Donal Warde spent nine years in institutional real estate investing before moving into real estate technology and, more recently, advisory work across investment analytics, capital formation and operations. This year, he’s bringing that perspective to the Blueprint stage as moderator of a panel on data-driven investment decision-making. The founder and principal of Folio Strategy Partners will convene leaders from Two Sigma, BGO, GIC and StepStone Real Estate. It’s a lineup spanning quantitative investing, data science, real estate portfolio management, and the institutional allocator perspective.
Ahead of the panel, Warde sat down with Insights by Blueprint to talk through where artificial intelligence is actually moving the needle in real estate investing, as opposed to where the industry just thinks it is. He breaks the opportunity into two buckets and argues that most operators misjudge where the real work happens: not in the first AI-generated answer, but in the iteration that follows.
Warde also walks through his four-level framework for data maturity, from spreadsheet-driven intuition to proprietary, AI-informed competitive advantage, and makes the case that most firms are stuck at level two. Along the way, he shares a hands-on example from his own work using a dataset of more than 270,000 hotel reviews to build a review-informed investment screen, and explains why he now spends more time telling operators not to build their own AI tools than helping them build.
Warde will moderate “Data-Driven Real Estate: Turning Information into an Investment Edge” at Blueprint on September 22, 2026, at 3:40 p.m., and anyone interested in the latest trends in data-driven real estate investment should consider making the trip to Las Vegas to hear what he has to say.
NP: Tell me about the panel you’ll be moderating at Blueprint. What are you looking forward to, and what are you hoping the audience will learn?
DW: I’m excited about it because we’re going to have leaders in real estate data and practitioners from both the technical side and the investment and operations side. Attendees are going to get a sense of what best-in-class analytics and data-driven investment and portfolio management decision-making look like at some of the largest investors in the market.
We have Drew Conway from Two Sigma, who leads data science for private investments, and Chris Liedtke, BGO’s chief data scientist. Ali Haidar, a leader in real estate portfolio management at GIC, and Nicholas Russell from StepStone Real Estate can speak to what institutional investors expect from managers, including minimum data requirements and best practices. Data inconsistency has been a real issue across the industry for years. Nicholas will also discuss REDI, an investor-led data model designed to make GP-to-LP real estate reporting more consistent.
So attendees should get a good sense of where the top players in the field are and where the trends are headed, but also — hopefully — some accessible, actionable takeaways they can bring home regardless of firm size.
NP: Tell me more about how artificial intelligence is changing the work you do in investments. It can sound vague when people talk about “AI in real estate.” How, specifically, is it changing things?
DW: I think about it in two buckets: data availability on one side, and analysis and decision-making on the other.
On data availability, the collapse in the cost of processing is enabling an explosion in unstructured data use for investment decisions. Data sources that were previously too expensive to analyze, or simply unavailable — things like hotel reviews, or permitting documents from local municipalities — can now be processed and made available for decision-making. That’s new potential alpha, new potential competitive advantage.
On the analysis side, the cost of intelligence itself has fallen dramatically, especially after Anthropic’s Claude Opus model was released late last year. That felt like a watershed moment for the processing capability of these systems. Being able to run analysis on both existing and newly available datasets gives firms a much more granular level of decision-making and a much deeper understanding of what’s happening at the portfolio and property level.
NP: Are there misunderstandings in the investment community about how AI is actually changing things? Things people can’t quite wrap their heads around?
DW: Yes, on both sides. First, a lot of AI users aren’t using the most capable models, or even paid tiers. Many people using AI at all are still using free models. So there’s still a small subset of people who really understand what these systems are capable of at the frontier.
Second, among people rolling out AI initiatives, there’s a misunderstanding of where the actual value sits. These models can produce an answer almost immediately, almost to a fault, so people assume the hard part is the prompt or the first response. It isn’t. The real work is in the iteration and how many rounds it takes to actually hit your goal.
People who are bullish on AI tend to discount how many iterations it takes to get a process working. That’s where the time goes, and that’s where any real investment ends up being spent, because the cost of the initial setup is now very low. It’s the iteration and validation that take time.
NP: Can you give specific examples of how you’ve used AI in your data and investment consulting work? Something that’s cost-effective and transferable to other investors?
DW: I use “beta” as shorthand for productivity tools, mostly proven SaaS products that improve efficiency, and “alpha” for competitive advantage, which is closer to R&D. There’s much more market demand right now for beta, because it’s proven and the time and cost savings are real and easy to hit. But the true edge is on the alpha side.
One productivity example: an owner-operator in New York City with a portfolio of roughly 2,000 units was struggling with leasing staff spending too much time on email response and scheduling during peak leasing season. We used AI to help create structured email sequences — not necessarily a fully autonomous AI tool, since some operators don’t want that — plus dashboards to track agent capacity and make sure nothing fell through the cracks during peak periods. Those changes improved speed to first contact by 61% and removed hours of weekly manual work.
There’s also a parallel, more fully agentic version of this happening across the industry now. Companies like EliseAI are handling prospect communication from first contact through the in-person viewing, and it’s a proven space with real adoption. My advice in general is: don’t build agentic tools yourselves, especially anything client-facing. Client-facing AI carries a much higher risk profile than internal tools — fair housing exposure, hallucination risk — so outsourcing that execution to a specialized third party makes sense.
The other example is more on the alpha side. For a Columbia Business School guest lecture a few months ago, I worked from roughly 272,000 TripAdvisor reviews across 470 Miami hotels (~80% of the market), cleaning the dataset to about 82,500 reviews for the final analysis. We turned that unstructured text into structured data using AI. That kind of processing simply wasn’t feasible for humans to do at that scale. The ROI wasn’t there before, but for AI tools it’s a great use case. The real value comes in the second step: turning the structured signal into actionable recommendations, which is where domain expertise matters — someone who can quickly fact-check the signals and judge which ones actually matter.
The output was an investment ranking and screener built from those signals. About 57% of negative reviews were primarily about operational issues — things on-site staff can fix, like communication or cleanliness. A second bucket was capex issues — things that require a contractor and a capital budget. A third bucket was structural issues that can’t be changed at all — being above a nightclub, or the views. That triage helps operators understand where the ROI actually sits, and it can be personalized: a vertically integrated operator with in-house contractors might lean harder into the capex signal, while a more operationally focused firm leans into the opex signal.
NP: The hotel review example seems very applicable to multifamily for reviewing unstructured data to find strengths and weaknesses. If an operator wanted to pursue something similar, what advice would you give them?
DW: First, you don’t need a technical background to execute this. I’m not an engineer, and I did the hotel review project myself. What you need is some domain expertise. Someone who can quickly gut-check the output, because these models will confidently produce answers regardless of whether they’re right, and without domain knowledge you can’t tell the difference.
Start with a pilot. Run the full pipeline on a small subset — one, two, three properties — before scaling. Step one is finding a data source, which is sometimes the hardest part; scraping can be difficult, but a new category of data marketplaces has emerged over the past year or two that aggregate this kind of information. From there, run it through the full pipeline to see where the bottlenecks are and whether the signal is strong enough to justify doing this at scale.
The technical barrier has collapsed dramatically over the last six to twelve months. The question isn’t “can I do this” anymore — it’s “should I.” And that’s a different question, because you’re not just doing this once. You have to maintain it, because the signal value decays with time. Any analysis is only as current as the data behind it.
NP: What are some cost-effective ways smaller managers can build institutional-grade data infrastructure without a big budget or platform?
DW: I think about data maturity in four levels. Level one is the historical default for real estate: spreadsheets, gut decisions, fragmented data, intuition-based decisions. That’s been the industry standard for decades, and it can work fine — if it’s working for a firm, there’s no requirement to change.
Level two is where a lot of firms have landed over the past year or two: data is available and turned into structured dashboards, largely driven by minimum investor reporting requirements rather than a genuine intelligence-gathering exercise. The challenge is that it doesn’t change behavior — it’s ad hoc, checked only when someone remembers to look or a question comes up.
Level three is repeatable analysis with a real cadence, where data and analysis feed directly into the property management and investment feedback loop — an essential, integrated part of decision-making rather than an occasional check-in. Most firms are trying to get here and are currently stuck at level two.
Level four is where unstructured data drives proprietary investment and operational insight that the broader market doesn’t have — a real, durable competitive advantage that merges quantitative decision-making with deep industry experience.
The interesting thing is that cost isn’t really the barrier across any of these levels. Level one is essentially free. Levels two and three use structured data already sitting in a PMS or system of record, so there’s not a lot of new technology spend involved. Even level four isn’t primarily a technology cost. The real blockers are governance, people management, and culture — for many firms, technology is no longer the main constraint. Now it’s about orienting the organization, which is a different problem entirely.
NP: Other than the panel you’re moderating, what are you most looking forward to at Blueprint this year?
DW: Seeing what “ordinary” users and everyday operators are doing with the technology. I want real user experiences and creative ways people are managing ops or analytics that I haven’t thought of or ever considered. Comparing notes with practitioners is going to be useful from an ideation standpoint, because the range of what’s possible right now is enormous, and I’ve inevitably missed things.
The other area I’m always interested in is build versus buy. Almost anyone can build these solutions now — that’s not really the question anymore. The question is whether they should. After five years running early-stage tech companies, I’ve become the person telling people not to build, even though that arguably works against my own interests. In most cases, it’s the wrong call: firms should focus on their actual unique value proposition and only build what reinforces that, rather than chasing interesting side projects. Focus is already scarce, and the temptation to build something just because it’s now technically possible is only going to get stronger.
I’m also very interested in the AI Use Cases panel on Thursday, September 24, featuring John Davis, Assistant Director of Process & Operations at Orsid. The panel represents a mix of vendor, third-party, and build-it-yourself approaches, so it should give people a useful spectrum of options.
John is a great build-it-yourself example. He had no technical background and was working in an admin role when he thought there had to be a better way. He taught himself Microsoft Power Automate, automated part of his own job, and showed his boss. That turned into a mandate to do it across the company, and he now leads much of its automation and analytics work.
Even then, he’s leveraging widely available tools rather than building technology from scratch. His story shows how early we still are in these automation cycles. A lot of it is basic admin work that arguably shouldn’t be done manually at this point. There were no layoffs involved. The company is growing, doing more with its existing team, and the staff is generally happy because you’re removing admin work.
– Nick Pipitone
