Latest / Private Equity Data Guy / The Compliance Reality Nobody Tells You About AI

Transcript

The Compliance Reality Nobody Tells You About AI · Private Equity Data Guy

• 0:00
  AI has kind of created an explosion of AI ideas where, you know, people are asking AI, where could I embed AI in the business?
• 0:08
  And it's given them like 100 use cases.
• 0:10
  And now you're at a point where like, no one's really moving anything into production.
• 0:13
  They just have this entire backlog of use cases.
• 0:16
  Right?
• 0:16
  AI is great if you know how to spot where it's making mistakes, but if you're just allowing people to go free reign and use these things and they're not like checking your work like it makes mistakes all the
• 0:29
  time.
• 0:29
  We're going to have a 15% reduction in force for AI, but they're reducing their workforce before AI even achieves what they're looking to do.
• 0:38
  And now they're scrambling because like, wow, AI is more costly than the people that we let go.
• 0:44
  Now they're straight away to, you know, like, hire these people back.
• 0:47
  And I'm not talking about JP Morgan, I'm talking about just companies in general.
• 0:52
  Behind every value creation plan, there's a data problem nobody wants to talk about.
• 0:56
  Fragmented systems, metrics nobody trusts, and decisions made on gut feel dressed up as analysis.
• 1:02
  Welcome to the PE Data Guy.
• 1:03
  Each week, host Graham Crawford talks to the operating partners, advisors and practitioners who are doing the work inside portfolio companies.
• 1:10
  If you care about what actually drives returns in the market, then you're in the right place.
• 1:15
  The PE Data Guy starts now.
• 1:18
  Brandon Miche spent three years at Capital One building one of the largest analytics communities in enterprise software, and then went on to lead AI and data initiatives at Citigroup, Southwest Airlines, EY and most recently JP Morgan Chase, where
• 1:30
  he deployed an LLM assistant to 27,000 users across the payments organization.
• 1:35
  He's built data platforms, AI products and analytics cultures inside some of the largest companies in the world.
• 1:41
  Today we're going to talk about what it actually takes to make AI work inside a big financial institution, what breaks when the data isn't ready, and what mid market companies backed by PE can learn from the playbooks that
• 1:52
  Brandon has run at Fortune 500 scale.
• 1:54
  Brandon, it's great to have you on.
• 1:56
  I know we've been trying to arrange this for a while.
• 1:57
  Happy to get on your schedule and yeah, great to chat again.
• 2:00
  We've known each other for a long time.
• 2:02
  Yes, likewise.
• 2:03
  It's great to reconnect with you and do this podcast finally.
• 2:06
  I know we've had a couple gaps trying to schedule it, but very eager to kind of jump in.
• 2:11
  I miss the partnership that you and I used to have at Capital One, we go way back, you know, so have been looking forward to this for a while now.
• 2:18
  Yeah.
• 2:18
  And I get to catch up on all the amazing things you've been doing rolling out AI.
• 2:21
  You know, we hear about all the.
• 2:23
  All the kind of ways it's gone wrong.
• 2:25
  So I'm quite looking forward to hearing about some of the successful AI deployments which are in the 5% minority according to MIT.
• 2:31
  So we will get onto that.
• 2:32
  But let's go back to where we first overlapped, right at Capital One.
• 2:35
  And one of the things you did there was build out the Tableau center of Excellence, up to 30,000 users across the company.
• 2:43
  I always say that's about big transformational change.
• 2:46
  That's not a technology project.
• 2:47
  That's a big change project and a culture change project.
• 2:51
  So one that is motivated by getting the entire organization to use data in a different way.
• 2:55
  So what's the.
• 2:56
  What did you learn about that and what did you take away from that that you've used ever since?
• 3:01
  Yeah, sure.
• 3:02
  So I think at first, when you are deploying a tool that could possibly go, you know, gangbusters, like it did at Capital One, I think Capital One had a really, you know, really embrace of way of, like how
• 3:15
  they embrace technology.
• 3:16
  They rarely ever told their internal customers no, but there wasn't really a lot of guardrails around it.
• 3:21
  Right.
• 3:21
  So I think the whole reason Tableau took off at Capital One was I think Rich Fairbank went around to his, you know, legendary town halls.
• 3:29
  I remember listening to him talk from like 8 in the morning to 5 and evening.
• 3:34
  Nothing Ever have I seen anything like it in my career.
• 3:37
  Yeah.
• 3:37
  Uniquely wonderful, I think.
• 3:38
  Uniquely wonderful.
• 3:39
  Yeah.
• 3:39
  In the way that he broke down technology and got every employee of the company excited about it.
• 3:43
  Really kind of, I think it just took off where next thing you know, 30,000 people went and downloaded Tableau.
• 3:52
  But there was no framework in place around best practices change, adoption, training, replacement, and legacy bi.
• 4:00
  So I think you had a culture where everyone embraced the new technology trends, but they really didn't take a step to kind of think about, okay, what happens if the scales to a massive scale?
• 4:12
  How do we ensure that people are using it correctly?
• 4:16
  How do we control the costs?
• 4:17
  How do we control licensing?
• 4:19
  Making sure that people are leveraging this tool appropriately.
• 4:23
  And I've seen the same thing across many organizations.
• 4:26
  When I was at another organization, Alteryx kind of came in and gave everyone a trial license.
• 4:32
  And next thing you know, you had 10,000 people at a large company using Alteryx for everything.
• 4:37
  That could be basic etl, right?
• 4:40
  Like you could have spent a couple hours teaching people some easy SQL SQL and they would be experts at what they're doing.
• 4:49
  So you got a 5000 hour tool basically doing basic ETL and then also enabling terrible practices on data infrastructure like our platforms.
• 4:58
  Right.
• 4:58
  Ultrayx ain't cheap either, from what I remember.
• 5:00
  Not cheap.
• 5:01
  And the output that like, I mean just basically create like a simple select statement with some like trimming and whatnot.
• 5:09
  Puts out like a 10,000 line code of output.
• 5:13
  Right.
• 5:13
  So you can imagine the strain on your data platforms.
• 5:16
  Like it wrote really, really ugly outputs where like that would have been a simple.
• 5:20
  Let's just kind of teach some people like how to go on fish and you know, do some data wrangling and then use Alteryx for what it's really good at, which was like advanced analytics.
• 5:30
  But it became a very difficult conversation when Alteryx came to renegotiate their contract.
• 5:36
  And they're like, hey, by the way, like, you know this intro period has ended.
• 5:40
  We now want $5,000 per license for 10,000 people.
• 5:45
  Right.
• 5:45
  And you know, for two years I was kind of always kind of cautioning like these guys are like drug dealers.
• 5:51
  Right.
• 5:51
  And I don't fall back for it.
• 5:53
  But you know, that's how a lot of these tech companies work.
• 5:56
  And I know because I've been through that rodeo with Tableau and a lot of other companies.
• 6:00
  Yeah.
• 6:00
  I mean it's been all the rage in private equity for the last 10 years.
• 6:04
  Like SAS platforms with the current revenue have been the gold that everyone's been chasing.
• 6:08
  You see it same with, you know, like, you know, AI now cloud before, like when you look at the economies of scale and like how your costs can balloon, I think that's what really keeps a lot of CIOs
• 6:20
  up at night is like, how do we control costs for unexpected surges.
• 6:24
  And I think you're seeing it across AI right now with you know, companies with these, you know, 500,000 hour token bills for like one user in one month.
• 6:34
  Yeah, token max, token maxing.
• 6:37
  You heard of that one, right?
• 6:38
  Yeah, exactly.
• 6:40
  Talk about setting the wrong KPI for the wrong reason.
• 6:42
  Yeah.
• 6:43
  So back to you know, the, the tableau story that was kind of classically how that job, you know, how I got brought into that role was I brought Tableau to two companies before Tableau went public.
• 6:54
  I brought it into PwC very early in my career because I started off, you know, working as a grunt analyst at PwC and I got tired of Recreating the same dashboards every month using month old stale data.
• 7:07
  Kind of set up a poc.
• 7:09
  Then I brought the booze down Hamilton.
• 7:11
  And then at that time, I guess, you know, if you remember, Capital One was looking for a COE leader.
• 7:17
  It was like a partnership between you and I and it's kind of like, you know, I was a huge advocate for the tool, but I also knew a tool is a tool and you got to have guardrails around
• 7:27
  it, you got to make sure people are using it for the right things.
• 7:29
  You got to make sure you're enabling best practices and training.
• 7:32
  And it's kind of new to me.
• 7:34
  A lot of this stuff I never really thought of before.
• 7:36
  So the things that you learned from that.
• 7:38
  Right.
• 7:38
  Say we've transported as we have like 10 years into the future and here we are now and you've got companies turning on AI like you described with Alteryx.
• 7:46
  There another example like you described with tableau at capta1, where even more so actually you can go in and press a button and suddenly boom, every employee in your whole company has got Copilot or Gemini or whatever.
• 7:58
  Or choose your AI tool of choice with no instructions, with free reign, with generic encouragement from the top down to use it.
• 8:10
  If you're in a kind of mid market company looking to grow and someone's just pushed the button and turned it on for everyone, like what should people do?
• 8:20
  What are the right steps to take to get a rollout that actually sticks, that actually delivers value?
• 8:24
  Yeah, I think first do it small scale, do it with a small subset of users.
• 8:29
  That's what we did with our first AI use case.
• 8:31
  About 100 power users.
• 8:33
  See how they use it, create training around it.
• 8:36
  I think to your question, should you just turn it on for everyone?
• 8:39
  A lot of risk involved in that.
• 8:41
  And I've talked to some organizations where they've done that.
• 8:44
  They've kept the light on.
• 8:45
  They're allowing people to do their shadow AI within the company because from a risk perspective, they're more afraid of them going using it on their personal device and then leaking out like, you know, confidential information through like their
• 8:57
  personal chat GPT.
• 8:59
  Which isn't a risk by the way.
• 9:00
  It's a reality.
• 9:00
  Right.
• 9:01
  And I, it's a reality and that.
• 9:02
  Keeps a lot of, you know, CSO's up at night.
• 9:04
  So some companies are just like you, we'll let people do whatever and then we'll monitor it.
• 9:08
  Right.
• 9:08
  Just to make sure it's still kind of within our geofence network.
• 9:14
  Let's just enable all the tools almost.
• 9:17
  As a risk mitigation ploy for that data leakage issue, Right?
• 9:20
  Yeah, exactly.
• 9:21
  So we'll monitor it, we'll see what they're doing, maybe we'll kind of learn from it.
• 9:24
  But with that comes huge risks as well.
• 9:27
  As you can imagine people leaking out data that they shouldn't.
• 9:30
  You don't have to write safeguards or guardrails around those tools.
• 9:35
  So I think the best approach is really, you don't want to tell people no because you know they're going to go use it on their own personal device.
• 9:42
  But you also want to make sure if you are rolling this out, that you've worked hand in hand with compliance to figure out what controls you need in place, make sure that you have training and best practices in
• 9:53
  place for people say, hey, if you're using Claude, here are some of the things that you should use it for.
• 10:00
  Here's some of the things you shouldn't.
• 10:01
  Right?
• 10:02
  Yeah.
• 10:02
  Here's what it's good at and here's what it's not good at.
• 10:04
  Right, exactly.
• 10:05
  It is great at this, but it's probably terrible at this.
• 10:08
  And you know, kind of really lay that out and work closely with the business to make sure that one adoption is, you know, growing steadily, but also make sure that it's growing in the right direction in the sense
• 10:19
  that people are using it for the right applications, the right use cases.
• 10:24
  Yeah, no, 100% the case for sure.
• 10:27
  And you know, as you and I both know, like, context is everything, especially in modern AI tooling.
• 10:35
  And there are still people out there.
• 10:37
  And it's interesting, you and I spend our days ensconced and involved in this.
• 10:41
  Right.
• 10:42
  But there's still people out there who think it's just like a better Google that knows everything about everything, which is a flawed understanding in how generative AI actually works, but a very common one, a very commonly held flawed
• 10:52
  misunderstanding.
• 10:53
  And it makes mistakes all the time.
• 10:55
  Right.
• 10:55
  I mean, I was just using it today to research something personally.
• 10:58
  Right.
• 10:59
  Like research like security cameras and you know, the Amazon sale that's going on and like looking for like upgrade my ring cameras and whatnot.
• 11:07
  And like it was going out there, it was doing all this research, but it was, I know which are the latest models and they're like, which one's 1080p, which one's 2k, which one's 4k?
• 11:18
  Like it was basically like regurgitating articles from like four years ago.
• 11:22
  So it's giving me wrong information.
• 11:25
  So that's the thing where like, you know, AI is great if you know how to spot where it's making mistakes, but if you're just allowing people to go free reign and use these things and they're not like checking
• 11:37
  their work, like it makes mistakes all the time.
• 11:39
  Like I remember recently it updated like a resume for me and it put on there that went to Harvard Business School and like, you know, I caught up.
• 11:46
  Lander went to Harvard Business School.
• 11:48
  No idea where to.
• 11:50
  I just thought that would be a good thing.
• 11:52
  That's a good thing to write for sure.
• 11:53
  Yeah.
• 11:53
  But it probably like looked in, you know, the context of like my history and found out that like one of my career goals is to get an MBA from Harvard Business School.
• 12:01
  So maybe just automatically thought I achieved it already, I don't know.
• 12:05
  Or another angle, like if you said like tailor my resume to maximize my chances of getting this job, then maybe having Harvard Business School on there would be the right thing to have.
• 12:14
  And this is the thing.
• 12:15
  Certifications while you're at it.
• 12:16
  Yeah, exactly.
• 12:17
  Like it's back to the context and guardrails.
• 12:18
  You didn't tell it not to mix shit up.
• 12:20
  So therefore then that's okay.
• 12:21
  Right.
• 12:21
  So it's, it's, it's a funny one.
• 12:25
  Well, I, I want to go back to the dashboard briefly because I know that you were commissioned to build specifically a dashboard.
• 12:32
  When we're at Capital One for, for Rich, the CEO, a common problem I see in in mid market portfolio companies is the board reporting or the leadership reporting or the monthly quarter reporting is not.
• 12:44
  They don't always choose the right KPIs.
• 12:46
  We talked earlier about token maxing, right?
• 12:48
  So you know, saying token use as a, as a proof of AI value is a terrible choice.
• 12:52
  Sometimes companies, you know, you can have all the revenue in the world, but if you're making a loss on everything, then revenue is actually a bad thing.
• 12:59
  Right.
• 12:59
  As an extreme example, one thing.
• 13:01
  And then secondly, the far more common, the wiring underneath those KPIs is never investigated.
• 13:07
  So these numbers appear, they look green every month, but yet actually they're being sourced in a, in a very kind of way that's optimized for making them looking good.
• 13:15
  There's no good data governance.
• 13:17
  There's no standardized definitions of terms within the company.
• 13:19
  So people have artistic license to make themselves look good.
• 13:23
  So knowing all of the things can go wrong.
• 13:26
  Like a Fortune 100 CEO, how does he operate?
• 13:29
  What was that project like and what sort of things?
• 13:31
  Obviously protect the confidentiality, but what sort of level of detail is he looking at?
• 13:36
  And then what decisions was he seeking to drive?
• 13:41
  Yeah, so, you know, it was high level basically to drive his board meetings.
• 13:45
  So of course they had to have like a tier one environment dedicated to it.
• 13:49
  It had to have 100% availability because.
• 13:52
  And another thing, he wanted it on his iPad.
• 13:55
  So the whole iPad design too was also a critical.
• 13:58
  I remember it was essentially.
• 13:59
  Yeah, yeah.
• 14:00
  So it was very essential that like the thing was up all the time, could not go out in the middle of a board meeting that would be catastrophic.
• 14:07
  Had to have, you know, very high level, like security around.
• 14:11
  It had to be iPad design, which when we got into iPad design, completely different, like thought from a product standpoint because, you know, like you're holding an iPad, you don't want certain filters on certain edges and whatnot because
• 14:25
  you might click them by accident.
• 14:26
  And that's the whole thing with dashboarding is depending on what you click on, like you might be feeling something completely different and then making board level decisions off like the wrong slice of the data.
• 14:37
  How was, how was the spec conversation?
• 14:40
  Like, did he come with like, I'm looking to drive the decisions or looking to update the board on these things?
• 14:44
  And again, you don't have to say the specifics, but like, how did he approach it as a leader in terms of what he needed to see?
• 14:50
  Yeah, so, you know, I didn't work with him very closely, but it was coming from like his like leadership team.
• 14:56
  There was like a finance organization had.
• 14:58
  Here are the like, high level, you know, areas that we want to kind of explore.
• 15:02
  Here are the KPIs that are like critical.
• 15:05
  So I don't know if you remember this effort, but I remember it was like, it was probably like 20 dashboards within a dashboard.
• 15:12
  So many different slices.
• 15:13
  Everything around, you know, financial, the revenue, you know, operating costs.
• 15:19
  There was like a dashboard for like every one of the functions, people.
• 15:24
  Very, very high level for Ridge.
• 15:26
  But if you remember, there was a lot of high level, like slices, if you will, that went into that.
• 15:32
  And we worked with, I think we worked with the organization, was head of pretty much top of the house reporting and finance.
• 15:39
  Got it.
• 15:39
  Yeah, it's.
• 15:41
  It's interesting to think.
• 15:42
  Yeah, I'm trying to remember a lot of the details.
• 15:44
  It's been like, when was that?
• 15:45
  It was like, of course, yeah, 20, 15, 16.
• 15:48
  Yeah, yeah, it's been a while.
• 15:49
  But yeah, interesting that, you know, the wiring and the data underneath that for all the technicalities of displaying on the iPad, like making sure that was correct and that's one of the things that I certainly took away from
• 15:59
  Capital One.
• 15:59
  The amount of rigor that goes into data governance, like, seems onerous at times, but I'm sure is hugely paying off in this AI era now.
• 16:07
  Well, back to your point too, about how the data engineering was done.
• 16:10
  It couldn't be row level data because if you remember Tableau, a lot of people in the company would load millions of rows of data into it and.
• 16:17
  Choke it to death.
• 16:18
  Yeah.
• 16:19
  It was not a data analytics tool, it was a data visualization tool.
• 16:24
  So you had to have one of those metrics, like baked it at like, you know, they're basically all calculated fields.
• 16:31
  Yeah.
• 16:31
  You had to do the ETL and the aggregation on the way in.
• 16:33
  Yeah, yep, yep.
• 16:34
  So in order to have the performance and that's the thing, a lot of people would try to like load row level data into it and be like, oh, it takes five minutes to load.
• 16:41
  This tool's terrible.
• 16:42
  It's like, no tool's terrible.
• 16:43
  Yeah.
• 16:43
  You haven't, you know, done the data engineering behind it to actually make it run efficiently.
• 16:48
  And for that level of reporting, you want it blazing fast, loading with it, you know, milliseconds.
• 16:55
  You don't want to have to sit there in a board meeting and like wait five minutes for a dashboard to load.
• 17:01
  So those were a lot of things that we had to get exactly spot on and correct in that initiative.
• 17:06
  Absolutely.
• 17:07
  I want to move on to a different big financial institution though.
• 17:10
  JP Morgan, you defected to the other side, where again, similar numbers you deployed.
• 17:15
  You oversaw the deployment of a Gen AI assistant to 27,000 people.
• 17:19
  Now, we talked about this before.
• 17:22
  We should measure the right thing.
• 17:23
  Right.
• 17:23
  27,000 is great, but it's no good if 27,000 people are making bad decisions as a result of it.
• 17:30
  How do you go about, especially in what would have been an earlier stage of AI than we're in now, rolling that out and making sure it's successful at such a big scale.
• 17:40
  Yeah.
• 17:40
  So I have to be careful what I talk about in terms of specifics, but I can tell you the first thing at any large financial institution is getting it through the compliance hurdle.
• 17:50
  I'm sure you can imagine our very first AI use case wasn't the most exciting slam dunk that I would want to go for, but it was something that had practicality and usefulness across a large scale, was relatively safe,
• 18:08
  didn't have to involve PII data, data risks, gdpr, stuff like that.
• 18:15
  It was basically a use case that was very, very low Risk.
• 18:18
  Even with that, still took a year to get through governance and compliance and everyone.
• 18:24
  Because AI was very new, right?
• 18:28
  Yeah, of course there's a huge risk to mitigate.
• 18:30
  Yeah, yeah.
• 18:30
  We need Gen AI use cases.
• 18:32
  So learned a lot from that experience, right.
• 18:35
  Like I'm sure the funny thing is a lot of people that were like leading those conversations and making those decisions didn't really know what AI was yet.
• 18:43
  Right.
• 18:44
  It was new to them, right?
• 18:45
  Not to their fault, new to me, new to them, new to pretty much everyone in our organization.
• 18:49
  So you know, they were probably using AI at home to come up with questions, right?
• 18:53
  Like we get questions every week like how do you control hallucinations?
• 18:58
  I'm sure I've seen the memes about it.
• 19:00
  What is the hallucination?
• 19:01
  Like what are you doing to, you know, like prevent that from happening?
• 19:06
  How's a human in the loop involved?
• 19:07
  So, you know, it was a very, it was a very interesting experience.
• 19:12
  Learned a lot from getting our first use case and then from those lessons learned, we're able to move quicker in the future.
• 19:18
  When we were scaling out to many more use cases, right.
• 19:21
  I think we grew to about 30 to 50 use cases, almost 30 in production.
• 19:25
  None of them were just different like rag models.
• 19:28
  Different like data being ingested into that rag model that we built.
• 19:32
  Right.
• 19:32
  So like we started with very simple use case for policy and procedures, eventually kind of evolve that to many different other systems.
• 19:40
  But like we learned a lot from that process to make sure like the next use cases that we're rolling out were more of like a one to two month governance process, if even that.
• 19:52
  Right.
• 19:52
  Because we've gone through before we built the framework, we knew how to move quickly, how to evaluate.
• 19:58
  I think the other big thing is, you know, when you're coming up with these use cases, making sure that you fail fast, quickly and determine ROI quickly.
• 20:08
  Right?
• 20:08
  Because you don't want to roll something out to production that's going to cost a lot more money to roll into production than it is going to save you in costs.
• 20:16
  So I think determining ROI upfront, when you're determining these use cases versus, you know, I think two years ago a lot of people just cared about how many use cases do we have, how many do we have in
• 20:27
  production?
• 20:28
  Right.
• 20:29
  They didn't really care about the roi.
• 20:30
  Now you're seeing a massive shift to we're spending billions of dollars on AI, right.
• 20:37
  Especially at an organization like that.
• 20:38
  I think 3 billion spent.
• 20:40
  Where's the ROI?
• 20:42
  I think that's where organizations are starting to lean into more now is okay, we have all these AI use cases.
• 20:49
  Is it generating revenue?
• 20:50
  Is it saving costs?
• 20:52
  Is it creating cost efficiencies?
• 20:53
  I think you've seen in the news a lot of companies have gone ahead and they have these metrics baked in at the top of the house.
• 20:59
  Like we're going to have a 15% reduction in force for AI, but they're reducing their workforce before AI even achieves what they're looking to do.
• 21:09
  And now they're scrambling because they're like, wow, AI is more costly than the people that we let go now.
• 21:15
  They're scrambling to, you know, like hire these people back.
• 21:18
  And I'm not talking about JP Morgan, I'm talking about just companies in general.
• 21:22
  Yeah, companies in general, yeah.
• 21:23
  So many companies.
• 21:24
  There's been some bold claims made for sure.
• 21:26
  Oh yeah, certainly across the, across the industry.
• 21:29
  Who was it Oracle recently laid off?
• 21:31
  What, 30,000 people?
• 21:33
  Yeah, meta as well.
• 21:35
  Meta made.
• 21:36
  I saw that story about Meta, like forcing employees to be monitored by AI and then using AI to replace them.
• 21:42
  That's a particularly brutal one.
• 21:43
  But you come to expect that from Meta, I think.
• 21:45
  Yeah, I have a lot of friends that work at Meta.
• 21:47
  You know, every, every month or two, they're always kind of like worried are they going to make the cut or what not.
• 21:53
  And yeah, I think it's across a lot of tech companies right now, life there.
• 21:57
  I think I like just one to shift some of the stuff we've talked about into context for my audience.
• 22:02
  So JP Morgan, when you were doing this gen rollout, obviously you've got a large team, you've got a large P and L behind you, but say maybe you're a smaller kind of mid market portfolio company that might only
• 22:14
  have 500k in budget.
• 22:16
  Right.
• 22:16
  For technology spend something much lower.
• 22:19
  What would the right approach be, do you think, to sort of translate the things that you've learned at J.P.
• 22:23
  morgan and take it and apply it in that environment.
• 22:26
  So they have 500,000 to spend.
• 22:29
  Yeah.
• 22:29
  So the point is that you've got a much smaller team and a much smaller budget.
• 22:33
  Like what still translates.
• 22:36
  Yeah.
• 22:36
  So you definitely need to get it right because you don't have a lot of money to kind of burn on failure and R and D.
• 22:43
  So I think first really kind of sizing where AI is going to have an impact depends on how the company does it.
• 22:50
  We were very big on process mining where we could kind of tell you, I could tell you maybe across 5,000 processes, end to end, what the baseline was before AI was implemented and then after.
• 23:03
  So you could look through each one of those digital processes and find out where there was inefficiencies, where there was the biggest time gap, where there was the most manual touch points, and then you could kind of break
• 23:15
  that down into, okay, this AI use case we're looking to implement, we estimate it's going to save 10 minutes, which equates into this much like cost savings.
• 23:25
  So I think, you know, with a smaller company you're going to definitely need to take a data centric approach and.
• 23:31
  Really be clear about the business case from the start.
• 23:33
  Right.
• 23:34
  And big companies too.
• 23:35
  Right.
• 23:35
  Because you see a lot of these big companies just burning cash on AI use cases.
• 23:39
  And I think a lot of them are coming from like the senior leaders or the business who.
• 23:43
  Everyone's got an idea now.
• 23:45
  Right.
• 23:45
  And I mean AI has kind of created an explosion of AI ideas where, you know, people are asking AI, where could I embed AI in the business?
• 23:54
  And it's giving them like a hundred use cases.
• 23:57
  And now you're at a point where like no one's really moving anything into production.
• 24:00
  They just have this entire backlog of use cases, right?
• 24:03
  Yeah.
• 24:04
  Side projects.
• 24:05
  Yeah, yeah.
• 24:06
  You got to be very careful of that trap as well and make sure like if you're a small or a company you're going to invest in AI, that you know what the costs are up front.
• 24:14
  You figured out like your infrastructure costs, you've put guard rails around like the token burns and everything and then, you know, really understand like what the net benefit is going to be.
• 24:26
  Yeah.
• 24:27
  So again, for the audience I work with, time is another thing that's important.
• 24:32
  Right.
• 24:32
  Particularly usually like the first 90 or 100 days after acquisition, we want to make some changes, we want to deploy some people and some technology and start expanding, growing the company or making it more valuable in some way.
• 24:45
  And just to set the scene for you, typically in mid market you're dealing with, you either have no technology team or a very nascent technology team.
• 24:52
  You have scattered data around across different commercial off the shelf products.
• 24:57
  You have not necessarily direct correlation between what the company's trying to do and what they're measuring.
• 25:03
  Like, so that's the general scene.
• 25:05
  Like if you were in that situation, you need to think, okay, what are some of the quickest ways to use AI to add value to a company in the first hundred days?
• 25:12
  What are some of the trees you think we should start shaking?
• 25:16
  Yeah, I guess it depends on the size of the company.
• 25:19
  I know with like mid sized companies and some of the smaller companies, they're looking to reduce a lot of their like back office type of operations.
• 25:27
  And that's where I think AI does a really good job.
• 25:29
  Yeah, there was a lot of manual processing and spreadsheets going on.
• 25:32
  Yeah, I would say that's typical.
• 25:33
  Yeah, there's a lot.
• 25:34
  But now that you even just that, I mean just like a matter of like, you know, like document intake and scanning, like processing of certain transactions.
• 25:43
  Like, I think there's a lot of areas in the back office where AI could definitely save a lot of money very quickly.
• 25:50
  And I think they're very standard across many different companies.
• 25:54
  Whether they're small, large, mega cap sized companies.
• 25:58
  I think there's a lot of areas where AI could be embedded there I think depending on like their sales funnels.
• 26:06
  I think also I think front office has a lot of benefit where like maybe you know, like lead generation, you know, out.
• 26:15
  Like I don't work very closely with the front office, but I'm thinking there's a lot of use cases, I think for like front office for like smaller companies where they could probably create a lot more like leads.
• 26:25
  Outreach, targeted outreach.
• 26:28
  Yeah.
• 26:28
  Marketing is an interesting one as well, you know, with.
• 26:31
  Yeah, it is interesting1 Because AI can produce decent marketing copy.
• 26:36
  But I've seen and you know, I run a small, medium sized business myself, right?
• 26:40
  So like I'm in this game, but I've seen that you can produce great volume of content.
• 26:45
  But one, it's not spectacular or unique in any way.
• 26:49
  And two, actually everyone's obviously pretty switched onto this, who's tight for resources and team.
• 26:54
  So it all starts kind of looking and feeling and sounding the same.
• 26:56
  Like you see it with the, you know, when there was people are building landing pages, right, Using, you know, cursor, whatever, you know, pick your vibe coding tool of purpose.
• 27:06
  Landing pages all started looking the same, right?
• 27:08
  They were all purple, they all had those rounded boxes on them, they all used the same typeface.
• 27:14
  It suddenly almost became an advantage to not use AI to do things.
• 27:18
  And in fact like with my own marketing copy, I've started like typing everything myself, like especially for LinkedIn for my newsletter because that's now the differentiation.
• 27:27
  Like yeah, you see a lot of.
• 27:29
  AI slop out there.
• 27:30
  Like it's all kind of become standardized, like web design, LinkedIn, post articles.
• 27:36
  That's a good point.
• 27:37
  I didn't think about that, but you're right.
• 27:40
  And like just through my process, like I do a lot of vibe coding but like I go through so Many iterations and I still kind of like draw out what I want.
• 27:48
  Like then I kind of like, you know, load it in and put it in figma.
• 27:52
  Like I, I do a lot of customization because if you went with something like I forget the tools.
• 27:57
  There's so many.
• 27:58
  There's like base.
• 27:59
  Oh yeah, that's right.
• 28:00
  Some other thing, I don't know.
• 28:02
  Lovable.
• 28:03
  Yeah.
• 28:03
  And I've seen some of that.
• 28:04
  Like a friend of mine is starting like a, like she wants to spin up a dating app that has a very unique aspect to it and she built her mvp.
• 28:13
  What's the unique aspect, can I ask?
• 28:15
  It's, I can't say because I don't want to like ruin her.
• 28:18
  Like, you know.
• 28:20
  Okay, we're pre launch stealth.
• 28:23
  Yeah, it's like, it's kind of like where EO Harmony was trying to go a long time ago.
• 28:29
  Right.
• 28:29
  But like I think it's using more like scientific psychology.
• 28:33
  Oh, can I go and date it?
• 28:34
  I know like, you know, like how people really like.
• 28:36
  I think a lot of people lie to themselves on like what they want and who they're compatible with.
• 28:40
  And I think social media has kind of tailored people towards like, hey, this is the life you want.
• 28:46
  These are the people you want.
• 28:47
  Like where I think they're not really kind of getting to the root of like what they really need and want.
• 28:52
  So it's more of like a target to help you find your person versus like just getting into the swipe game of constantly swiping the same people or not.
• 29:00
  It's unique but like when you get into like the MVP that she built and lovable like it's absolutely like just throw away like I would have to rebuild the whole thing from the back end to the ground up.
• 29:12
  It's like, it's just a lot.
• 29:14
  You still have to think like a silver engineer.
• 29:16
  This is what I always tell people.
• 29:17
  Like I have clients sometimes show up with front ends, pre built front ends saying, oh look, I already built it.
• 29:21
  I'm like, you know, building the backend data structure behind this and then wiring it together so that everything like that's the actual hard bit.
• 29:28
  Like you like you changing the color of the buttons on the front end is not difficult.
• 29:32
  And never.
• 29:33
  And it never has been.
• 29:35
  Never has been.
• 29:37
  And I think when you use tools like that you don't think of a lot of those things.
• 29:40
  Like so I used Cursor originally to build my website and then some apps that I've built personally and I over engineered the whole path of production.
• 29:49
  Where I built a true CI CD pipeline, where every step of the way it is going through rigorous checks to determine should this move to production or not.
• 30:00
  I can tell you, over 50 deployments, I haven't had a single one fail in the last six months.
• 30:05
  Right.
• 30:06
  But people aren't thinking about that type of rigor when they're building this AI slab.
• 30:10
  Like they're just throwing in there, you know, gonna roll it out.
• 30:13
  The production they're not thinking about.
• 30:15
  It works once for one set of inputs.
• 30:16
  Yeah.
• 30:17
  Maybe PII data gets leaked out or personal financial data gets leaked out into, you know, GitHub for everyone to access.
• 30:25
  Like especially for a dating app.
• 30:27
  Right?
• 30:28
  Yeah, you would like those are a lot of things that you have to think about.
• 30:32
  So you can't really.
• 30:33
  I think AI has enabled a lot of people to come up with these MVPs, if you will.
• 30:38
  Everyone can kind of create apps, come up with business ideas, but unless you have experts who know how to scale and build those and architect them from a enterprise grade, I wouldn't bank on them.
• 30:50
  So I think that also too for a lot of your PE companies that are starting to use these tools like cursor and Claude code and whatnot, like they need to kind of start thinking about like best practices from
• 31:03
  an architecture standpoint, like for back end data security, things like that and having the people.
• 31:08
  On staff who know how to do that stuff.
• 31:10
  Yeah.
• 31:11
  Because many of them don't.
• 31:11
  Right.
• 31:12
  Is I think where people are getting stuck.
• 31:14
  I'd say a lot of them don't.
• 31:15
  In this whole mindset that you're not going to need architects, you're not going to need engineers in like eight months.
• 31:20
  I think Elon Musk is saying that recently, like I get where they think you're going to shift to this like master product owner who is like an expert in every single area.
• 31:31
  But I don't think it's these amazing.
• 31:34
  For deployed engineers, the Palantir, these data.
• 31:38
  Scientists unicorns that we were looking for back in Capital One space.
• 31:42
  Right.
• 31:42
  Like you, there was like four areas where you truly had to be like a top tier data scientist.
• 31:49
  Right.
• 31:49
  You need to have the business knowledge and be an expert in the business.
• 31:53
  How do you take this idea?
• 31:54
  How do you take these models and convert them into roi?
• 31:58
  Are you an expert in deploying your own infrastructure?
• 32:01
  Are you a master coder in SQL Python?
• 32:04
  R.
• 32:05
  Right.
• 32:06
  And do you have the advanced statistical knowledge what data scientists are Now?
• 32:11
  They might be good in one or two of those areas, but I still very, I struggle to find data scientists who are masters in all four of those areas.
• 32:19
  It's almost like data scientists were just unicorn and then they continued to dilute what it was considered to be a data scientist.
• 32:27
  And now you just, you know, basically have people who are just like data engineers.
• 32:31
  Yeah, they did stats at college, they did stats at college and other stats.
• 32:34
  In college and they took a Python course.
• 32:36
  Yeah, they go, oh, data scientists, that's me.
• 32:39
  Is, yeah.
• 32:39
  It's easy to get the label, isn't it?
• 32:40
  For sure, yeah, yeah.
• 32:42
  Give someone label now than give them the money like this.
• 32:45
  For deployed engineers, I think they're even more ridiculous than the data scientist thing we're talking about because like we're talking about And I think OpenAir and Anthropic are promising.
• 32:54
  They have these four deployed engineers as well.
• 32:56
  So these are people who are expert software engineers, expert data engineers, expert product managers, and great at talking to executives and understanding what it is they actually want compared to what they're saying they want.
• 33:06
  Nobody can do all, I don't know anyone who can do all those things.
• 33:09
  I don't know anyone who can do all those things.
• 33:11
  So I'm sure there's some people lacking in a few areas and I know some incredible software engineers in my time, but would I put them in front of a CEO of a mid market company?
• 33:21
  Probably not.
• 33:21
  Very rare.
• 33:23
  Probably not.
• 33:23
  Yeah.
• 33:24
  They're like a needle on a haystack to find someone who's that technically adept and then put them in front of a CEO to speak it in like lamest terms.
• 33:33
  Yeah, in terms that they'd understand.
• 33:35
  Yeah, exactly.
• 33:36
  I want to close with this interesting question because I know you're, you're free to talk about it to some extent right now more than you have been before.
• 33:44
  Like I know from your great Instagram content, Brandon, that you know stocks in private markets well and you know, you've also, you're also someone who's built great data infrastructure inside loads of different companies.
• 33:55
  So right now, like from a private equity perspective, if someone is looking to put private money into a company, what are some of the things they might miss about data when they're evaluating a company?
• 34:06
  I think a lot of people right now like you can see it with like the SpaceX acquisition, right.
• 34:11
  A lot of people buying it at like 220 where like a lot of people that I know were like shorting it at that price.
• 34:17
  Not necessarily shorting it, but definitely like they knew it was way overvalued.
• 34:21
  Right.
• 34:21
  I, I think companies have long term,.
• 34:25
  Did you get any?
• 34:26
  I didn't get in.
• 34:28
  Now it's looking more, more.
• 34:29
  No vanity purchase for you?
• 34:31
  No, no vanity purchase for me.
• 34:33
  I, I've learned with a lot of these IPOs that they, they seem to skyrocket in the beginning through all the hype and then they kind of come down to like where they're at.
• 34:41
  But you gotta look at the long term potential.
• 34:44
  I think a company like SpaceX has huge long term potential.
• 34:47
  But is the revenue generation there, is the ROI there right now?
• 34:52
  Probably not.
• 34:52
  It's going to take some time to get there.
• 34:54
  So I think with like, I don't even know if we might have to scrub this part.
• 34:58
  I don't even know if I can talk about like stocks and what I would invest in.
• 35:02
  Oh, it's not, just say it's not financial advice.
• 35:04
  Brandon.
• 35:05
  We're fine.
• 35:05
  Yeah, okay.
• 35:06
  Definitely not giving financial advice.
• 35:08
  You have to look at like, you know, one of people are investing in like the long term value of a company now like they're looking like five to 10 years out.
• 35:16
  And that's like what's so difficult about like picking the correct like tech companies and where they're going to go.
• 35:21
  Like I would have never thought two years ago that Micron or Sandisk would be trading at where they're at.
• 35:28
  So will they probably come down to Earth?
• 35:31
  Who knows?
• 35:32
  Probably eventually, right?
• 35:33
  Or maybe they'll just keep going.
• 35:34
  But like, you know, when I'm looking at these companies, I'm really looking.
• 35:38
  I don't invest unless it's something that I personally am passionate about.
• 35:45
  And I think that's why I've been successful in all my picks for a very long time is I've always invested in things that I truly believed had value.
• 35:53
  And that you wonder and that you seek to understand to a level where you understand that value.
• 35:57
  Right.
• 35:58
  It's not just a hype train certainly back to SpaceX.
• 36:00
  I will buy it eventually because I do believe in the future of space exploration and what it could do for, you know, especially when you talk about like mining, mining asteroids and all the.
• 36:10
  Like, you know, the rate they're churning shuttles out is incredible.
• 36:14
  What is the, the, the rate that they're manufacturing space shuttles at is incredible.
• 36:18
  Incredible.
• 36:18
  Bandwidth of that production plant is wild.
• 36:21
  There's hardly any room for error.
• 36:23
  Like I, I feel like they're so efficient and they continue to evolve.
• 36:27
  I mean if you look at like the rockets where they were like a couple years ago when they were first developing the rocket boosters into where they are now they truly took MVP and made it best in class.
• 36:42
  And that's what I think a company, if I was investing in a company like SpaceX or a lot of these startups, I would look at what they consider excellence, where they're headed, where they could go 10 years down
• 36:57
  the road.
• 36:57
  I think there's, you know, Kathy woods tries to do that and I think she was really successful in the beginning at doing that.
• 37:05
  Is like really kind of looking at like where are companies headed in a 10 year horizon and invest in those companies because it's going to take them a while to get there.
• 37:12
  Especially when you're talking about like these PE backed companies.
• 37:16
  I mean the pe, but like I'd say expected hold period for PE is about five years.
• 37:19
  So you know, we're looking on a similar time horizon.
• 37:22
  Yeah.
• 37:22
  Maybe slightly shorter, but there's no doubt about it.
• 37:26
  SpaceX, their efficiency and their knacking for excellence and perfection is unmatched.
• 37:35
  And I have no doubt they've invested in data and I have no doubt that they're AI ready.
• 37:38
  Right.
• 37:39
  So I had a guest last week talking about AI readiness and the investment that has to go into the people, the processes of the culture before you even get to the technology platforms and the data, before you can
• 37:50
  actually truly use AI as a competitive advantage.
• 37:53
  I'll end on this because I knew we're out of time but you know, one mistake I made in my career.
• 37:57
  Right, interesting.
• 37:58
  Speaking of SpaceX, they reached out to me in 2017 for a tableau, like a tableau like developer role and you know, I, I think I, I didn't even entertain it because I didn't want to move to California.
• 38:11
  Still don't want to move to California.
• 38:12
  That's what's like a lot of tech companies.
• 38:14
  I've turned down Meta, a lot of others because I'm just like, I don't want to live in that state.
• 38:18
  It's a beautiful state, I love to visit but the cost of living would kill me, regret it because like now that they're in Texas it's like, man, like SpaceX would be a dream to work for just the stuff
• 38:27
  that they're innovating and what they're doing.
• 38:30
  Like you could be a part of like real technological advances that are like unmet anywhere else.
• 38:37
  So.
• 38:37
  Yeah.
• 38:38
  And whatever your opinions on the, the, the value of the stock right now or our IPO 2017 stock would have certainly been worth having.
• 38:47
  Yeah, definitely.
• 38:47
  I mean, you know, look at, look at people working in the cafeteria becoming millionaires.
• 38:51
  Like, overnight.
• 38:52
  Like, you know, so definitely, definitely a company that's going places.
• 38:57
  That's what it's all about.
• 38:58
  So cool.
• 38:59
  Brandon, thanks so much for coming on.
• 39:01
  You're a man who's going places as well, so great to catch up with you.
• 39:05
  Likewise.
• 39:06
  And great to hear everything that you've been up to.
• 39:08
  And thanks for sharing all the audience.
• 39:10
  Much appreciated.
• 39:11
  And yeah, best of luck with whatever is on the horizon next for you.
• 39:15
  Thank you.
• 39:16
  I appreciate it, Graham.
• 39:17
  All right.
• 39:18
  Good to see you.
• 39:19
  Bye.
• 39:20
  Thanks for listening to the PE Data guy.
• 39:22
  The place where private equity meets data.
• 39:25
  Please forward this episode to your favorite private equity friend.
• 39:28
  Thanks for listening.
• 39:29
  See you next time.