AI After Dark

00;00;00;00 - 00;00;23;11
Unknown
All right. Patrick, what is managing Director head of Innovation mean? What is innovation mean for? For a real estate fund like Griffin Partners? What does innovation mean? Yeah. So I think we we spend a lot actually a fair amount of time probably the past couple of weeks talking about this. And it's it's been really interesting because of all times when you say innovation, everyone automatically assumes tech technology.

00;00;23;13 - 00;00;54;11
Unknown
You implement technology, you build digital innovation by technology. Yeah, it's all called pushing that part of it in the future. And the thing that we've been trying to communicate and, and tell the story of a little bit better than we have in the past, is that innovation is really a people focus thing. It's really around call it the way that you communicate, the way that you change as a company, the way that you call it adapts to new situations.

00;00;54;13 - 00;01;18;01
Unknown
And technology is a major part of that because in the world we live in, right. You can't go anywhere without seeing AI, right? And that's a big part of what we're adapting today and adopting today. But, but that that didn't just start with AI, right? We've been innovating by. I mean, shoot, 20 years ago, SAS was an innovation, right?

00;01;18;04 - 00;01;36;09
Unknown
And so it's like, really the focus of innovation is kind of on how we adapt to change within our organization, how we make space for people to be able to do that. Because making space is a lot easier to change if you if you don't have any time, like naturally, it's like not going to deal with that thing.

00;01;36;16 - 00;01;50;07
Unknown
Sure. You want to give me Claude? I don't really care right now because my hair's on fire in ten different ways. It's like the no cloud can actually fix a lot of like, streamline or make things a lot of fat a lot faster for you. Yeah, but I had this argument with my wife all the time. Yeah, I don't have time to deal with it.

00;01;50;07 - 00;02;07;04
Unknown
I don't have time to learn it. It's like, well, it's really not that complicated to learn, but there's no space, right? So it's a how do you like organizationally create space for that change and that that's the biggest part of what I think. When you talk about what innovation is, it's that, oh man, you're going to be my kid's therapist.

00;02;07;07 - 00;02;27;04
Unknown
Okay. So so how do you in your organization, how do you create space for change? So the easiest answer is it has to start at the top. And I don't mean just like call it the CEO or anything. We've been I've been fortunate at Griffin, that we've gotten to go through kind of a changing of the guard.

00;02;27;06 - 00;02;52;02
Unknown
We've had many people call it and count 2 or 3 people that were well tenured there 15 plus years at Griffin that had retired, call it in my first year or two being there. And so with that, you saw people come in with a different perspective. And so in terms of that, there's a team of eight of us, that really, really work together.

00;02;52;02 - 00;03;09;27
Unknown
And naturally, I've been there five years, so that that team has gotten closer and closer and closer as we've done more of this and as you've gotten to see more of it take effect. But really the the buy in as you start doing some of these things from that team is the way that you create it, because you can't create it unilaterally.

00;03;10;00 - 00;03;34;09
Unknown
Right? And you can't create it by mandating by mandate because that never works. But partnering and serving each group and each executive in a little way, right, kind of starts the snowball rolling right? Hey, let's do this little thing here just to make this little thing better and start to create some competition amongst the people. Like this pure healthy competition type psychology.

00;03;34;09 - 00;03;58;15
Unknown
Or is it is there a different layer? No, I think it's I don't think it's really call it competitively driven. I think it's called everyone's got their hair on fire because there's I mean any sort of business that is not call it fortune 500, right. There are more problems than there are solutions, right? And they're constant problems. And that's why they have a whole division of operations.

00;03;58;15 - 00;04;20;22
Unknown
Which operations is just all the stuff that's going on that people have to deal with, right. And so when you get into like kind of the the different operational things that are going on, how can you give someone a small thing that will actually serve them and then allow them some space? Because by serving people, right, there's always the values of reciprocity, right.

00;04;20;27 - 00;04;40;17
Unknown
You do something, help somebody else out. Well, naturally they'll be open to listen the next time you come around and say, hey, why don't you just try this too? It may be horrible. May make your life awful. And if it does, just tell me it will be on upfront and honest about it. But that's where you go back to, like, it's the space in the communication and the comfort to fail.

00;04;40;19 - 00;05;11;21
Unknown
Because like luckily, coming out of the more startup world, we're a lot more comfortable with failure. When you move into court, private equity or anything where the results drive a lot of what we're doing is we've got to get results for investors, gotta get results for partners, things like that. Those results don't just kind of happen, right? They happen because a lot of little things that have been done and everyone's so focused on that, that it's tough to be like, hey, just just set that aside for like five minutes, right?

00;05;11;23 - 00;05;26;04
Unknown
I know, I know, you got to get this out by the end of the week, and it's Thursday night and you're really worried about what's gonna happen Friday. But just like, can you wait for a second? Try this one thing? Yeah. And it's a lot easier to do that if you're like, okay, well here we did this one small thing and it did make your life easier and listed make your life easier.

00;05;26;04 - 00;05;49;18
Unknown
And that that's the snowball, right? That starts to grow because that space can't be created right. Without that, without proof. That's why I wanted to talk to you. Because you're you're the first person we've had on the show that isn't a true startup. Like a tech startup, right? Yeah. And so you're dealing with a different profile of individual in terms of how AI gets adopted, how AI gets built, how how you even invest in AI.

00;05;49;20 - 00;06;12;10
Unknown
Yeah, than than any other founder. And, and in fact, one of the founders mentioned, he's like I try to implement like thinking is working my taking time and finding the time, creating the time to think is working. I don't think I've ever experienced that in the corporate environment. Yeah. Ever. I mean, it's it's not something that's common.

00;06;12;12 - 00;06;34;07
Unknown
And because so many organizations, especially as you go corporate, right, top down, take orders from above, and it's not truly entrepreneurial as you would think of like typical effect typical founders would be. And so I think a big portion of the call, it the founders and the people that are entrepreneurial, that understand the time that they spent doing the dishes.

00;06;34;11 - 00;06;52;23
Unknown
Right. To the to the Jeff baseline, right, where your best ideas come from when I'm doing the dishes. But that's time you're spending thinking. Right. And and just kind of saying, all right, like what? What can what can be done about this other thing or in the shower, like all these other times that if you're, when you truly like, say, own something, right?

00;06;52;26 - 00;07;11;13
Unknown
It kind of engulfs a lot of your life. It becomes your little baby kind of thing. And so that doesn't happen in the corporate space because there's not a lot of ownership and accountability like that. True. There is at the top, but there's not at all levels. Do you think you take longer showers as a head of innovation and your peers?

00;07;11;15 - 00;07;41;18
Unknown
My wife would say probably, yeah. There's a that's probably true. My water bill is just astronomical. No, no, I would say probably, I probably should do the dishes more and I'd get more out of it. Yeah, it's probably a better answer. No. That's interesting. So, No, but you're right. I think, allowing yourself the space in, in this, in this case, head space, you know, to be able to kind of think on how to make change, drives the innovation forward.

00;07;41;21 - 00;08;02;18
Unknown
I, I to that end, do you feel like because maybe we're not creating as much change this like I adoption or AI builds are just causing us to create more work ultimately. So yeah, I think I think we're I think we're in the really early innings of it right now. No one really knows how it's going to shake out.

00;08;02;20 - 00;08;23;12
Unknown
I do think, like I was on a really interesting call a day where we were talking about, proptech vendors specifically, and the fact that there's, there's a group of men where we kind of go through, and really it's called the Emerging Manager Tech Council. So it's a bunch of companies roughly the same size we are. It's a good peer group.

00;08;23;14 - 00;08;44;16
Unknown
But as part of that, that group, we track it every Proptech startup we can find. There's over 6000 right now. And it's like, that's great. But is there really that much of a budget in Proptech for 6000 startups like. And they all sort of, I mean, assume there's some stratification but mostly point solutions. They're mostly all point solutions.

00;08;44;16 - 00;09;01;26
Unknown
And, and sure, like even if some of them started to roll up or like there's still too many and it's because it's really easy to build a point solution right now. And so problem is, is you have a point solution. You're like, great, this says this one little thing. Oh, now it has a CRM the next day. Oh now it has this other and you're like this.

00;09;01;26 - 00;09;21;17
Unknown
It's it's just kind of growing unwieldy. And it almost makes you like as a buyer you'd be like, all right, we think we're going to wait and see how all this shakes out, because so many of those point solutions just don't provide enough like leverage to move the needle. They just don't. We were looking at one solution that's an eye for Hvac, right?

00;09;21;17 - 00;09;44;01
Unknown
It's going to constantly tweak everything so that our buildings will be more efficient. We looked at it and we're like, great, let's see how much this would save. 18,000 bucks. Not small like we would. We wouldn't end up paying anything. We'd save 18 grand like it's a great win win. You're saying the net of what you'd pay for the software mind like minus what you're.

00;09;44;02 - 00;10;01;23
Unknown
What we'd save is $18,000. Oh, okay. So? So, yeah, it's a good deal. It should be a good deal for us. But then you turn it on, ask what happens. What's the risk? If, like, our tenants are uncomfortable for a month, as the software figures it out, it's not even close to worth 18 credits, right? Yeah. Like, am I really saving anything?

00;10;01;23 - 00;10;21;04
Unknown
Is it worth the risk? Well, there's going to be 50 of these. There's like, I don't know how many. There's a bunch of these vendors all promising the same kind of thing. We'd be better off waiting till it shakes out and seeing proof, rather than being an early adopter of this. In this instance, we typically like to be the early adopter, but it's not really going to be worth it from a risk reward basis.

00;10;21;10 - 00;10;41;18
Unknown
So if, if, if you guys are typically the early adopters and you're not, what is the profile of real estate fund that does adopt early. So so we are mostly an early adopter on many technologies. My comments more for like the huge landscape of tech that's out there. You have to move the needle so far to even get an early adopter to take the risk.

00;10;41;20 - 00;11;04;04
Unknown
Yeah. Right. Yeah. And I mean, and this is a pretty clear cut ROI calculator, most of these likely aren't the things that what you're looking at. Most of them are coming in the form of our saved, which is the biggest, I don't know, just it's made up. Yeah. It's made up. Not accurate. It's it's arbitrary. But the biggest problem is it's not actually return to me.

00;11;04;04 - 00;11;22;07
Unknown
People treat it like it's a return, but it's like, great. You saved our team ten hours a week. Like, on paper. That sounds awesome. Does that mean we're spending more time on social media? Like we don't save any money from it? There's no actual return. Yeah, yeah, you still pay exactly what you're going to pay. And so it's like, what do we actually return?

00;11;22;07 - 00;11;44;09
Unknown
Well, and we spent more on this product. So net loss right. It's not a return. And so that's where like time served their time saved isn't really isn't really valuable unless you're truly talking about the ability to move the needle at scale. We added ten buildings and didn't hire a person. Different story. Right. And that's where it's got to be.

00;11;44;09 - 00;12;06;06
Unknown
And they're there occasionally, like we've got products that we've used that are like that where it's like, we did this and this was a product process. We were struggling to even complete. Now we have a product that processes now running perfect kind of thing. That works for both products. We've purchased products, we've done internal builds on, right with code and Codex and all these different tools.

00;12;06;11 - 00;12;29;22
Unknown
It's so easy to build now your own internal point solutions that don't really require much maintenance. We're not going to go out. We're not going to build something that we're going to spin out and start selling, but we can build something that, hey, will handle the specific workflow that moves things between teams. We did something about six months ago for our accounting team, and actually got to look at it earlier this week and said, hey, how's this performing?

00;12;29;29 - 00;12;57;17
Unknown
And got to have code kind of build us an analysis tool over the top of the data that we've been capturing for the past six months and go, oh, we actually found out our accounting teams performing better than expected, which is awesome. And so we can actually move the timeline and actually come closer to meeting requirements or be it early on some of our requirements for reports or for different things, we have to report out to them for like loans or or any sort of partner, or joint venture agreements kind of things.

00;12;57;21 - 00;13;18;18
Unknown
How how do you what is your framework for buying versus building versus buying to help you build? Yeah. So in terms of a pure framework, a lot of people like to think of like the call it Gartner quadrants or whatever you call it with like and really I'm only going to build the things that are like unique to us.

00;13;18;20 - 00;13;42;23
Unknown
I tend to think of it a little differently and a little more in flux of I'm only going to buy something that has a clear return because call it the idiot index on software is still obscenely high compared to like what it cost to actually build something now. And that's coming down rapidly. Like we've seen we've had multiple groups quote us, call it half of what we'd be paying a software vendor to build us something.

00;13;42;26 - 00;14;09;06
Unknown
And it's like, well, that's great like that. We can build that and we can use you to maintain it at a lower fee than we would pay a vendor, and we can have our own custom thing. But when you kind of start stacking all those things up, it's like one, I don't have the belief that that will stay around too long, like the longevity of those things or the custom builds, the longevity of them.

00;14;09;08 - 00;14;28;16
Unknown
As costs change for the models. It'll be interesting to know how long that will hold on. And so skeptical on that type of thing for non-technical groups that are doing that. And so tend we tend to kind of say, if we're going to do it, we'll, we'll do that ourselves because we have some tech talent in on the team.

00;14;28;18 - 00;14;57;15
Unknown
And so in terms of building there, there has to be something that that almost makes you go, wow, like the makes me go, I haven't seen this before, or this is going to solve a problem that the not only is it something that's inefficient, but it's something we're just not doing today or we're struggling to get done, and we can do it in a way that doesn't cost our tenants a bunch of extra money, doesn't put a huge burden on corporate, those kinds of things.

00;14;57;17 - 00;15;16;03
Unknown
And so you're seeing that cost come down in software. I know people talk about the SAS, Pocalypse and all that stuff, and not seeing it personally, at least in the space we're in, we're just seeing the pricing models change. And so as the pricing models get more friendly, it becomes easier to to lean more towards the buy stuff.

00;15;16;05 - 00;15;42;00
Unknown
But I don't think of it as rigid as call it. We only build or only buy certain things because everything changes. And right now shoot in two weeks that will be something new, right? And some new development that causes people to shift the way they think about things like even as simple as we rolled out Clod Enterprise, looked at a couple and Claude has been better for our team.

00;15;42;03 - 00;16;01;06
Unknown
Granted, we do a lot more finance type of thing, and so we rolled it out to to 2 or 3 specific teams. So it's not rolled out across the organization. They're looking at it just to determine versus what they were using before. And and kind of seeing how they like it. From the feedback, it seems like it's light years ahead of what we were using.

00;16;01;08 - 00;16;16;28
Unknown
But even in the like, hey, this we're going to do this for 30 days. Well, fable got released, right. Like there's there are these other things that are now coming in as part of that that's like, well, if we'd known that, we might have evaluated this differently, right? If we don't, everyone's going to get a 14 day preview of fable.

00;16;17;05 - 00;16;35;00
Unknown
We'd probably get rid of the whole company and said, yeah, you guys go try this. Yeah, yeah, get as much sunshine. So then are you finding companies trying to build on top of furniture models a little bit more so that when the financial model improves, they also improve? So yes, we've definitely seen some people build on top foundational models.

00;16;35;03 - 00;16;52;10
Unknown
I tend to be more skeptical of those even because they don't control their core business, which means we're not in control of the business. We're not control of what we pay. Because the answer is always oh well it got more expensive chat got more expensive. Right. What were the pass through. And now we're paying a margin on that.

00;16;52;12 - 00;17;09;07
Unknown
And so it makes me less likely to want to work with a group like that. There's some groups in our space that are actually really interesting that are more like, hey, we will, we'll build you something specific, we'll support it, and we'll use the models and it'll just be a pass through. That's far more interesting than a group.

00;17;09;07 - 00;17;31;00
Unknown
That's kind of a wrapper on top of a cloud or a chat or something like that. Yeah, yeah, yeah. Just because I tend tend to be a little skeptical of their data security, if they're a wrapper on top of it and not that they can't secure it. But it tends to be something that we've got too much investor and sensitive data for one of those vendors to get it wrong.

00;17;31;04 - 00;17;54;11
Unknown
Yeah. So how do you compete then with I mean, going back to the viruses built piece, right. If you're saying, hey, I may not buy you, especially if you're a wrapper or if you're reliant on the foundation model, which like candidly, most are and and so therefore, like, I may just want to build it myself because then I have at least a little bit of guidance or control manipulation on, on what my variable cost structure looks like.

00;17;54;18 - 00;18;26;23
Unknown
Yeah. It is. I mean, how does that tie into how you decide whether to buy versus build where your trust points? Yeah, I think one of the biggest trust points is how much you control of your own model and your own data. So there's we've done some, some projects with, with certain groups where we've had to go acquire data sets, some paid, some call it now, I say old fashioned way, but some kind of grinding through an old fashioned way of building our data sets that we couldn't completely put together.

00;18;26;23 - 00;18;43;11
Unknown
And we had to kind of come up with that on our own, is how are we going to how are we going to get at this information to be able to do things the data makes makes it a lot easier to trust and buy when it's like, all right, here's here's the data source in the data set we have, because that's the foundation for sure.

00;18;43;16 - 00;19;08;27
Unknown
You can use the foundational model. You can train your own. And I'm not going to say you're going to get the same results right. But you can still get really good results if the foundational models get prohibitively expensive in the future because you have the foundation. Yeah, yeah, I think the the thing I've seen work well in some of the market portfolio companies have been like, I'm going to I'm going to do the R&D on the latest and greatest models, the fables.

00;19;08;27 - 00;19;26;17
Unknown
But then once I've got it figured out, I'm going to move to Deep Seeker, Cami or something. You know, more of an open source that produces and use a DSP framework. That kind of helps me mitigate some of those costs. Yeah, and I can run it locally, and I don't have to trust data security because it's in a box on a room or it's on a cloud server I control or whatever, whatever else you think.

00;19;26;18 - 00;19;50;26
Unknown
Yeah. And and yeah, the design process, the, the R&D process changes. Yeah. And so I definitely think that the point about call it as, as we look at those foundational models, right. There's a huge subsidy for those foundational models currently. And no like no one has a great handle on what that is going to cost. And some groups are already seeing it.

00;19;50;26 - 00;20;08;22
Unknown
I know, like there was a whole story about Uber. If you are a month or two ago, they blew through their AI budget like months or for months. Yeah, yeah. And just kind of crushed that whole and you're like, all right, well where did all that spend go? And so it's yeah that's true that no article talks about what, what was the ROI of that three four months spend.

00;20;08;23 - 00;20;34;16
Unknown
Yeah. They just said, oh they spent it. And it was kind of you're going through that token vaccine period, right. Where everyone's like bragging about how much just tokens, how many tokens they could use. And it's like, this really isn't a good use of dollars, even not just tokens. And so it's like as you start moving into this world of like, we're going to see, shoot, open it, open a vehicle, open OpenAI and entropic both have like filed for IPO right.

00;20;34;19 - 00;20;53;18
Unknown
It's not going to be too long till suddenly they're going to be trying to make money. And when they, when they start taking actual external funding and they actually have to start being profitable, the cost for those tokens can be wildly different. Yeah. And where is going to be the good enough model that everyone gets comfortable using?

00;20;53;21 - 00;21;11;03
Unknown
But that's going to be a real interesting question. Yeah. In fact, we were just talking about this. It was just like the the pace of the technology is exponentially greater than our ability to know how to use it or learn how to use it. And so we are now the bottleneck, right, of our ability to actually leverage all this.

00;21;11;05 - 00;21;32;00
Unknown
And and so is it this the equation as simple as, oh, this, you know, these the model that we'll use is is aligned with where we're at in our ability to actually execute from that. And obviously from a cost, you know, benefit standpoint. Right. Or is it something where it's like, hey, look, I'm willing to invest, you know, in this failure mode right now, right.

00;21;32;00 - 00;22;02;17
Unknown
Part of the token maxing to be able to prepare myself to for the latest and greatest model constantly. Yeah, I, I don't know if there's a right answer. I don't I think I think that's the experimentation that we're in right now. Yeah. But I, I do find that, like, we're just for not just at a individual level, but to your point, at an organizational level, just not understanding where to even start from an innovation standpoint, and, and then just trapped in analysis paralysis there in between FOMO and this fear of being an idiot.

00;22;02;19 - 00;22;23;20
Unknown
But I do think there's a big part of when we talked earlier about creating space for it, when we like using the cloud roll out as an example. So we started with a team of called 3 or 4 people and we're like, hey, we're going to pick one person from each different business unit and say, go use court like here, just have it to fixed fee license.

00;22;23;20 - 00;22;41;04
Unknown
We're not going to even deal with token usage if you if you run out of your usage, wait six hours and and kind of go that route. They had some problem. Some of them maxed it out, some of them didn't. But then we started getting interesting use cases that no one could have been like, hey, go try this this way and tell me if it works, right?

00;22;41;04 - 00;22;56;28
Unknown
Like lease abstraction is a big deal when we're buying an asset because we want to know if what we're getting, if the data actually lines up with what we said we were going to, what we thought we were going to get right. So it's like we need to figure out what's in these leases. We usually pay a firm to do that.

00;22;57;00 - 00;23;16;05
Unknown
And the last, the last few we'd done, we'd we'd use an AI provider also to say for a minimal cost to say, hey, abstract this. Well, we tried giving this to Claude. This time, and the results from the Claude study was basically better than the other two, right? And it's like, awesome. That's 5000 bucks. We don't need to spend.

00;23;16;05 - 00;23;36;00
Unknown
Suddenly that's better for investor pricing or for the for like the purchase. We're going to spend less on diligence. Really I realize it's marginal in the in the scheme of things. It's not that big of a deal. It's just one thing. Right. And so then we had same analyst that there was going through and working on that was like, hey, I can generate these other documents.

00;23;36;03 - 00;23;52;12
Unknown
Oh, this other set for this other assets, if I look I can generate these. I don't need to pay a lawyer. Oh I don't need to do like I generate these and send them off and like oh I can do and sort of doing this all on his own. Right. That space from sitting in a room be like, I need 30 minutes.

00;23;52;12 - 00;24;11;00
Unknown
We're going to go through and run this little abstract together. Suddenly turned into, I'm going to write here, I'm going to write here, I'm going to write there. That's where you get that snowball effect, right? And saying, all right, now, now I can go do this. And I've generated these, these 40 documents that it was going to take me two days to generate or yada, yada, yada.

00;24;11;00 - 00;24;35;29
Unknown
And it just builds. Right? I'm still stuck on, and maybe I'm feeling pessimistic, but it's like, you know, the financial models continue to just eat up the world, so to speak. Right? And so then, like, how do you build something that's defensible, you know, how do you build something, how do you stand out in that world of, of where, like, progressively more and more people are presumably going to be just saying, hey, I'll just give it to OpenAI, give it to anthropic.

00;24;35;29 - 00;24;54;14
Unknown
Well, yeah. And it'll perform maybe 60, 70, 80%. That's that's fine. That's enough. Well, I think a big part of it is you ultimately have to ask, like what's different? Like the only way you're going to stand out, right, is you got to be different than that, which in my mind is like you have to have the human component.

00;24;54;14 - 00;25;08;27
Unknown
That's a big part of it. And that's why I kind of talk when we asked what the difference is with innovation. Innovation is really a human focused thing. It's a change focus thing. It's it's how do we as humans adopt all of this stuff and make use of it productively, not just is this better tech, right. We've had better tech.

00;25;08;27 - 00;25;30;16
Unknown
Shoot, the Betamax was better tech, right? But like but still it was change manager. Sony had bought change management for it. Yeah yeah yeah. So but but to that I mean to that effect it's like consumer tech scares the shit out of guys. But the human aspect is like one of the big things where, where are you different.

00;25;30;16 - 00;25;54;03
Unknown
Right. Like the human like sorry OpenAI anthropic. They can be as great as they want. Would you trust them with your investment? If you're a pension fund? No. Like in short, because like you like it if they do something wrong. What, like you have no recourse, right? Like there there's this big, huge element of trust. There's the big element.

00;25;54;03 - 00;26;13;27
Unknown
And one of the reasons why I, I switched careers five years ago in terms of like, I was in the energy in software space and, and sorry, the energy software space and switched to called being a company that were focused far more real assets because it's kind of what we saw coming is there's going to be this huge disruption, right.

00;26;14;02 - 00;26;35;04
Unknown
And you can have AI and anthropic do as much as you want, but there's still people with boots on the ground that have to do things right. And so there's going to be a real asset component to things that are successful. And whether that's hardware and AI and hardware, because as much as you want, you can't cover foundational model on a small thing yet, right?

00;26;35;04 - 00;26;59;29
Unknown
I'm sure you will be able to some day, but like today, you're not gonna be able to have that call it in a small, small device on in a physical space somewhere doing something hyper specific and. Right. So that's where I think you also get into if you have to also remember, these foundational models are huge generalists. And like if you take parallels back to like people being really big generalists, it's great that it's made generalists better.

00;27;00;04 - 00;27;21;26
Unknown
But the people that are like, you have someone who's a back surgeon, you're not gonna let them work on your brain. Right. There are so many of those types of parallels where specialists are needed. And these groups may be great at the general, but there's so much opportunity for specialization. Yeah. And for the people that know how to orchestrate between the two and choose, there's so much opportunity.

00;27;21;26 - 00;27;38;01
Unknown
But does that does that line move. Like I feel like over time everyone's going to push it through the to the generalist first to see if it works. Yeah I think feeling like they have to specialize. No. Okay. There is that there's a there is a point where there's just you can't I think it's and I think it's a cost point.

00;27;38;04 - 00;28;03;12
Unknown
Right. Because for as great as Facebook is going to be and as many things it's going to discover and are going to be discovered and why they're restricting access to it because they're scared of what's going to happen if they release it to like potentially people who are bad actors. Right? You you're still only limited by the amount of stuff that it can do at once, and which is far more than like any of the other models have been far more probably than like I can sit there and do like on paper, hands down, right.

00;28;03;15 - 00;28;29;18
Unknown
But you're going to get to a point where, like it is trained on knowledge that exists. It's not trained to find new knowledge, it's trained to find problems with existing knowledge. And sure, can it can it take, can it solve different protein folding issues that would have taken forever to solve? Yeah, sure it can. It can it find bugs in code that I mean, because it understands the rules.

00;28;29;19 - 00;28;50;04
Unknown
Excellent at that. Right. But like, fundamentally these these large limbs, the foundational models are really like predicting what is most likely. Yeah. Yeah. So how does it discover something new. Well and you can fit it. You see what good looks like right. You're saying, hey, I did this in this protein folding situation. This is what it needs to be.

00;28;50;06 - 00;29;13;28
Unknown
Yeah. And I can't figure out how to get from where I'm at today to where it needs to be. Yeah, but you can. Yeah. Try, try 100 billion times. Yes. Yeah, yeah. And do it fast. Yeah yeah yeah. But if you don't know how to explain what it needs to be. Yeah. So like a, like the research, the future state defensible state is one in which you are appreciating, understanding, learning, maybe even guessing at what the thing should be.

00;29;13;28 - 00;29;35;05
Unknown
Right. And that's one of the things, honestly, in using just the generalist tech. But I found a lot of success that if you give it a, hey, this is how I see the future and how I see it, what would it look like to get there. Right. Like that's super something that's super intriguing and something that honestly the general is really good at is being like art here.

00;29;35;05 - 00;29;57;19
Unknown
I can draw you this map and say this is how like I ran 10 billion scenarios and this is the most likely way to get there. Yeah which is awesome. But you still have to have the vision. It's not going to have the vision for you. Yeah yeah yeah. Right. So I mean it doesn't have we captured then your role at this point it's like yeah you are the vision of still like what what good should look like for Griffin.

00;29;57;25 - 00;30;21;07
Unknown
So I don't think that that is unique to me though I think that is something our team does. Oh you're like the central node or. Yeah. Yeah. And and maybe and you could think of it more as maybe like a unifier or the person. I'm kind of connecting the dots between the teams. And, and having the ability to like, I've got a computer science degree, so I do speak a little bit of the tech, in terms of being able to help translate that and saying, all right, here's the business goals that we're taking.

00;30;21;07 - 00;30;42;28
Unknown
How can we leverage what what's out there to data to help us achieve that sooner? Yeah, absolutely. But in terms of saying, what does the vision look like for the future that that's all that's that's our team. And that can't just be one of us. How much do you feel like you're doing. Yeah. Of just like creating space for others by as presumably solving problems that are your preexisting and you know what good looks like and yeah.

00;30;42;29 - 00;31;19;14
Unknown
And it's just a time element. Yeah. Versus trying to come up with with new things, you know, builds new builds on your own versus trying to think through what ideal future state looks like. That's a great question. And it's probably always the wrong balance. Right. So implies advisor learning. Yeah, exactly. So, I mean, I do spend a fair amount of my time, looking at, call it point solutions are the right way, but but really solutions to problems that individual teams have in terms of with the common theme.

00;31;19;16 - 00;31;37;20
Unknown
And really that theme is really around data capture. Like how do we take a process that's been running and get to the point where we can capture information so that at some point in the future, capture relevant information? We want to capture everything. But so at some point in the future, we can then have insight that we didn't have before.

00;31;37;23 - 00;31;56;04
Unknown
Right. That there's that frame and the insight we didn't have before naturally feeds the vision of the future. And like where, where we can go in terms of saying where are we going and that kind of thing. I mean that's something we do quarterly is an executive team get together and and sit down. All right. Here's all the stuff I'm seeing here.

00;31;56;04 - 00;32;15;15
Unknown
Stuff you're seeing like how do we all think this? How can we synthesize all of this information and say, where should we go? What's most important for us to do next? And so those are always honestly, really fun sessions. And then in terms of thinking about, call it proof of concept, some was call in terms of something we don't do today.

00;32;15;15 - 00;32;19;20
Unknown
And it's like, hey, let's go figure out if we could do this.

00;32;19;22 - 00;32;46;27
Unknown
The breakdown unfortunately, today is probably like 75% in the like. Let's solve problems today and get and get data out and more 25% in the hey, let's try something crazy new and see what the business model looks like. See how this would shake out if we did it, depending on how one of those takes off. And because we've done others right, we had one two years ago that's turned into the underlying technology for our fund, that we're working through now.

00;32;46;27 - 00;33;11;18
Unknown
And it's like that fund, the tech for that fund came out of something that was a, hey, let's see what we can do, see if this is possible. And it spun up into, oh my gosh, this is possible. This is a better mousetrap. And we need to make sure that we we leverage this technology to, to do something that at the time, not many others were doing it or now people are, but we've got two year head start kind of thing.

00;33;11;21 - 00;33;31;03
Unknown
And so that kind of thing is, is really where we're saying, okay, like this is this is something that is, is really has now moved from that proof of concept bucket into something that is now more we've got a solution. And now we're kind of saying, all right, what are we capturing about this? How do we make sure that this keeps running.

00;33;31;03 - 00;33;46;12
Unknown
How to make sure this keeps operating. And how do we get information that can feed us insights into the future for future proof of concepts. And so it's kind of that flywheel, right? It's it feels a bit more like a flywheel of problems. Right? Where you're just like as you solve, you're like, oh my gosh, that made me really unearth 15.

00;33;46;12 - 00;34;06;02
Unknown
It's or what is the Hydra where you cut off ahead and three more sprout. Exactly. Yeah. But but like to to be like 100% like transparent. I feel like that's a lot of what businesses though. Yeah. Right. Businesses is learning and being able to adapt quicker than everyone else. Right. Because you know, we I forgot where it was.

00;34;06;05 - 00;34;25;12
Unknown
Maybe it was, no idea where it was, but what's the definition of a successful business. And it's called one that's still running. Right. Like it's really simple. Yeah, yeah. When you look at when you look at businesses that are 100 years old, it's like, is it a great business or still alive? So clearly it's good. You might you're going to get in it.

00;34;25;12 - 00;34;40;21
Unknown
And every business has its own version of dysfunction and problems and stuff that people are working through. And they should be, because if they weren't, the business would be dying. Yeah. So yeah, I mean, I don't know what an ideal business even looks like, right? Because I think everywhere I've been, it's been a bit of a shit show.

00;34;40;23 - 00;34;57;26
Unknown
But it's almost by design. But that's a good thing. Yeah, exactly. Yeah yeah yeah, yeah. Constantly solving. Yeah. That's fascinating. So. Okay. I am curious to talk a little bit more granular about some of the builds you've made personally. Yeah. As the technology. I mean, you're the only technologist there. So. Yeah. So it's it's me.

00;34;57;26 - 00;35;15;18
Unknown
And then I've got, some teams, some people in India that support us, and what we're doing. So. But yeah, that's it's a small team. So me and basically one other person dedicated and, and recently more of, brought someone in part time and, called more of a business analysis role. And it sounds like it's been much more finance focused.

00;35;15;21 - 00;35;47;22
Unknown
So no, it's been it's been all over the place. So we have probably eight different business lines that we're focused on or business units that we're, we're focused on. Where whether we built things for our asset management team or for accounting or for property management. There's been a ton of not a ton of differentiation. So whether it's deal sourcing or measuring tenant retention or, tracking how we're doing in a month and close like there's, there's, there's a variety of, all of that.

00;35;47;25 - 00;36;10;04
Unknown
So which I mean, dive into 1 or 2. What would, would have, you know, the famous questions like, what have been some failures that have led to, to finally succeeding in getting these bills done? Yeah. So very interesting one. Has been and probably what called the worst data mess we've gone through, was actually our tenant retention.

00;36;10;06 - 00;36;32;19
Unknown
So started out in tenant retention. We're like, this should be tenant retention should be on paper, very easy to figure out. Why is it called. Because the concept simple. It's called my my tenants lease ended on call it May 31st. Was the tenant there on June 1st. If they were there, I retain them. If they weren't, I didn't, should be that signer.

00;36;32;20 - 00;36;54;26
Unknown
Yeah, yeah, neither were there or they weren't there. And that's it. And when you go back and look, basically, you find out that the tenant data is a mess because so many people were just making the decision that at that time I needed to to get this in so that I could either build the tenant or so that the tenant would show up in a report or so that I could do something in that system and then move on.

00;36;54;29 - 00;37;15;27
Unknown
And so you go in and you look and you're like, wait a second. This this tenant with three amendments has seven renewals. You're like, this is this is just a mess. And so we honestly had to use a little bit of AI to help us go through and identify all the gaps and all that. Hey, what are the patterns here?

00;37;16;00 - 00;37;37;24
Unknown
Why are people why is this data consistently wrong? So that we could then go change our process? Because until we change the process right, she's going to keep rearing its head. We're going to cut off ahead two more come right kind of thing. And so as we we went through figured out the data problem and said, all right, now we're going to fix the process first.

00;37;37;27 - 00;37;55;05
Unknown
And this is a big part. And and maybe an overarching theme of the way we think about data governance, which is don't fix your data first. If we want a hot take, don't fix your data first. Fix your governance problem first. Fix the governance problem, which is the process and the people using that process. That's got to be corrected first.

00;37;55;07 - 00;38;15;15
Unknown
Then go clean your data because then you won't get more bad data, right? And so fix the governance, then fix your data. And then you can have something on top right. Sharpening the ax before you try to chop down the tree. Yeah. And and so as we go, as we go through kind of that kind of overarching process, that's the same thing we've used for a lot of different, different projects.

00;38;15;17 - 00;38;35;29
Unknown
But as we went through this with this one, we we came to find out that there's a gap in the way people think about underwriting properties. Most everyone uses a 75% retention rate. There's not really a published version, and we're just kind of like, yeah, so for fun, we are so required. Good kind of thing. Yeah, yeah, we're a great operator.

00;38;36;01 - 00;38;57;19
Unknown
Everyone's a great operator. Right. Several. So so we go and we look, can we say, all right, well, hey, like we can't find a root retention rate plot. Here's what we think the retention rate should be. Go go run analysis. Try to find as much information you can. What studies can you find? What is retention rate actually what is retention rate actually across the country.

00;38;57;21 - 00;39;18;06
Unknown
But don't we? The studies were there and the studies were had information. And we haven't done any sort of super deep dive. But the numbers that came back with were around 48 to 50%. Right. Which shows you there's a huge underwriting gap which either people intentionally don't know or people sorry, people don't know or they intentionally ignore. Right.

00;39;18;06 - 00;39;37;15
Unknown
Because this is what the market's pricing out. And that's how I have to price it to be able to compete. Right. Not that that's a bad thing, but as we suddenly start to measure our retention and we see how close we actually are and how much closer we are than the benchmark, what we're kind of saying, all right, this is where we think people actually are, that process.

00;39;37;15 - 00;39;56;07
Unknown
And hey, we cleaned our data or sorry, we fixed our governance, we cleaned our data, and now we've got this thing. We've had it up and running for about almost 18 months now. We can confidently say one hour retention rates increasing because suddenly we're paying attention to it and we're doing a lot of things. And Kelly on our team has done a great job of of making that a focus.

00;39;56;10 - 00;40;15;07
Unknown
But we being having access to that data suddenly now people are going, checking, saying like, oh, like I want to see how how is this doing? How are we doing each quarter kind of thing? And that that growth has been like, all right, we're closer and closer and closer to that 75%, which suddenly now presents an opportunity to go speak back to investors, right.

00;40;15;09 - 00;40;34;06
Unknown
And say, hey, first off, do you even know about this gap? And if you do know about the gap, then gray, you understand it. Do you understand how far a typical operator is away from this gap? Do you understand how far we are away from this gap? There's a significant capital savings if we if we renew 10% more tenants.

00;40;34;08 - 00;40;54;24
Unknown
Right. If every tenant cost $50 a square foot to replace, and you're going to say we're going to save 10% of them, mean on a 100,000 square foot building, that's 50 grand at least short. Simple math. But this this just in the sheer like amount that they're paying you. Not to mention all the work you would have to do to replace them if you.

00;40;54;25 - 00;41;14;04
Unknown
Well, that's that's the $50 in improvements in the work to replace a mine that doesn't count the vacancy cost or any of the other stuff. So a math might not be right. 50 times. Yeah. 50 times. No. Sorry, 500,000. That's where. So 500,000. The costs that you would have for that building for for not retaining 10% of the tenants.

00;41;14;07 - 00;41;35;16
Unknown
Right. Which is a direct hit to your total return. Yeah. And that's that's the needle moving return number. Yeah. Right. Yeah. So but that all started because it was called curiosity around what is our retention rate. And like all right. Well government like we'll figure out how to clean this up because a huge problem that people take on and not to generalize all of this, but a huge problem people take on as they try to.

00;41;35;18 - 00;42;03;18
Unknown
We're going to govern all of our data. We have a governance strategy across the organization. I hate to tell you, but you won't for five years. Like if you try to take that big, broad swath approach, it's like pick up, pick a problem, solve the problem, govern that data, and then let the snowball build to a what I'm hearing is like where AI has been helpful today is is in helping us actually measure something either more actually or measure something that we weren't measuring before.

00;42;03;18 - 00;42;29;28
Unknown
Right. And then from there being able to, like, make better decisions and then from there being able to like action on those decisions in a better way. Right. Using that microcosm as an example, I mean, we will call the foundational models, let us clean our data faster, let us identify the gaps in the processes, in terms of figuring out why people were doing what they were doing that created the bad data.

00;42;30;01 - 00;42;48;28
Unknown
And then, I mean, we leverage cloud code to actually build us something to explore the data. Right? And so, I mean, we use it in each component, which great success in this, in this instance. And then we've started to take it the next step, which is well now we want us to start predicting and we're not having AI predicted.

00;42;48;28 - 00;43;06;19
Unknown
Because again, I go back to the whole we don't know what this is truly going to cost, but we can use AI to build deterministic predictors. Hey, we know these tenants are renewing already. They're already in the jail as a renewal, and it's not going to happen for three months. We know that's a renewal. Hey someone has told us that this tenant's vacating right?

00;43;06;19 - 00;43;38;21
Unknown
We know they're like, oh, there's already a termination amendment in our GL. We know that tenant's vacating. We can predict what our renewal and our retention will look like, which means we can project capital costs. Right? Which means we can project what we need to reserve. Right. AI's helping us build the deterministic process to really capture that insight and to capture that insight on a regular basis to where we can use AI to do that with some other process and not need AI consistently to do this, because I think there's a big dichotomy between like when we've talked about plenty of times, right?

00;43;38;27 - 00;43;55;27
Unknown
The build AI versus the build with AI. We're real estate firm. We do not need to build an AI model right? We have we want to build with AI. But could you yeah. You're saying sure we could. And we could tell everyone we're doing it. And I mean, I'm sure it would make for some great marketing news, right?

00;43;56;01 - 00;44;14;21
Unknown
But I mean, it really does it move the needle nose flashy? Probably, but it's going to be far better to build with AI, build something that's sustainable even when I changes, when ups downs have suddenly I mean, I think it was like a month or so ago when cloud had its problems and suddenly you can't really use it.

00;44;14;21 - 00;44;39;19
Unknown
It's like one prompt in your usage is gone kind of thing. It's like if you get to that world where your rate limit it suddenly to all these solutions people built, just stop working. And it's like, do we really want that to be our risk? No. So, yeah. Yeah, I mean, I think I'm, I'm not as, I'm not as down on that cataclysmic event.

00;44;39;20 - 00;44;58;28
Unknown
I think I think, well, obviously, in fact, I think a lot of people are building now to be able to like hot swap models depending on which layer. And hopefully that kind of mitigates any Y2K moment of, you know, all of this coming crashing down. Yeah. But, but I guess to that end, what, you've mentioned cloud code, you mentioned Codex.

00;44;59;01 - 00;45;22;03
Unknown
What's the tech stack you're using today to build? Yeah. So, so we primarily build building cloud code. Codex is really the group, the one that acts as a code reviewer. That's kind of the way we've we've approached it. Yeah. I think it's fairly standard, fairly simple to how people are using it. The reason why I talk more about cloud going offline is because where we use cloud for teams and that kind of roll out, right?

00;45;22;05 - 00;45;38;23
Unknown
It's almost like, hey, when when Microsoft Data Center goes down and you're on outlook and everyone's on Microsoft Shop and everyone's like, why is an outlook working? It's not like outlook is going to be down. It's not like your email is going to be offline for two weeks. Right. But for the four hours, I mean, you would think people are running around with hair on fire, right?

00;45;38;23 - 00;45;54;24
Unknown
And so you'll see, I think those things happen because static that they don't have the chugging. Exactly. They might be they might be happy. But I do think you're going to see some of those types of things because it's going to be too expensive to be. Hey, we're a clod Gemini and open air shop for everyone.

00;45;54;26 - 00;46;12;08
Unknown
Right now, sure. Are there going to be marketing teams that may have Cloud and Gemini, or are there going to be specific teams that may use multiple LMS, probably like coding teams that will have a Codex license and a cloud license. Do. Yeah. Makes total sense. Yeah. But do I think there's going to be times when you're going to see some outages.

00;46;12;13 - 00;46;35;05
Unknown
Yeah, I've, I've played around with an open source. Not not really. And a lot of it's been because of call it just. We're trying not to take a lot of the risk on for having to maintain this. So the more that we can rely on the vendor, the, the foundational models, as the vendor to do a lot of this work, like that's perfect.

00;46;35;05 - 00;46;54;09
Unknown
So we leverage our MSP to kind of run our servers and things like that where smaller teams. So it's like, all right, if I go stand up an open source server or say like, hey, we're going to set up Cami or we're going to set up something like that. Now, I'm also creating a bottleneck where I can't spend as much of my time focusing on solutions and new proof of concepts.

00;46;54;09 - 00;47;15;20
Unknown
Right. Not to say that it's a distraction. It's good to understand some of that stuff and and definitely play with some of it at home. But professionally, that takes away from the time focused on the two areas that are really should be focused on. So what, like what I this is, this is an unfair question, but what's what's the biggest unlock you've had with AI?

00;47;15;22 - 00;47;35;06
Unknown
Probably in the last two weeks, because naturally, like you, as you look across and go, all right, we got eight different groups we're working for, kind of, in my mind, a different customers like these, all of our executive team, when you're saying that it's individual executive team members or it's different business channels or different business unit types.

00;47;35;08 - 00;47;52;18
Unknown
No, no. So business units, each kind of owned. So the eight different business units owned by all of our executives, some own may own multiple. But for the most part it's this kind of broad priorities board. And you're constantly shuffling those around of like what's top priority? What's the biggest thing we're working on now? What's the biggest thing today?

00;47;52;18 - 00;48;24;07
Unknown
What's the most? How do we move the needle today? And so Claude and co work went through and kind of had it build a personal like personal assistant. And at first I was like, all right, what I really just want this is I really just want you to review my email and meeting minutes and meeting minutes and like, teams messages and any voice notes I leave myself and just kind of like, keep it in my to do list because, like, I've had to call it a pad that I keep everything on for a long time and it works well enough.

00;48;24;13 - 00;49;04;16
Unknown
But when you start getting this, there's always periods and work where it's like every time you uncover a rock, you find three more things you got to fix kind of thing, and that list starts to become unwieldy. And so the biggest unlock I've had is being able to have that actually move those things into the the future box, where it's not that I don't see them, but they're out of my head because I think one of the things that and I was talking with the guy, for a corporate, prep call, yesterday and he's like, in this world of call, I being able to communicate at that speed, I've been he's like, I've

00;49;04;16 - 00;49;28;27
Unknown
never been this busy. And that's how I was three weeks ago until we finally got this system in place where it's like I've got I'm kind of just managing the tasks and surfacing the ones that are important today to where it's like now, instead of just being busy, I'm busy in the right way and focused and and not having to worry that I have to move at the speed of AI.

00;49;28;27 - 00;49;49;13
Unknown
Talking to AI, because I think that's the ideal that everyone, everyone hears AI and what they actually think is right. A gentle guy doing all my work for me and so I can just sit back and relax and it's like three years off that, today. Now, are there some things where that can happen? Yeah. But you're going to find out that the majority of that is just deterministic non AI doing the work.

00;49;49;16 - 00;50;11;18
Unknown
Yeah it was built by AI but yeah yeah yeah yeah yeah yeah I, I God I just think more like macro socially. Yeah I worry our short termism that again you know that has been amplified by but especially by social media is just going to be worse because everything's going to be so quick at our fingertips that that's going to be our expectation for all knowledge.

00;50;11;18 - 00;50;28;23
Unknown
Two day shipping is not going to be fast. Yeah. No, no. It's like, well here's what yeah, the DMV will still take five seconds. That's Yeah. When everything else takes a, you know, something like that. It's just it. Yeah, it's it's both terrifying and exciting. And it's it's cool to have people like you working in it.

00;50;28;25 - 00;50;43;19
Unknown
I was told recently that the ideal podcast length is 46 minutes. It might have been a made up stat, that I'm choosing to believe, but blown right past that. So. Well, hopefully you can cut some of this. Yeah. Thanks, Patrick. This is awesome.