AI for CRE: The Bleeding Edge

Introduction

0:00Great to have you all here. Great to see familiar faces. So we're going to get started.

0:08You keep hearing AI is revolutionary. You keep hearing about everything that's happening with AI, and everybody's so blown away by what it can do. But from my conversations with people in commercial real estate, it's not really clear what that revolution looks like. So that is what I'm going to try to answer in today's webinar.

0:32A few words of introduction. I'm Or, I'm the founder of Trophy. Trophy focuses on AI systems for Manhattan office firms. Some of you know me already, but I'm going to give you a bit of background on how we got to this point.

0:49About a year ago, Trophy was just like your regular business, with traditional operations, and our focus was property marketing and office marketing. And then something happened. That thing was the advancement in AI, which was dramatic for anybody who's in the weeds. We're at an inflection point, in the last few months, six months more or less. As things were happening and I was hearing more and more about it, also coming from a tech background, I started tinkering and building internal tools for Trophy. As this became more and more meaningful for us, I decided to pivot the company and start working on these AI systems and automations for our clients, for Manhattan office firms. And today we're at the point where I can confidently say that we are AI native. You'll see what that means today.

1:50Before we start, here's what you won't get today. I'm not going to show you any special tools. I'm not going to give you magic prompts. I'm not going to explain how to do things. And I'm not going to go into technical details, by definition, because I want you to just get a sense of what's possible and how far this thing can go. To do that, we're going to have to skip a lot of details, and we're going to go into those details either in conversations or in the next webinars.

2:24Here's our agenda for today. We're going to start with before and now: how Trophy used to work and how it works now. We're going to walk through four examples of how we operate day to day. We're going to talk about knowledge management, how it's done and what's important. We're going to talk about collaboration with AI. We're going to talk about the relationship management system that we run, and about data collection, which is also very meaningful, especially for real estate. At the end, we're going to cover the takeaways and recap why this is so important. And then we'll have time for questions, so if you have any, try to write them down and save them for the end.

Before and now: how we work

3:10Okay, so before and now. This is how Trophy used to work, the list you see on the left, and I think most of you work in a similar manner. You have disorganized spreadsheets. You have your inbox, which is full of stuff. You have your CRM. You make notes on your phone, in your notebook, whatever. You have your calendar. You manage some freelancers or offshore people to do some manual tasks for you. And even though you have all these things, you still end up doing manual, repetitive tasks, which you know are not your highest leverage. They're not the most valuable things you can do, but you just don't have anybody to give them to. That was where Trophy was before as well.

3:55Today, this is more or less what it looks like. We have Claude in the middle. In this example we work with Claude Code, but there are many tools that can function as this kind of brain of the organization. Everything that we used to use, and that you use today, is now connected to Claude Code. So instead of going to all these different tools and platforms, I funnel everything through that AI tool, and that is incredibly powerful.

4:26The other two things you see on the top and bottom: Claude obviously has web search, so any task I do that needs information from the web, it gets it. And then automations, which Claude both builds and runs for us, which is very powerful. We're going to review a few of those today.

4:46So let's go to the real examples. Even though these are Trophy's actual operations, we can't really show the details, because they hold our clients' information and we respect their privacy. So we're going to look at diagrams. I think that's also going to help you understand, and simplify things a little bit.

Example 1: Knowledge

5:04Example number one is knowledge. This is where knowledge lived before: a lot of stuff is in your head, like it was in my head; the inbox; the notebook; and messy drives and folders, which everybody deals with. Where did I save that? What's the name of that file?

5:26Now, with AI, Trophy has a very comprehensive knowledge base. We have call transcripts of almost all of the conversations that we have, which is very useful. We have emails that Claude can easily read. And we have our CRM. That's more or less where all of our knowledge is.

Company knowledge

5:46Now let's talk about company knowledge. The problem is that to do a good job for your company, to be useful, AI needs to know a lot of things. If you try to recap everything you know about your business, about your role, about your clients, there's a lot of information in there. So for AI to be useful and to act as your aide, it has to know a lot.

6:15However, most AI tools, the consumer tools so to speak (Claude, ChatGPT, the apps you have on your phone), operate on memory. From their conversations with you, they pick up different facts and store them as very shallow pieces of information. So maybe you do something in Midtown East, and it's going to remember "Dave did a lease in Midtown East." It's very disorganized. It's not deep. And it's hard to read back when you actually need it, the next time you do something similar.

6:54The solution is to create a full company knowledge base. It's not something you do in one day. It gets built up as you work more and more with AI, and in the end, if you do it right, you have this very comprehensive asset that makes AI as smart and as knowledgeable about your company as you are, more or less. Obviously there are edge cases, but it's going to know everything it needs to, and what it doesn't know, you're going to teach it. So it becomes an asset for the company.

7:25This is a whole topic in itself, and I decided not to go into it; we're probably going to do an entire webinar on company knowledge bases. But it's critical for getting the results that are possible to achieve with AI.

Meeting prep in three minutes

7:39So now let's look at things we do with that knowledge, starting with methods. Part of what AI needs to know to be useful is the method: how we do certain processes. Let's talk about meeting prep. Every time you meet somebody, you want to know a lot about your past interactions, who the person is, and so on. Claude has a skill that lets me do meeting prep very fast and very efficiently.

8:18Let's say I'm meeting a person called David Cohen. I have the method in my knowledge base, so I just say, "meeting prep for David Cohen." This is what Claude Code goes and does. It reads our past conversations and our emails. It researches that person and their background, if we don't have that information yet. It asks me what I want to talk about, and I tell it: I want to talk about A, B, C, D, and at the end it's important that we don't forget E. Then it writes the notes for me in the format I like. Not AI slop, tons of paragraphs I can't follow. I like a very certain format, and it writes exactly the way I like it, every single time.

8:55And at the end, another very important step: it fact-checks its own work, or an external model checks the facts, because I don't want to come into a meeting with the wrong number or the wrong address and make a fool of myself. Then I have the meeting notes, done for me. Imagine I just throw in this prompt, "meeting prep for David Cohen," go do my stuff, and three minutes later the notes are ready. I can do tons of stuff in between.

What did John from CBRE ask for?

9:25The next example is internal knowledge and conversations. I mentioned we have transcripts of our conversations and we have emails. That gives me a perfect memory of anything that ever happened in the company. Imagine I have this person, John from CBRE. I spoke to him a couple of times and I want to know where things stand. Without AI, I would have to read through email threads, try to remember, find my notebook from that period. But being AI native, I just go to Claude Code and say, "What did John from CBRE ask for in July, and did we follow up?"

10:09Claude has a process it runs through, and it has access to all the systems, so it gets perfect information. It finds John in the CRM (if there are a few people, it asks me). It finds the meetings with that specific person and reads through them. It finds out what we agreed on, then checks whether I followed up and what happened since, and then it summarizes and compares everything. So I get this answer: "John asked for a case study for a big pitch in the fall, and we sent it. He then said the pitch got delayed to November. Now is a good time to follow up."

10:47I give one simple-looking instruction, which requires a few different sources, and I get a perfect answer. I never miss something somebody said. I don't have to sit down and make notes every time I talk to somebody. Instead of trying to note down every fact, name, building, request and decision from your conversations, it's all just there for the taking, always available. That's very useful.

Which buildings sold for over $10 million, and who do I know there?

11:27Now let's talk about something more complicated. Sometimes you have questions that need even more sources. Let's say I want to know this: which Manhattan office buildings sold for over $10 million in the last six months, and am I connected to anybody there on LinkedIn? That's not something you could just do in your regular Claude, and it would be very hard to do from memory.

11:54So I give this to Claude Code. It finds all the sales that happened and filters for the ones I care about, in Manhattan, over $10 million, searching our market database (we'll talk at the end about internal databases and why they're so powerful, especially for real estate). Then it finds all the people on those sales. Knowing about a deed that changed hands is different from knowing who did the deal. It goes through our broker graph, an internal Trophy database, and finds the relevant people. Then it searches the web again, to widen the net and check some facts. And at the end it goes to the CRM and my LinkedIn and checks who I'm connected to in this group of people.

12:44So I throw in this pretty complicated question, and I get a list of all the deals, the numbers and my connections. You can see how far this could go, how advanced your requests could get, while you still get very clear, factual answers.

Example 2: Collaboration

13:11Okay, so that was the knowledge part. Now I'm going to talk about collaboration, which is another very powerful aspect of AI when you have it set up correctly. Let's say you have a rough idea. You woke up with it, or it came to you in the shower or at the gym. But it's just a rough piece of clay, with a million details you need to figure out. If you try to do it, you're probably going to go through a few rounds of trial and error, try different things, and it's going to take you a long time to get it right. And you have to do all of it yourself, assuming you're not yet proficient with AI.

13:51How Trophy operates is that Claude is a brainstorming partner, to think together and figure out what we want to do. It's also an execution partner, because it's connected to all these different tools: when I'm done deciding what I want to do, it just goes and does it. And, very importantly, and this touches on the knowledge base idea, the method for that particular task is saved for the next time you want to do it. That is how the knowledge base expands and AI becomes more and more useful for you every day.

From a rough idea to a webinar

14:36Let's look at an example. Let's say I have this idea, only theoretical: I want to run a webinar on AI for commercial real estate. I would go to Claude, give it this kernel of an idea, and then we would have a conversation, assuming it has the knowledge and I'm able to steer it to the right points. It could go straight to creating a webinar for me, but I know there are a lot of unresolved questions about that kernel of an idea. So Claude and I discuss the strategy. Who is this for? What am I trying to get out of it? How long should it be? When do I want to do it? There are so many aspects AI helps me figure out in conversation, like an advisor. But unlike an advisor, at the end it can go and execute, and actually do the work for me.

15:40This webinar, what we're doing today, is exactly the result of that kind of collaboration. Obviously it's more complex than that, with more than one conversation, but you see how it works. In the end it created the deck for me, then we refined it, and I was able to collaborate with it.

15:56Another important point, again, is the documentation. Once we do one webinar (and we're doing this right now for Trophy), we end up with knowledge files on how we do webinars and how we distribute them. It's not enough to just create the webinar; you want to get it out there. Do I do an email campaign? Do I post on LinkedIn? Do I talk to my friends? And how we design slides. So every part of the work, done right, becomes a piece of knowledge, and the next time I want to do it, it's so much faster and easier. In the end you get to the point where, like with the meeting prep, the method is crystal clear. You press go, and so many things happen exactly the way you wanted.

Executing a documented task

16:59Now let's talk about executing a documented task: something where Claude has all the information and I just want it to do a lot of work for me. Here's an example. We have 39 West 37th, a building we work on with our friends from Noah & Co. Let's say I get an email from my friend Ben, the leasing broker, with a list of availability updates: this floor leaves, this floor we split up, we're going to do this kind of build-out. Before AI, I would have to go through the list, work out what changed compared to the current situation, then go to the website and update it by hand.

17:44Instead, since Claude has access to my inbox, to our information on the building, and to the website (which we built with AI, and which can be edited with AI), I just go to Claude and say, "Read Ben's email and update the availabilities."

17:59What does Claude do? It reads the email and checks our methods for how we update: what information goes on the website. Maybe Ben slipped something into the email that we don't usually publish, like the rent. So it makes sure everything is correct. Then it goes to our internal records and updates the availabilities for 39 West 37th, so we have one source of truth. Then it updates the website, and the broker folder, which has all the information (maybe a floor plan changed, so we want to take it off). It has perfect knowledge of how this task should be done, and it just executes. Something that would have been very complicated, very long and very manual becomes almost fully automatic, because Claude is connected to all these systems and has all this knowledge.

Example 3: Relationship management

19:05Let's talk about the third example, something a bit more specific: relationship management. We have a few brokers here. You're all about relationships, about managing the people you interact with: your clients, your landlords. You've got to manage a lot of relationships.

19:22This is how most people do it today. They have a mental list of contacts, or maybe a CRM, but a basic one, not comprehensive. They have a memory of what they need to talk about with each person and what they've talked about in the past. And they just try to remember: "Oh, I haven't talked to John in three months, maybe I'll shoot him an email." The result is that relationships drop. You forget about somebody, maybe it's been a year, so you shoot them a note, but it's gone cold. It's very hard; you have so many things on your mind.

19:58This is how Trophy works now. All of our contacts are tracked, and all of our conversations are logged, in the CRM, the inbox or the knowledge base. And we have a system that is systematic and automatic, so nothing ever falls through.

The follow-up system

20:19This is what it looks like. Since we track all of our connections, the system can easily tell how long it's been since we spoke to any single person. Let's say again I have this person, David Cohen, that I haven't spoken to in 51 days, and my system is set to remind me after 50 days. It finds the people we need to talk to today, the ones past that period. For each person, it works out what our relationship is like, what we've talked about in the past, when we last spoke and what that conversation was about. So it has perfect context about the person and our relationship. Let's say I spoke to David Cohen on May 28, and we talked about his portfolio and about asset management; maybe they have some challenges managing all of their assets.

21:16I don't want to just send him a note, "Hey David, what's up?" I want to be helpful. I want to be useful. I want to add value. So the next step is to match the person with a piece of content that might interest them. Maybe we posted something on our blog, or there's an article that might be interesting to them. Our system goes through a library of content and tries to find a match for that person, checking whether they've seen it before. In this example, the system suggests sending him the blog post we just published, "Top five use cases for AI in office asset management." Makes sense: if they're struggling with asset management, maybe this could help.

22:06The next step is to write the email. I could do it by hand, but I prefer AI to help me, so the next step is to draft an email to David: "David, hope leasing is good." Again, I have perfect context; maybe he told me about 227 Broadway. I want to make this personal, and make it feel like an actual email, not mass spam. "You mentioned... here's a link."

22:39This drafting step might seem simple, but in reality it's pretty hard. Once you have it set up right, with writing guidelines for how I want to sound, how I write emails, what I do say and what I never say, plus a few subsystems that fact-check and make sure I'm not making mistakes, most of the emails turn out very well, very similar to how I would write them. So I have no qualms about sending people AI emails, because they're very directed by me, my personality and how I want to sound.

23:17The last piece is what I personally get every single morning. This card in my dashboard says: you haven't spoken to this person in X days, here's a piece we might want to send them. Then I have options. I can just send it, if I'm happy with the email and nothing happened in the meantime that the system isn't aware of (maybe I saw something in the news that changed my mind). I can snooze it. Or I can edit it, and it appears in my inbox ready to go, and maybe I change a few details. Following up with people, being useful, maintaining relationships, can take 15, 20, 30 minutes per person. It becomes a two- or three-minute exercise I run through at the beginning of my day. It saves a bunch of time and makes sure no relationship ever gets missed.

Example 4: Data collection

24:27Let's talk about data collection, our last example for today: market signals and information. We're all in real estate, and real estate is an information game. Negotiations are all about information. You want to know about things when they happen, you want to know as much as possible, you want comps. I don't have to tell you all that. But how do you get that information today?

25:03This is what I used to do. I would read a bunch of newsletters every day and scan them for what's interesting to me. I would scroll LinkedIn looking for the pieces of information relevant to my business. I would log information by hand, a note to myself: "455 Fifth just sold, reach out to that company." And I would miss a ton of opportunities. That's what it was like before, and maybe how it is for you today. Obviously there's CoStar and other databases that can bring you that information, and we'll talk about those in a second.

25:45This is what it looks like for Trophy today. We have a private database, which is basically a set of online spreadsheets that automations can read and write. We have all of these data sources: Traded, Commercial Observer, ACRIS (which is public), the Department of Buildings, newsletters, LinkedIn and many more, that we're able to query and scan daily, collecting the information we care about. So nothing gets missed, and we're constantly growing this asset for the company.

26:20Let's look at examples. For building sales, if that's what you care about, there's ACRIS, Traded, Commercial Observer and other sources. This is what we track ourselves: a database of sales in Manhattan, which is where we work, and at any point, like we saw before, we can go through it and pull out what we care about. There are leases, obviously, that you can track: what lease, where, sizes, rents, and also who's on the deal, through the broker graph I mentioned. I can look at all the deals of the last month and see which brokers are the most active. There's a ton of information you can pull from those.

27:11In Trophy's case it's not so much our focus today, but in the past renovations were a big thing for us. When our business was marketing office properties, the biggest marketing push would come after a big renovation or a sale. So you can see how I can go to this cloud of information online, pull exactly what I care about, and save it for whatever use I may have later.

External databases vs. a private database

27:42You probably have CoStar; there's CompStak; there are lots of data products for real estate. Here's my opinion of the differences between these external databases and a private database. With an external database, you get the same data as everyone else. You usually have to search yourself. There are features like automatic reminders; I haven't heard of a lot of people getting much use out of these, but maybe you do. And whatever data you find stays in their product. They own it, which has a lot of implications.

28:28With a private database, if you're able to set one up, you control all the sources. You're not dependent on their sources or where they get their information, and you have a bigger menu to choose from. It's all linked to your CRM and your other systems, so you can always compare and contrast and make complex queries. And what we're going to look at now: these private databases can also trigger action when things happen.

The renovation trigger

28:58For example, let's say I'm an owner with a portfolio, and I want to know what my competition is doing. I want to create an automation where any time a building within three blocks of my portfolio files for a major renovation, something happens. That event triggers an AI automation that does deep research on the renovation. What are they trying to do? How much are they going to spend? What kind of amenities are they going to create? It gathers information on the event I care about, and creates and sends a detailed report to me and my team. You can see how valuable it is to know everything you can about an event right when it happens, or right when it's published, as opposed to weeks or months later, and being behind your competition.

The impact of AI and automations

30:12Okay, so those are our four examples. Let's talk about the impact of these changes and these technologies that are becoming available. First, ideally, you want to go as much as possible from doing to thinking. Instead of doing manual tasks every day, you want to spend more and more time thinking, making decisions and working out the strategy, as opposed to hacking away at things you'd prefer not to do.

30:47Second, you move faster. If you have a big team and every person benefits from this implementation of AI and automations, every employee is going to be more useful and get more work done. That has a big impact on your company.

31:05You're going to save time and money on outsourcing the tasks you used to pay people overseas to do, which AI can do now. And it's not only the actual cost of the freelancer; it's also your time managing them. Maybe you're not happy, it takes a few iterations. With AI, everything is very fast and very efficient.

31:26Then, and this is the database example, you're going to have more information. You're going to see more opportunities and be smarter when you pursue them.

31:39And the last piece: if you go all in on this, if you make it your goal to implement these technologies, you're going to get ahead of your competition. I'm not trying to scare anybody, but I genuinely feel this is happening. This wave, this tsunami of change, is coming, and it's much better to get on it than to do nothing, and much better than going against it: entrenching your manual processes and hiring more and more freelancers instead of moving to AI. If you ride against that wave while everybody else gets on it, you're going to find yourself behind. So I really urge you to look more and more into this.

Recap and more use cases

32:33So here's what we reviewed today. We talked about knowledge: the knowledge base you need to build, the methods, and how you can access your conversations and have a perfect memory. Collaboration with AI, on new tasks and on documented tasks, and what that looks like. How you can manage your relationships with AI and automations. And how you can collect data, and why that's useful for you. These examples are just the tip of the iceberg. There is so much you can do, and I urge you to find your own examples and see how you can implement AI in your business.

33:11Here are some more use cases we're looking at; not a comprehensive list. Again, this is just a small example of what this could look like, and you have to figure out what it should look like for you. For everybody here, everybody watching: I would love to talk about your business and see how you can implement more AI and benefit from it. Even if you do it yourself, I can help you figure out where you should be looking and what you should be trying right now.

Thank you

33:45So thank you, everybody, for listening and for joining. Feel free to reach out and ask any question you want; we can find a time to chat. There are going to be many more webinars on more specific topics, so stay tuned. I'll make sure you get the invites to the next ones, and we're also going to share the recording of this webinar, so you can go back to the parts you care about. So again, thank you, and I'll see you next time.