In the Vault: Assessing AI Impact vs. Hype with John Stecher of Blackstone

Episode Transcript

This transcript has been condensed and edited for readability and clarity.

David Haber: You’ve spent the last two and a half decades living at the intersection of finance and technology. I’m curious: what has changed about how these financial firms operate in that period?

John Stecher: I think there’s a lot that has changed in financial services and there’s a lot that hasn’t. The fundamental problems you’re still trying to solve are exactly the same. It’s basically: equities trading is still equities trading. People want to buy and sell stuff that exists in every single market out there. From the retail banking perspective, people still want mortgages, they want loans, they want credit cards, they want deposit accounts, and then, ultimately, every single bit there wants some form of security around the money that they put into the bank or any type of asset manager. And so I think the core business doesn’t change. What has happened is there has been a lot of focus on reducing friction so clients can actually get access to their money faster. They can transfer money faster. They can buy things online faster than they ever were able to before. And that’s where I saw early on — 2001, 2002 — technology was at the forefront of making that happen. So you had really smart people in the finance space that knew how to build products, but they couldn’t get anything done unless technology was woven into every single product that they built. And so what I think has really transpired from that timeframe to now — and it’s going to continue — is that a lot of the folks on the “business side” have to really be tuned into what technology can and cannot accomplish for them.

The way the best teams are formed out there is at the intersection of really smart business people that understand the products that they’re working on, combined with folks on the engineering side that understand what they can bring to bear. If you look around, the best fintechs have that type of talent, the best asset managers have that type of talent, the best banks have that type of talent, and they really foster those teams to work together. 

Technology-wise, the biggest change that I would say I’ve seen is people moving from their own data centers to finally trusting moving out to the cloud. That has unlocked a ton of innovation across legacy financial institutes, as well as all the fintechs that exist.

The other big bit that I’d say has evolved is a lot of banks and financial institutions have gotten a lot more comfortable at not building everything proprietary. And so they’ve looked for solutions with partners on the outside that build infrastructure products that can sell to various different banks, asset managers and whatnot because they’ve looked at as, Hey, this is just a way for us to basically not pay as much cost for technology and focus on product development and servicing and customers. 

David: Totally. Yeah, I think what you just touched on in terms of the culture at a lot of these firms changing — I know you were at Barclays, you were at Goldman [Sachs]. I’m curious if you saw that evolution then.

John: Completely. And, I mean, I remember some of the first banks I worked with in IBM were JPMorgan or the Tokyo Stock Exchange, from an exchange perspective. And so much of the stuff was built proprietary, from nuts and bolts up. Custom hardware to custom operating systems to custom racks that everything sat in to custom network infrastructure. I mean, the time it would take just to get writing the first line of code was insane for most developers. And all that has now compressed so much.

David: One of my favorite lines that I repeat often is: Opportunities live between fields of expertise. And, you know, when I think about alternative investment firms, Blackstone is the ultimate archetype for this because you guys have such a unique opportunity to express conviction in big, secular themes across different asset classes. So you can believe in ecommerce and invest in an ecommerce brand, but also buy the underlying warehouse infrastructure or AI in the same capacity. Is that core to how the firm operates?

John: It’s exactly core to how the firm operates. I mean, take AI. The growing theme was basically: as AI — predictive AI, now generative AI — kept growing, you need data center space. We were way ahead of that curve because we were investing in the ecommerce boom. What does the ecommerce boom and online transaction processing take? It takes data center space. And so that was all part of the reason we invested in QTS back in the day. We’ve invested in a bunch of power infrastructure and other things because you look at: what are the foundational elements that you can’t really go wrong with from an investment perspective, along a theme? And so that is really where we’ve ended up right now with a lot of the AI investments we’ve made — we are really looking at that infrastructure provider space.

So just like you do, I’m sure, we do a ton of looking at: What are the current cool things? How do you map that thematically? And then look for opportunities in those spaces.

David: I’m sure a lot of folks are familiar with Blackstone, but I’d love to dive deeper into your role. How do you view your role as CTO? And what role does technology play here at Blackstone?

John: It’s the coolest job I could have ever asked for when I got into this game. It was the culmination of all the stuff that I’ve liked throughout my career. I have four functions that I play, and depending on the day it varies how much of my time I put into them. 

So first and foremost, it’s everything engineering and technology around the firm. Making the trains run on time is the way I think of it. And if you make trains run on time, from a technology perspective, you’re in the right to keep doing more and more advanced stuff. Then we get into the actual data infrastructure and space that we have. So over the past four and a half years that I’ve been here, we have invested heavily in unlocking our data advantage. Then you layer cybersecurity on top of it, which I find to be one of the most interesting things in the world because you get to play like you’re James Bond for a little bit — How can I defeat the villains? It’s always fun. So that’s job one. 

Job two is I run Blackstone’s innovation investment. And it’s one of the most fun parts of my job. It’s looking at early stage companies — fintech, proptech, cybersecurity, and enterprise tech ones we’re using or our portfolio companies are using — where we have conviction to actually pay them money to consume something. Like, we want to invest with them because we believe across our network we can really help move them forward quickly with low friction. 

Job three is working with all of our portfolio companies. And so I spend a ton of time with the various different CIOs, CTOs, and CISOs across our portfolio companies.

And the fourth job is I work with all of our investment teams from the technology part of our investment thesis. So it’s helping define strategy. It’s actually doing diligence. It’s spending time with our investment committees and actually going through what are the gives and takes around, Is this a good platform? Does this make sense the way they do product development? Does this make sense, the technology that they built? No different than we look at a DCF for a company. And so, it’s fun to come into work.

David: I’d love to unpack a few of those different pieces. You oversee a really big engineering organization and are responsible, I imagine, for allocating those resources. How do you determine what you actually choose to build internally versus those third-party vendors versus the innovation portfolio companies you chose to partner with and invest in? How do you think about that trade-off?

John: I learned a lot of this from Marty Chavez at Goldman, when I worked for him and with him. He had the “download it/build it/buy it” kind of mentality, and that is something that has always stuck with me. There was always the theory of, if other people need to build the same stuff that you’re building, why not partner up and basically build the highest quality product at the lowest cost and just open source it? Not give it away, but have other people contribute to it and then focus your engineering talent and brain on building the best possible value creation tools for what you’re doing. And so that has stuck with me. Use your brain to solve the hardest problem that brings us value. And so when we look for new stuff that we have to do to provide technology for the firm, it’s: Is this proprietary or not? That’s filter function number one. If it’s proprietary, if it will make us value — definitely something that you would build in-house.

If it doesn’t fit that criteria — not proprietary — the second look is: what is around in the marketplace? So, is there something that you can procure from a large vendor that matches up and serves exactly what you need? And so [this is] why we’re on AWS, why Azure is popular, why we leverage a lot of platforms as-a-service from the outside: there’s no reason not to swim with all the other fish. They’ll fix bugs for you faster than you can, etcetera. So we take that as a second look. 

If there’s not something that’s a perfect fit out there, or there’s something that we believe is maybe proprietary and unique to this industry — where we have us, KKR, BlackRock, Apollo, we all need the same thing — that’s really where the investment bit comes in. And so we start to take a look, at that point in time, into strategic investing for the BXTi (Blackstone Technology and Innovations) world. We look at, is this something that across the industry that my peer at Apollo would go off and buy? And if that is true, and we could find a company in the outside world that will do it, it’s worth it to put money into them and to pay for that. And so investing in those core thematics is what we’re focused on.

David: Totally. What advice do you have for early stage companies trying to sell into a firm like Blackstone?

John: The easiest way for smaller companies to sell in is to just be honest and up-front with what your product can do, number one, because most people are smart enough to sniff through the sales pitch and get down to brass tacks pretty quick if you’re working at a tier one institution. So that’s number one. 

Number two is: Have a product that very easily creates value. What’s the time to prove value and what’s the time to purchase? And so the more complex a product is, the longer it’s going to take you to do a PoC, which means it takes longer than you would ever expect to actually have somebody get it installed, run it, prove value, etc. It just makes that sales cycle really long. 

So I would tell entrepreneurs: Focus on out-of-the-box experiences being like, “Holy crap, this is a wow moment, I need this,” because it’ll result in sales faster. It’ll result in technologists pushing their procurement teams to get you onboarded as quickly as possible in the pipeline. 

And the third thing I’d say is that I think it’s really important for young teams or young companies to actually be able to understand the industry they’re trying to sell into. We struggle with this every so often: you know, a new tech company will come in, want to pitch us on something cool, and they have no clue what we do. They don’t understand the cyber requirements we have, the regulatory requirements we have. There were a lot of AI companies early on that were like, “Oh, just give us all your data.” Well, what’s your security profile and what’s like your data loss prevention program? “We don’t have one, just give it to us!” And that’s just not going to fly. The best companies understand who they’re selling to, what industry they’re in, and how that person in that industry can extract value quickly.

David: What are some of the other qualities that you look for in the types of companies you like to get involved with?

John: Leadership team. Early stage companies, you’re backing a version one, if not an alpha, of a product and how much you trust “Sally the CEO” and “Johnny the CTO” to actually build the thing. And so, that is a challenge. And that’s where I think you really need to spend time and push on those people quite a bit to see: are they marketing material or are they actually legit? We do a lot of reference checks, asking them combative questions. It is “battle it out to have the best idea win.” You have to push people pretty hard in these spaces. Nice thing is, given the market slowdown, capital raising is a little bit harder for some of these companies, so they’re more willing to have these conversations versus just blowing through “Here’s a term sheet, sign it or leave,” which existed in the market in the 2020 and 2021 timeframe. 

The other big thing I’d say we look for with our companies is how well they can talk with the proposed buyer inside of a firm. So if they can engage with our CFO, our chief accounting officer, the head of valuations, our head of cybersecurity, and have a legitimate conversation back-and-forth where you can see they understand the space, they understand the problem set, and they can actually really convey how they’re going to go off and be able to change the world for them, that’s a high degree of conviction, as well.

David: It’s a good segue way, and I want to spend a bunch of time talking about generative AI. I recall from our lunch last year talking about some of the challenges of actually deploying this technology in an organization like Blackstone. And I recall you mentioning the challenges around permissioning access to different data and NDAs and what sounds like basic infrastructure, but it’s actually really complicated. Maybe talk through what you’re still working on to deploy this technology.

John: Nothing has changed a ton. If you think about when ChatGPT came out the first time and OpenAI started saying, like, “Hey enterprises, come play here.” They didn’t even have any type of real data security policy. And so I think that was the first thing people were like, ugh, I can’t do this, right? Some folks rushed in, sent in a bunch of information and didn’t think about it too much, and then kind of like — hey, what’s going on with all that? — asked a bunch of questions. If you’ve worked in the financial services industry for any amount of time, you know how closely guarded we are with our people’s and our clients’ information. And so, the first pass is like “Spidey sense.” What are the data policies? There’s a bunch of different stuff from a legal perspective there.

What we’ve had to focus on — and we’ve had great partnerships with the AWS folks and the Azure folks — is making sure the core models and core infrastructure that’s out there is secure, that it has all the encryption capabilities we want that it has all the data rights that we want, that it’s everything that we deem necessary. But then on top of that — and they still haven’t built this, what they’ve been building — we have to worry about things like NDAs. So if we have a deal come in and we want to pass a bunch of deal documents, let’s say we get 100 documents for a specific deal and we have a two-week NDA that we can actually view those documents on. We have to legally prove that those documents go away [after two weeks], right? Used to be easy, just delete them off of your desktop. If you stick them into an [AI] model, there are a lot of folks from a legal perspective saying you have to be able to prove that you can remove them from the model. And so we’ve had to spend a lot of time building our own infrastructure around the Azure stack and the AWS stack to do that. So those are a lot of the challenges.

You have other things as simple as: You load up the model with stuff. [Say you and I are] entitled to a deal; we both work at Blackstone. Melissa, who works on a different team, can’t see the same stuff. So she asks a question of the model. It can’t have any knowledge of the documents that we have. And so this just creates a whole different paradigm from what I think the folks that were building the stuff out of the gate, and even now, were thinking about. And so there are now startups in the space doing more stuff like this. There are a number of interesting ones out there, but nobody that I’ve seen has really pushed the envelope on this. And if I talk to peers at different banks, a lot of them are having similar problems. They’ve solved it in different ways, but they’re facing similar problems to those we’ve had.

David: Does it inform how you think about open-source LLMs in any way? Using, say, something like a Mistral or LLAMA and rolling it in-house and hosting yourself?

John: It does. And we went down that path. What we found was, you know, the power to be able to switch the models underneath the covers actually was far greater than us building our own models. And so we wanted to spend more time building the security layer around it so that we can do all this stuff so we can switch from GPT 3.5 to 4 to 4.0 to Mistral to Anthropic and figure out what is the best for the different use cases. And that’s one of the very interesting things: every model is very different at reading and actually extracting information from different types of docs. Anthropic is good at certain things. GPT 3.5 is amazing at some stuff and terrible at other things. And then you also have the performance criteria you have to deal with. So we focused on building a layer that abstracted us away from what I consider just to be underlying infrastructure. I think the basic model stuff is going to become underlying infrastructure, no different than cloud providers. So I think it’s just going to become a game of the hyperscalers and a few players that actually build very interesting model infrastructure behind the scenes. And then everybody else will run possibly some open-source stuff. But most people are just going to say: that is the game. They’ll focus on building the middle layer, like for permissioning and everything else. Somebody will eventually solve that. And then the real value is in the verticalization of AI, so doing a whole bunch of interesting things that ultimately make jobs easier for people at the end of the day.

David:  Given that response, would it ever make sense for Blackstone to build your own LLMs or maybe train an existing model on your own data?

John: We’ve done numerous experiments down that path. We have a wrapper around the basic models out there called Secure Chat that people inside the firm can use. It protects information from leaking outside the firm and gives responses back like you would get from ChatGPT or Anthropic. 

Then what we’ve really focused on, beyond that, is the document bit. So being able to search and ask questions of reams and reams of documents instead of having an analyst go into it. [An AI that can] read through it, highlight it, mark it up, summarize it. The ability to just take a first pass crack with a tool that can extract it out, deep-link you back into where these things are in the document, it just makes our analysts more effective. And so, it could be a private equity analyst, it could be a legal analyst — it’s all the same problem: I have a bunch of documents I need to rip through, extract information out, clean knowledge on, and then rock on from there.

David: I’m curious, how does the role of that junior analyst change in the world of generative AI? Is there any anxiety among the most senior folks at the organization? There are some fundamental skills that you learned going through an analyst program here, I imagine — how to run a DCF or build a merger model, whatever it is — that candidly can now be done by a lot of these models. And so how do you think about a new tool — which exists in every technology cycle, you can’t avoid it — while also needing to learn the fundamentals of the job?

John: This is an apprenticeship business. You learn how to do stuff by grinding through these things. And so there is a little bit of trepidation around, well, but this is the rite of passage. You learn how to analyze reams of documents and extract out key information. You understand how to write really concise summaries of those documents so that you can pull together an actual investment committee memo. So you can make a deal pitch, right? So there’s definitely concern around it. My view on it is that this is no different than every bit of evolution of technology we’ve ever had. Technology keeps evolving forward. It doesn’t mean you don’t still understand the core skills. It’s just an enabler for you to do more faster. Technology has compressed cycle times and made stuff more efficient. And so I just view AI as doing the same thing.

David: You touched on this a little bit, but I think it’d be fun to walk through the different asset classes and imagine the impact of generative AI across each. I feel like we’re seeing more private equity-like deals in venture land than I ever have, candidly. How does generative AI impact the way that your private equity teams think about their investments and the potential value add that you can deliver to those?

John: A requirement for the way we evaluate our investments is, what is the impact of both generative and predictive AI continuing to evolve on this company? One of the things that we looked at early on was around, what is success and what does great look like? is the data moat that these businesses have. So how much of the data is really proprietary, where if you layered AI on top of it, you could actually extract out a ton of interesting information to make the business more valuable. So that’s, I think, what most people are fishing for.

David: Totally. It feels like there could be sort of a — I’m going to exaggerate — an extinction moment for a lot of private equity firms who don’t understand technology. Because the historical playbook — again, I’m oversimplifying — probably wasn’t that.

John: I mean, I think it’s key. I think it differentiates us here. And on my team, I have a bunch of people that work nonstop with our investment professionals that are subject matter experts in database technology, ETLs, cybersecurity, and bolt them into the investment teams and help them share knowledge. And so that definitively, in my opinion, differentiates the way we invest. It’ll differentiate returns in the long-run, is my guess.

Now that said, the world is made up of a lot of businesses where technology doesn’t really matter to them. It might matter to the input, but like: cement industry [for example]. There’s a whole bunch of stuff there. They’re great businesses; there will still be a ton of alternative asset managers and private equity firms that [consider that] their bread and butter. They’ll continue to crush it there, because it’s such a knowledge game.

David: I was going to ask about real estate. These are industries, at least from my point of view, that have historically been slower to adopt technology. Is that what you’re seeing? Or are you seeing opportunities for gen AI to have an impact on your real estate business?

John: I think everybody got all hyped up when all the gen AI stuff first came out thinking of the creative play. And now it has pivoted back to: we can actually just leverage it to make ourselves more efficient. And then we can take that cost savings and reinvest it into actually driving the business forward. That, to me, is what we’ve seen a lot of in the real estate play. There are a whole bunch of platforms there that I think are interesting, but that’s what we’ve seen in that space.

And then getting into the other hard assets: power. Microsoft wants to buy tons of alternative power, right? There’s so much power draw that’s going to be needed here over the next decade. When you think of the electrification of everything, let alone data center power draw, this is a huge space to be in. The supply chain to make AI work or to make ecommerce work starts with power, at the end of the day.

David: Maybe we’ll touch briefly on a couple of the other categories. Credit, in my mind, is probably one of the most document-intensive investment asset classes — just all the different flavors of covenants. And I imagine you guys are already…

John: Spot on. I mean, that’s exactly what we’ve been looking at so far: parsing all of our credit covenants, making sure investor guidelines and mandates, we can extract them, true them up against our compliance engines all the time. If there are changes, ensure that everything gets trued up together, as well as human-intensive tasks before. And this is still evolutionary. These are early innings. We’re working on building this stuff out.

The other thing is predictive AI. So it’s like the uncool, original older brother of generative AI, but we leverage it to do risk management, pricing, a whole bunch of things in that space, as well. That’s really where we’re focused in the credit bit.

David: I guess to close, in the next 5 or 10 years, how do you see Blackstone evolving? And what role does technology and AI play in that future?

John: Technology is becoming the backbone of the firm in terms of how we run and how we operate. We’re still a collection of insanely smart investment people. That is ultimately what the firm is at the end of the day. And that’s what we do, that’s our job: to deliver the best possible returns we can for the LPs. But to enable them to be as smart as they are, human-wise, they need technology to reduce friction in their life and to augment the intelligence that they have. So that’s really where I see technology going. You can see how AI plays directly into that, with the ability to actually recall information. As I’ve gotten older — two and a half decades in the industry — I forget a lot more stuff than I remember anymore. So having a copilot that I can ask, do you remember the conversation that I had with David back at lunch? I can actually look it up. That’s awesome, it makes my life easier. But these are the tools that I think are going to differentiate it.

It [may not sound like] “leap forward” stuff, but after being in this world for 25 years and seeing really, really cool stuff, technology just takes a long time to adopt. From a Blackstone perspective, I think the top of the firm is amazingly dedicated to continuing to grow this business. That’s where we’re trying to push the whole firm — more assets under management. That means more investments. That means all the more need for technology as well. The only way you can manage more investments, make more investments, is if you are able to see more investments, synthesize them, determine which ones you want to go after and do deep dives on. It’s all technology-driven. 

And so I think it’s just like the evolution of banking. It’s like, GS probably back in the ‘80s was open pit trading, talking to each other, right? And you look at it now, it’s technology through and through. I see this industry evolving very much the same way.

David: John, thank you so much, this was an amazing conversation. It was a really good time.

John: I really appreciate it. It was a lot of fun.