is absolutely what we're thinking about: how important and critical context is for using these AI models and LLMs. Context — whether that's structured data or a prompt — is what makes them efficient. And it's not just about what's included in the prompt; it comes down to how the systems behind those agents are structured, such that they can give the right context for a particular problem. We see this across the problems we're trying to solve, whether from an underwriting or investment perspective: is it referencing the right data set? Does it have the context? How does that relate to other data sets we have access to, or are acquiring, or building ourselves? So when we talk about systems, it really comes down to providing these tools with the appropriate context for the problem you're trying to solve. We're spending a lot of time thinking through how to frame and structure that so we can be the most efficient.
Where the edge comes from now
Raj Singh: Even in the way you both describe getting the most out of agents, the conversation has clearly shifted over the last few months. Six or twelve months ago, the divide was between folks who were using these agents and folks who weren't, and there was all this stuff being thrown around — "you're not going to be replaced by AI, you're going to be replaced by somebody using AI." Fast forward six months and everyone's using AI. So the agents and the models are no longer the threat or the differentiator. It's now about what context you can give the agents, and how efficient you can make them with the right tools and information.
That takes us into this whole discussion around differentiation. Both your firms, like a lot of real estate firms, are very committed to making the most of AI — but you all have access to the same models. Maybe some firms are using Codex, some are using Claude, but you're all fundamentally using a frontier model of some sort. So how do you each think about where you get an edge, whether in your use of AI or even outside the realm of AI? To switch things up, I'll have you start first this time, Graham. How does a real estate firm think about advantage as these commoditized models get used more and more?
Graham Russell: I'll speak in more general terms than specifically to QuadReal. One way is proprietary data sets. The ability to harvest new and unique data and connect it into decision-making is becoming a lot easier. As firms think about their competitive advantage, those with real scale can leverage unique data sets they have access to, and the connections that come from them. That's a real competitive advantage we're going to see in the market. There have been plenty of aggregations among real estate managers over the last twelve months. Some of that reflects today's market factors, but some may be about aggregating data. As large firms get larger, they have access to more competitive data than some smaller groups — though at the same time, accessing larger data sets for a group that doesn't own them is becoming easier too. So is that a true competitive advantage? That's a little left to be seen. The pace at which everything is changing makes it hard to define a competitive advantage, or at least the timeframe we have to evaluate isn't long enough to determine "yes, this worked" or "no, it didn't."
The other piece is doing more with less. Can you expand the number of deals you're looking at or underwriting, more efficiently, with the human capital you have? Part of that creates a unique data set in itself: we looked at this many deals last year, this is how we'd view value, this is where they traded. That kind of backward-looking analytics on the team is a really interesting use case as investment and acquisitions teams look to become more efficient and develop their own competitive advantage in the market.
Raj Singh: Really interesting — you touched on at least three threads we could go further on. I'll let Dave choose which to pick up, but I heard unique data sets and the scale of that data. I also heard the idea that your team is now the people you have plus the agents — and the agents are commoditized while the people aren't. So how do you get the most out of your team? Dave, respond to any of those, or anything that hasn't been mentioned around how you're thinking about differentiation.
Just get started
Dave Welk: Graham touched on a lot of good and important points, and we can double-click into some of them. But I want to pull up to thirty thousand feet, because we're talking about the industry as a whole — not Origin or QuadReal per se. I think the people on this call are probably in the top 10 to 20% of AI adopters in the industry, so we're a bit in our own echo chamber. There are a lot of folks out there who may be listening to this who've heard about AI — you can't go anywhere without hearing about one of the LLMs and the new breakthroughs — but who haven't even started. They haven't found a way to integrate it, and there are a lot of questions about how to use AI in your daily work.
So my advice is: get started. The easiest thing is to go into Claude — or GPT, or any of the other LLMs — and just ask, "how would I integrate this into my workflow?" It's amazing how easily it'll give you better prompts. It has to be ingrained in the culture; at the firm, you have to make integrating AI a high priority.
On the differentiation side: there's a distilling down of competitive advantage. The whole 80/20 Pareto principle I mentioned earlier — 80% is the commoditized LLM layer everyone gets; that 20% is where the magic happens. That's your proprietary data, how you use it, how you interpret it. And the one thing AI is not going to do for any of us is instill discipline. It'll get you to an answer, maybe a more refined answer more quickly; it'll leverage different systems and data sources efficiently; it may let you scale — screen more deal flow without more people. But you still have to remain disciplined in our business at the end of the day. That sounds tongue-in-cheek, but it's true.
That's the risk in all this — that we outsource a lot of our decision-making to an AI agent that, as we've seen over and over, hallucinates all the time. It's not accurate. I've been through some of the Claude code that comes out in Excel and it's just not accurate. You have to be very aware of garbage in, garbage out, and you have to vet the output of these models, especially if you're using it to augment decision-making. That part will never be commoditized in my mind. We're going to get tons of scale, speed, and efficiency — but stay disciplined. Like Graham said at the outset, question the assumptions, question the output. That's the thing we all need to be aware of.
I also have a general concern about the next generation of analysts coming into our industry, because they don't have a baseline for taking a model or an investment from soup to nuts the way a lot of us were trained. There was a lot of trial and error — almost an apprenticeship, where you learned to build a model from scratch. Building an Excel model from scratch is gone now. You can produce an output instantaneously from one of these LLMs and basically tell it to input certain assumptions. So my concern is: as we evolve from here as an industry, how does the next generation of talent develop? I don't have a good answer for that, but it's a real question.
The biggest pitfall: trusting the output
Raj Singh: A lot to unpack there. A good reminder that there are still folks earlier in this learning curve, and some pitfalls to avoid. I also heard discipline as a potential differentiator, which I completely agree with. These models let you do the right thing quickly and efficiently — but they also let you do any number of wrong things quickly and efficiently, and they don't help you prioritize, focus, or decide what to do with the information. That layer of focus and discipline is very, very important.
Graham, when you think about where the pitfalls were as you made your way through this learning curve, what was the biggest course correction along the way — for the folks Dave just reminded us of, who might be where we were a little while ago?
Graham Russell: Dave nailed it: questioning the output is the biggest pitfall. The second might be giving up. Earlier on in my own adoption — and broadly, as an organization — it was easy to dismiss some of these tools as less efficient, less sophisticated, or not as smart as advertised.
Sticking with it is the best advice I'd give anyone. One of the really interesting things, as Dave was speaking, is that this is one of the few times I've come across a tool that can teach you how to use itself — basically a personalized tutor. That goes from structuring prompts all the way to more complex database development. Using it to learn how to use it is one of the genuinely unique things about this technology, at least in my view.
Raj Singh: I echo that. One of my biggest learnings early on was that your instinct is to tell it what to do — but if instead you ask it to ask you what to do, it asks really good questions, and you end up telling it things you might not have thought to, the context and color it needs. So putting it in the driver's seat for the questions is helpful. Although you never want it in the driver's seat on the outputs — you've got to remember it's the subordinate.
Advantages of being a smaller firm
Raj Singh: Graham, you mentioned scale and the importance of unique data sets, and that it's getting easier to aggregate data through third parties. There's a big difference in the scale at which QuadReal and Origin operate. There are obvious advantages to being a larger organization — Dave, from your perspective, what are some of the advantages of being a bit smaller, potentially more agile or more focused? How do you think about unique advantages related to the size of your organization?
Dave Welk: Starting with the fact that we can play like a much larger organization by leveraging other third-party data sets. Very early on, we built our own proprietary rent-forecasting tool, called Multilytics. We own or developed about 20,000 units internally, which we used to build a robust rent-forecasting model. The thing we ran up against is that we just didn't have enough units to produce the level of granularity we wanted to feed and train the model — we ran into a data challenge. This was a few years ago; since then, tools have been developed that you can leverage in the industry to do the same thing. But we were one of the first I know of that built a web-scraping tool to scrape the number of units, basically on a daily basis. We went from our own internal data set, plus a little third-party information from rent rolls in the marketplace, to a web-scraping tool that dynamically tracked eleven million units overnight. It took us a while to build the scraper and house the data, but it dramatically improved the model's forecasting ability once we could intersperse those eleven million units and track movements over time.
For a company like ours — we don't have a Greystar or Blackstone-type data set — we can get a lot more information than we used to, and, importantly, analyze it. Before AI, extracting eleven million units of data and analyzing it would have taken a team weeks; now we can do it almost instantly. So the advantage for a smaller firm is that you can close the data gap a bit. The proprietary-data advantage still exists, but the information gap that existed two years ago has narrowed dramatically — at least in multifamily, which is what we focus on. It's harder in other product types like industrial, office, and retail, but that's one big advantage.
Another advantage of being smaller: we have funds regulated by the SEC and we take that seriously, but we're allowed to be a little more nimble organizationally. We don't have quite as much compliance overhead as some larger peers, so we can move fast and build the plane while we're flying it when it comes to adopting AI and inserting new tools. Graham mentioned the rate of new model releases — just from 4.8 to 4.6 over a couple of months has been a light-year improvement in analytical capabilities and the speed at which Claude can process information. We can jump on that and ride the wave nimbly, whereas some larger peers have to get approval, go through compliance review, and involve several attorneys. We can move agile and fast.
Raj Singh: Historically there's always been a cost to being large — companies grow because there are advantages to scale. In some respects, the power of these tools gives smaller companies the same or even greater capabilities, and reduces some of the advantages of scale. Proprietary data is obviously difficult to replicate — but what I've found in my career is that proprietary data is often the hardest data to access. Publicly available information usually comes in a beautifully packaged API or export. Your proprietary data is the opposite — I shudder to think what it takes to aggregate data across hundreds of Yardi instances at a very large national operator. Just because you have the data doesn't mean you can use it.
Team, culture, and changing skill sets
Raj Singh: Maybe switching gears one more time, to talk about team and culture. Dave, you already mentioned you worry a bit about the next generation coming up without some of the first-principles experience you had — building models, looking at deals without the advantages of these agents. For both of you: how do you see a typical real estate organization evolving in terms of skill set, and the balance between senior and junior folks? Is it just fewer junior folks and more principals? What are your thoughts, Graham?
Graham Russell: I actually don't see the team structure changing dramatically. As Dave said, these tools can make you more efficient at underwriting and evaluating opportunities, but they ultimately can't make decisions — that still comes down to the structure we have in place. They do let us operate the team differently and perhaps expand the number of opportunities at the top of the funnel that we're evaluating. Even within our own portfolio, they let us leverage our own data to evaluate and improve how we operate assets, and truly grow value and NOI. All of these tools apply to those use cases. But specific to the team, I don't think it changes the structure dramatically, at least not in the short term — which may be a misconception in some of the market commentary right now. What it will change is the type of work, and how we use our data and tools.
Raj Singh: That might play into skill set. Traditionally a field like real estate investing has a heavy emphasis on technical, quantitative, and analytical skills. Maybe it shifts a bit more toward things that are more subjective — discipline, focus, being able to work alongside an army of agents, which from our experience is almost a people-management skill. You need to know how to coach them, how to rein them in, how to verify their work. So maybe the teams are roughly the same size and structure, but the skill set is a bit different. What do you think, Dave?
Dave Welk: I generally agree with everything Graham said. From a structuring standpoint it may look a little different, but in terms of headcount, I think we're about where we need to be to scale and grow our business. I don't think we're adding more people per se, and I don't think we're losing folks either. The real benefit of AI is that you can do more with the same; I don't think we've proven we can do more with less yet — that remains to be seen.
For the next generation: it'll be incumbent on those of us who are more senior to truly create an apprenticeship-type program, where we carve out time to sit down with entry-level folks — or folks who've been in the industry for a couple of years — walk through the model, and understand the logic that Claude or one of the other agents is using. Let me show you what to look out for. We're going to have to get back to that level of training. The work won't be "let's build the model" or "let's create the investment-committee memo" — we're already seeing those processes automated very quickly. It becomes "let's understand the fundamentals": what's important in a particular location, what locational advantages make one investment better than another, the macro and micro risks to a submarket, supply. You move out of the technical skills and more into the pure economics of investing — what's really going to move the needle at the deal level, the city level, the street level. Let's look at the macro forces that will impact our ability to achieve a business plan.
And then operations, as Graham pointed out — you're going to see a huge benefit from AI there. It used to take a long time to pull in underlying data for a property, synthesize it, analyze it, and put together scenario analysis: if we push rents here, what does that do to vacancy? All of that can now be generated in a scenario-analysis environment inside an LLM. The focus on operations, and the ability to get more from your portfolio performance, is going to be a real value-enhancement tool for all of us. As an industry, those who really focused on driving operations over the last four years have distanced themselves, because cap-rate compression is gone. That bailed a lot of firms out — if cap rates went from five to three and a half, you did great, and you didn't really have to be a great operator. None of us are forecasting cap rates going back to three and a half again, so we're really focused on operations.
Raj Singh: Absolutely. And one thing we firmly believe is that these technologies don't smooth out the differences between firms when it comes to their existing operations — they magnify them. To the extent you've invested historically in good data, good systems, and good processes, when you add the power of AI, it just magnifies that advantage.
What they're most excited about next
Raj Singh: Last question for both of you. As you think about where you are on the journey — what you're using this tech for, and what you're not using it for yet — what opportunity excites you most about where this can go, within your firms or your lives? What aren't you doing with AI that you can't wait to do? Graham, I'll start with you.
Graham Russell: I think we're still early — and I wonder how long we'll keep saying that, because it might be a long time. That's probably the most exciting part: if you'd asked me this two years ago, I'd have been excited about things that are standard and table stakes today. So I'm excited for what's to come, while fully aware that I don't know what's coming. Specifically, there's a lot that can be done around connecting data, and we're still in the early innings. We've done a little of it and we're getting better, but there are going to be really interesting insights that come out of our own QuadReal portfolio, and then more broadly in how we connect that into the wider market. There's an enormous amount of information that remains to be connected to drive meaningful insight. It's the type of insight that would change how we'd otherwise look at a decision — something we couldn't have decided without the connection that was created, or the insight an AI model drove.
Raj Singh: It's interesting you landed there, because that's been the biggest problem in real estate investing for as long as I've been at it: the fragmentation of data, which fundamentally stems from the fragmentation of deal flow. So much of what we do doesn't live anywhere central — it's disconnected and siloed. There have been so many attempts over the years to solve for that data fragmentation, and AI may be able to make some inroads. Dave, how about you?
Dave Welk: Similar to Graham — what we're putting a lot of focus on is integration, getting systems to talk to one another. People who are earlier don't always appreciate this, but those further along keep being surprised that there's no out-of-the-box solution off the shelf — no download or subscription where suddenly everything's fully plugged in. I don't know, frankly, that one is even close; it could be years away. So we're making a push week by week to get another system integrated and connected.
I'm excited for when we hit — pick a number — 80% of our systems fully integrated and talking to one another, to see what's possible. I was talking to a colleague last week who sits right behind me; while I was on vacation, he said he'd had the most productive week of his entire professional career, because of the centralized overlay we've built and the amount of information you can pull without sifting through files and folders. It's a prompt: "generate narrative text for me," "do this analysis." It's amazing to watch in real time. What excites me is the time we can reallocate to focusing on strategy and investment-level decisions, instead of producing work — it's evaluating work.
We pride ourselves on being analytical, disciplined, and thoughtful about the investments we make, and we'll be able to spend even more time looking backward — what decisions were incorrect, what about a business plan went wrong. A lot of postmortem evaluation, so we can do it right and improve our prompts and data aggregation on the front end going forward. Our work is going to become much more enriched over time. It's less about collating the information, and even analyzing it, and more about: we have all the information here — what do we do with it? That scenario-analysis work is going to be fun, including for the junior folks. I've been in the business twenty-three years; going back to the early 2000s, I pulled multiple all-nighters building Excel models and programming macros. I tell the analysts I've hired over the years, "you have no idea what that was like." Putting together those 120-page investment-committee memos was a badge of honor. That's gone — there's no need for them to spend that time anymore, even if they wanted to. So quality of life gets better, and they'll spend more time in the trenches on investment decisions earlier in their careers. I'm excited for that. Our work is going to become a lot more interesting.
Closing thoughts
Raj Singh: That's a very positive way to end. For all the doom and gloom out there about how this will impact junior folks especially, I think you hit the nail on the head — there's a lot to be excited and optimistic about. The ability to do more value-add, more interesting work early in your career, and avoid the drudgery a lot of us had to do. The thrill that comes from crushing a week's worth of work in half a day — this stuff can make work a lot of fun. So we'll leave it there. Graham, Dave — thank you both very much. Maybe we'll check in in six months and see where we're at. I'm sure there'll be more lessons to share, and more things we weren't expecting will probably have happened by then. Thank you both very much.
Graham Russell: Thank you, Raj.
Dave Welk: Thank you, Raj. Good being on.
.webp)
