Latest / Private Equity Data Guy / Private Equity Leaders Could Not See What AI Was Doing

Transcript

Private Equity Leaders Could Not See What AI Was Doing · Private Equity Data Guy

• 0:00
  Resistance to AI is almost never about the model.
• 0:03
  You know, from an employee's perspective, it's about their work becoming visible.
• 0:07
  And from an organization's perspective, it's about visibility into what AI is doing.
• 0:11
  The question isn't what AI cut this year, it's where the next dollar of value is and whether we're positioned to see it before our competitors do.
• 0:18
  If your information is scattered across individual inboxes and, you know, in people's heads, AI is not going to fix that.
• 0:25
  It's just going to make confusion faster, faster.
• 0:28
  Behind every value creation plan, there's a data problem nobody wants to talk about.
• 0:33
  Fragmented systems, metrics nobody trusts, and decisions made on gut feel dressed up as analysis.
• 0:38
  Welcome to the PE Data Guy.
• 0:40
  Each week, host Graham Crawford talks to the operating partners, advisors and practitioners who are doing the work inside portfolio companies.
• 0:47
  If you care about what actually drives returns in the market, then you're in the right place.
• 0:51
  The PE Data Guy starts now.
• 0:55
  When I first met Phil Murphy, he was sitting on a panel at an ACG National Capital event.
• 1:00
  And it was a fascinating panel because it was contrasting the differences between a founder, CEO, staying post, private equity acquisition, and a CEO, which in this case was Phil being placed into a company after acquisition.
• 1:11
  And they talked about the many different ways that it could work.
• 1:14
  So Phil joined a company called Synergy ecp.
• 1:17
  But I was super impressed with the way that Philippines Phil talked about bringing a new culture to an organization.
• 1:22
  And we're going to get deep into this today, but in particular, implementing AI in the back office and implementing AI properly in a way that truly realizes the competitive advantages of it.
• 1:32
  So I was super keen to have Phil on the podcast.
• 1:34
  It's taken us a couple of months of calendar back and forth to get him here, but here he is, Phil Murphy, the president and CEO of Synergy ecp.
• 1:41
  Welcome to the show, Phil.
• 1:43
  Thank you, Graham.
• 1:43
  Really appreciate the time.
• 1:44
  I'm glad we were finally able to get the calendars to work out for this.
• 1:47
  Yeah, exactly.
• 1:48
  Back in the same place, albeit virtually this time.
• 1:50
  So private equity is such a fascinating vertical to exist in, and I think there are very few teenagers who, you know, are in junior year of high school saying, I want to be in private equity.
• 2:00
  Maybe there are, I don't know, but I haven't met them.
• 2:02
  So everyone's always got a circuitous and wonderful route that took them there.
• 2:05
  So tell us a bit about your journey and how you ended up in the wonderful world of pe.
• 2:10
  Yeah, I was definitely one of those teenagers.
• 2:12
  One of the Normal teenagers who knew nothing about private equity didn't know it existed.
• 2:16
  My background is technical.
• 2:18
  My degree is information systems with a specialty in forensics.
• 2:22
  When I got out of school, I went straight into, you know, on contract technical delivery and spent the first close to 10 years of my career doing individual contributor technical work.
• 2:33
  Right.
• 2:33
  So I've always been, you know, hands on with the mission.
• 2:36
  I've been in, in customer spaces, I've worked in the skiff.
• 2:40
  I understand what it, what it takes and what the, you know, employees that are on site go through.
• 2:44
  So I think that really helps from part of the culture perspective and what I've now transitioned into my leadership style.
• 2:50
  But there was a shift somewhere along the line when I was on contract one day where I realized I was doing fine work from a technical perspective.
• 2:58
  But my skillset, for whatever reason really lines more in finding the right team to solve a larger problem than what I can solve as an individual contributor, hands on keyboard.
• 3:08
  And at that point I started to shift a little bit more down the business path and, and you know, project manager, program manager and kind of eventually into a vice president role of running an actual portion of a
• 3:18
  business.
• 3:19
  So I spent the first four and a half years of my career at a company, you know, Booz Allen Hamilton.
• 3:24
  We then jumped into a small company at the time called Strata Solutions.
• 3:28
  A handful of us moved over.
• 3:29
  We grew that company from probably 15 people to about 100 when, when we were acquired.
• 3:35
  And when we were acquired, we were acquired in, I think it was December of 2018 and we were acquired by a private equity firm.
• 3:42
  That was my first introduction into private equity.
• 3:44
  I didn't know it existed.
• 3:45
  I didn't know what it was.
• 3:46
  Into the deep end.
• 3:47
  Exactly.
• 3:48
  I didn't understand this concept of, you know, rolling up acquisitions onto a platform to make a larger entity and eventually exit it one day.
• 3:55
  But honestly, I fell in love with it.
• 3:57
  I, I fell in love with the concept of exactly that, the M and A concept behind rollups and you know, the having more capital behind you to do real investments.
• 4:05
  You know, when we're, when we were a small business, we had a lot of great ideas, but we're like, man, this, you know, that's, this is coming out of our pockets out of a founder owned company at the
• 4:13
  time.
• 4:13
  Whereas with private equity the risk is still real and you still have to be very mindful of that.
• 4:18
  But there's more capital behind the scenes to actually drive more impact, which I have found to be very interesting.
• 4:24
  So introduced into private equity in 2018 we ran that entity until we exit that entity to CACI in November of 24 or October of 24.
• 4:36
  So you've actually sat in both seats then.
• 4:37
  You were the private equity, you were the founder based leadership team that then you then transferred to PE and then stayed.
• 4:44
  Yes, I've done both, both buy and sell side with PE firms, which has been great.
• 4:49
  And then so once we exited to caci, I came over here and joined Synergy, which is owned by Valfurious Capital Partners, headquartered out of Charlotte.
• 4:58
  Love it.
• 4:59
  Tell us a bit more about Synergy.
• 5:01
  Yeah, so today we are solely focused on the intelligence community and Department of War.
• 5:06
  We specialize in software and software development, infrastructure and data engineering and cyber oper with, you know, my technical background and some of the leadership team that I've brought in.
• 5:17
  Since we've been here or since I've been here, we've been heavily focused on, on permeating technology through the company, through the culture shifted, the type of work that we go after from a prime perspective, kind of shifted some
• 5:29
  of the vehicle, the contract vehicles along with obvious government, government mandates lately, you know, this administration around, buying solutions versus, versus manpower.
• 5:39
  So we're, we're about 240 people today, about 100 million in revenue.
• 5:43
  Most of our work, I'll call it 70% of our work is at Fort Meade, about 20% of our work is in Westfields, and then 10% or so is Cybercom.
• 5:51
  Yeah, it's been a busy year for the Department of War as well, right?
• 5:54
  It sure has.
• 5:55
  It sure has.
• 5:56
  Yeah, indeed.
• 5:57
  I'll, I'll, I've got a bunch of, like, I've, you know, living in the, the D.C.
• 6:02
  metro area.
• 6:02
  I have a bunch of friends involved with, you know, confidential, secure information and government work.
• 6:08
  And I always, I've, I've learned to park the questions that I want to ask, that I know where the answer is, that I know that they can't be answered.
• 6:14
  But.
• 6:14
  Yeah, we'll save.
• 6:14
  That for a different.
• 6:16
  Yeah, we've been quite busy this year along with all of our peers in the industry around Epic Fury and things of that nature.
• 6:22
  So it's been an interesting year.
• 6:24
  Yeah, absolutely, absolutely.
• 6:26
  I definitely want to talk about AI and one of the things that stuck in my mind from your panel presentation was the way that you had implemented AI.
• 6:33
  And I think that, forgive me if I'm paraphrasing Roland, quoting directly here, but you said you didn't pilot AI.
• 6:39
  You mandated it.
• 6:40
  And do you know, I love that approach because I think pilots are sometimes A weak excuse that leadership teams can give for like, stop bothering me about this.
• 6:48
  I don't want to do it.
• 6:48
  I just want to satiate you and I have no intention of doing it further.
• 6:51
  So actually, when I do pilots with my clients, I'm always very clear of like, this is a fail, succeed experiment.
• 6:56
  We are defining what failure and success is and what the subsequent actions are before we start.
• 7:00
  Pilots need very careful management like that.
• 7:02
  Otherwise you end up with multiple pilots and they.
• 7:04
  And they go off.
• 7:05
  But you went straight for the mandate, right?
• 7:07
  And what did that mandate actually say and how did it land?
• 7:11
  Yeah, we did.
• 7:12
  And I mean, to your point, honestly, I think pilots are how organizations decline to decide at the end of the day, right.
• 7:18
  You get a volunteer, deadlines that nobody enforces.
• 7:21
  You get a report that just has a promise in it.
• 7:24
  From my perspective, I wasn't interested in a promise.
• 7:26
  I was interested in again, changing operationally how the company functions and operates from a back office perspective all the way through delivery.
• 7:33
  So the mandate was simple.
• 7:34
  Every functional leader had to take at least one core reoccurring workflow.
• 7:39
  Real world, not demo work, just something they do every day and run it with AI and come back with a baseline they started from and then, you know, use a sanction path so that it was visible, we could
• 7:49
  see what it was.
• 7:50
  What I deliberately left open and didn't do was the how.
• 7:54
  Didn't tell them which workflow, which tool, how to approach it.
• 7:58
  That was all theirs.
• 7:58
  I didn't want a central team designing workflows for functions that they don't run.
• 8:03
  People doing the actual work, they know where the friction is.
• 8:06
  Right.
• 8:06
  Like, I don't need to tell them where.
• 8:08
  Where friction in their job is.
• 8:09
  My job is more take away the option of not doing it, not leveraging AI, and hand them the answer.
• 8:14
  We're going to use AI.
• 8:16
  You just figure out what to do.
• 8:17
  The question of what broke is interesting because it's not technology.
• 8:20
  The technology worked on day one.
• 8:21
  You know, at the end of the day, AI is a mirror.
• 8:24
  You can only run a process that you can describe on data it can find.
• 8:27
  Yeah, human APIs, I like to call them.
• 8:29
  Exactly.
• 8:30
  So, you know, when we looked closer, it went further than that.
• 8:34
  We had the same functions, just not doing the same job the same way.
• 8:38
  For example, staffing.
• 8:40
  Great example.
• 8:41
  Same function, same titles, different processes.
• 8:43
  It's not a technology problem, it's a standardization problem.
• 8:46
  Again, the AI can only be as good as the consistent, the least consistent version really of the process.
• 8:50
  It's Looking at.
• 8:51
  So very quickly the mandate became, without building consistency, one process, one data structure per function, because that's turned out to be the real ceiling of what the tool could do for us.
• 9:00
  So it was interesting.
• 9:01
  It wasn't a technology problem, it was a process problem.
• 9:03
  You know, from a, from a resistance perspective, it actually really didn't come from where I expected it to.
• 9:09
  It wasn't about the tools.
• 9:10
  It came from functions with the most, you know, tribal knowledge underneath it.
• 9:14
  It was fair for everybody.
• 9:16
  I get it, right, that for years most people's work has been judged on output and effort when suddenly all this was going to be visible.
• 9:23
  Right?
• 9:23
  Cycle times, volumes, exceptions.
• 9:25
  Yeah,.
• 9:25
  Absolutely.
• 9:25
  Yeah.
• 9:26
  It's a real thing, right.
• 9:27
  It's a real thing to ask of people.
• 9:28
  And you know, once we were honest, that the visibility into the work was the point, not the capacity or visibility and capacity point, not the headcount.
• 9:36
  Right.
• 9:36
  This wasn't a.
• 9:37
  This wasn't a, let's see if the AI can do it faster than you.
• 9:40
  Answer.
• 9:41
  That so that we can.
• 9:41
  Shuffle you out the door.
• 9:42
  Yeah,.
• 9:43
  Exactly, right.
• 9:44
  I mean, sure.
• 9:44
  So what I have seen is, you know, resistance to AI is almost never about the model.
• 9:49
  You know, from an employee's perspective, it's about their work becoming visible.
• 9:53
  And from an organization's perspective, it's about visibility into what AI is doing.
• 9:57
  Those are really kind of the two areas that you see.
• 9:59
  So the mandate actually surfaced things that had nothing to do with AI, you know, duplicate systems doing the same job of two functions, manual reconciliations that only existed because two systems never talk to each other.
• 10:10
  People who are already using AI on their own, actually, before I ever even mandated went out.
• 10:16
  Yeah.
• 10:16
  And.
• 10:16
  I guess this brought that into visibility.
• 10:18
  That work then becomes visible and shared.
• 10:20
  Right.
• 10:20
  Rather than siloed,.
• 10:21
  You know, and then really the last one like that, the finding out that people were already using it, that's really what was the start of something bigger.
• 10:29
  And exactly what I meant when I said that an organization's resistance to AI is visibility into what AI is doing.
• 10:35
  Yeah, absolutely.
• 10:36
  And the thing I love about the mandate as well is not specifying a certain tool that does one first.
• 10:43
  It gives people the creative freedom and puts the honest back on them to design the way they want to design it.
• 10:48
  But the second thing, and this is important to me, any company that has its context and its data and its processes that are well set up for AI, high quality, well governed, consistent, actually changing a model or changing
• 11:01
  an AI LLM that you use or whatever else actually has Very little impact because if the context is set up outside the model and not stored in the model memory, first it's way more efficient and secondly, it shows
• 11:10
  you've thought about the right things in advance.
• 11:11
  So just to pick a random example, if the Department of War suddenly says you can't use Anthropic anymore, then if you're set up and your context and your data structure is autol, then actually transitioning becomes a lot easier,
• 11:21
  I found.
• 11:22
  Yeah, I think for most people's usage of AI, right.
• 11:25
  I mean, at the end of the day models are most likely going to be plug and play.
• 11:28
  Obviously there's specialized models that are better at some things than others and I get all that.
• 11:33
  But to your point, if Department of War goes through with it and says Anthropic is no more, you know, pulling Claude out in the various cloud models and you know, shifting over to OpenAI or Bedrock or whatever it
• 11:43
  may be, it should be seamless at the end of the day because the structure that's in place and the process and the data sets are really what, what matters more.
• 11:50
  Yeah, exactly.
• 11:51
  You know, the LMS are tool on top of it.
• 11:53
  They are not the, the, the be all and end all or the means to an end.
• 11:56
  They are the, the thing you use to take advantage of your well ordered data in my book.
• 12:00
  So running AI through the back office, I can see how that would land differently to different eyes.
• 12:05
  So, you know, if, if P buys a company, you go in and you establish this and you're making things efficient and more visible, transparent, then it can be received well.
• 12:12
  If P is buying a company and that's already in place, maybe that's received differently because you don't know that it's done well or it might create some suspicion.
• 12:19
  So how do the, how does it land differently in a board pack versus a diligent room?
• 12:23
  Yeah, it's a great question.
• 12:25
  My goal at the end of the day, and at least the way it is with, with Valfurious, is is both conversations are actually focused on innovation and industry positioning.
• 12:33
  You know, are we far enough ahead from a corporate adoption and usage perspective to create real advantage versus what I think happens a lot of times with this conversation and where some of the negative connotation comes in is
• 12:44
  it's focused on margin and cost reduction.
• 12:45
  Right.
• 12:46
  From a fal furious perspective, nobody across the table in any of my board meetings, you know, have asked, hey, did AI cut some costs?
• 12:53
  No.
• 12:53
  The questions they're asking are whether or not we're far enough ahead on adoption.
• 12:57
  To be finding value before anybody else has mapped it.
• 13:00
  You know, for us, it starts with the philosophy just AI is an enabler.
• 13:04
  It is not a replacement for people.
• 13:06
  We think about it kind of in two buckets, which is really efficiency and innovation.
• 13:09
  You know, the first efficiency is, is what I said earlier.
• 13:12
  It's doing the work we already have and we've been doing it, but doing it with less linear GNA growth.
• 13:17
  Right.
• 13:17
  So top line can grow without the back office having to grow right alongside it, but also, but also without putting an outsized burden on that existing back office staff.
• 13:27
  Staff.
• 13:28
  Right.
• 13:28
  So that's table stakes.
• 13:29
  Do more with less.
• 13:30
  But it doesn't mean you.
• 13:31
  We have to cut.
• 13:32
  It's.
• 13:33
  You know, Phil, we should definitely clip that bit because it's like it's, it's a very common fear.
• 13:38
  And yet you're talking from the PE side and the Portco side.
• 13:41
  I've been a service provider across multiple PE's and port calls.
• 13:44
  I have yet to come across someone who has said, how can we use AI to get rid of people?
• 13:49
  But that's the stigma.
• 13:51
  I've not seen.
• 13:52
  It once, I've not heard it once across multiple opportunities.
• 13:55
  To have heard it in places where always.
• 13:57
  And this is the thing I love about pe, the focus is how do we grow the value of this company.
• 14:02
  So you'd think that harder edge might create it more, but I'm yet to see it.
• 14:07
  You know, I mean, from an innovation, I agree with you.
• 14:09
  Right.
• 14:09
  I mean again, from the board side, my sponsors never asked me for headcount or seat count.
• 14:15
  Right.
• 14:15
  Again, what they watch, quarter to quarter is more of a GNA and percentage of revenue, how much incremental GNA takes to support next dollar revenue.
• 14:22
  And that's where AI from our perspective really shows up.
• 14:24
  You know, I mentioned to efficiency and innovation.
• 14:26
  So from an innovation perspective, you know, our team, we're utilizing AI to capitalize on new technical capabilities and open new growth areas.
• 14:34
  Right.
• 14:34
  And in a diligence room, that's the bucket that matters more because it's evidence, you know, we're creating new value, not just trimming cost.
• 14:41
  And that's really like the key in my opinion here, the path that actually kind of the part that surprised me was the integration, right.
• 14:48
  As we're buying companies and back offices are integrating.
• 14:51
  There's, there's this integration engine.
• 14:54
  An acquired, acquired company shows up.
• 14:56
  They've got their own systems, their own process, their own knowledge.
• 14:59
  With the mandate to, to force us to standardize integration actually became a Data problem instead of a headcount problem.
• 15:06
  And data problems at least have a known playbook right now.
• 15:08
  It's also a means that we can take an acquisition and we can plug it into our current AI adoption curve much faster instead of just starting from zero and trying to figure out what, you know, what do we
• 15:18
  do with this new, this, this new entity.
• 15:20
  So when it comes to an exit, you know, I, I, my goal is we don't just leave with ebitda, we obviously we lead with EBITDA and current market positioning.
• 15:27
  Right.
• 15:27
  So the question isn't what I cut this year, it's, it's where the next dollar of value is and whether we're positioned to see it before our competitors do.
• 15:35
  You know, a one time cost takeout is a one time story.
• 15:38
  A demonstrated, you know, govern capability to keep, to keep finding efficiency and innovation ahead of the market.
• 15:44
  You know, that's what compounds and at the end of the day the multiple isn't paid for what you cut, it's paid for how far ahead you can see.
• 15:51
  Yeah, well the even simpler and even less useless take that I see people take on it is like oh, we're using AI as some sort of indicator of potential future value without the recognition of the fact that if
• 16:03
  you're using AI in the way that many people do of like oh it's just a replacement for Google search just a little bit smarter.
• 16:08
  That's never a competitive advantage because everybody else is doing that.
• 16:11
  Like if you're not building proprietary first party data or building proprietary capability inside your company using these tools that nobody else has got, then it's not a competitive advantage.
• 16:20
  And that's, I struggle to get that across the people sometimes.
• 16:22
  Yeah, I couldn't agree more.
• 16:23
  I mean in this case just you know what, I kind of have lifted to this point in time and gaining efficiencies and just back office daily administration.
• 16:31
  That's table stakes at this point.
• 16:33
  To your point.
• 16:33
  Everybody, if they're not doing it yet, everybody will be doing that.
• 16:37
  Or even add it onto the end of your company name.
• 16:39
  Phil, if I can may suggest synergy, ecp, AI then you know, that's worth at least a multiple.
• 16:46
  There you go.
• 16:47
  I'm going to write that down.
• 16:48
  Graham.
• 16:48
  I.
• 16:48
  Appreciate that.
• 16:50
  I've seen people try it.
• 16:51
  I've seen people try it for sure.
• 16:53
  So talk about the enterprise AI governance capability that you're launching believe as a product.
• 16:59
  Right.
• 16:59
  What did you see inside your own walls that convinced you that correct visibility and governance was the product?
• 17:05
  Yeah.
• 17:05
  So this is where it gets fun.
• 17:07
  Again, everything I said to this point was really more the mandate was around, how do we gain efficiencies, how do we, how do we again, not grow the back office in a linear fashion as we grow the company?
• 17:18
  But to your point, that's everybody's doing that or is going to be doing that.
• 17:21
  You know, how do you take it one step further?
• 17:23
  So for us, it was actually, you know, it came down to one simple question.
• 17:27
  You know, somebody asked, I think it was our COO asked, well, you know, where is AI actually being used, by whom and on what data?
• 17:34
  And I couldn't answer it.
• 17:35
  I mandated everybody to use it, but then I couldn't tell you what they were doing.
• 17:38
  And it's not because anybody did anything wrong.
• 17:40
  The usage was, was widespread, multiple.
• 17:42
  Yeah, they were doing what.
• 17:43
  You asked.
• 17:43
  Dozens of tools, each with its own console, and none of them talking to each other, none of them producing a record that I could see or hand to anyone.
• 17:50
  And I asked them to do this to your point.
• 17:52
  So, you know, at the end of the day, I run a company whose business is built on being trusted with sensitive work, and I couldn't answer the question of what is AI doing and who's doing what with AI.
• 18:01
  So it's not that the mandate created the problem, it just actually happened to reveal a problem that we then took internal and said, let's take a look at this problem.
• 18:10
  So, you know, everything that we have done here and everything I'm talking about is, has been 100% done on our back office corporate data.
• 18:17
  This is not something that we've, we've pushed out to customers just yet.
• 18:20
  And it's intentional, right?
• 18:21
  We wanted to take this capability, we wanted to deploy it inside of our, our back office in production and actually run with it and, you know, more or less eat our own dog food for, for lack of better
• 18:31
  term.
• 18:31
  So the biggest takeaway and really what we were seeing inside, though, like what we're seeing inside of our walls and inside of our back, matching what we've heard from customers.
• 18:40
  And if you remember, you know, being on the technical side, back in the early cloud days, developers suddenly had a console.
• 18:48
  They could spin up whatever data, you know, whatever they wanted whenever they wanted.
• 18:52
  Providers were shipping features at an incredible rate.
• 18:55
  In my opinion.
• 18:56
  AI is at that exact moment right now.
• 18:58
  Everybody's trying to figure out how to adopt it, how to adopt it safely, effectively, efficiently, you know, and it's changing daily, literally daily, as it continues to change.
• 19:09
  This problem is actually getting more and more Urgent.
• 19:11
  So with generative AI, you can review the answer.
• 19:15
  Everything for the most part is safe.
• 19:17
  With agentic AI, the tool starts to take actions.
• 19:20
  You can't go back and review every action that it already executed because it been executed.
• 19:25
  Visibility is where the key became.
• 19:27
  Right.
• 19:27
  And that's gotta be built in up front and not audited after the fact that.
• 19:32
  So that's actually, that's exactly what we were focused on and that's where we're helping.
• 19:36
  Right.
• 19:37
  So we operate, you know, in environments where we can't deploy something we can't explain, you know, deny by default, record everything, produce evidence on demand.
• 19:45
  Like that's the price of admission at this point, supporting.
• 19:48
  You're so ahead of the game with it as well, because you've done that work.
• 19:51
  And then when was it that Jev came out, like two months ago that Jev made a big splash on the scene as a, as an AI powered decision engine.
• 19:59
  And it's jumped like OpenAI and anthropic into releasing things that do the same.
• 20:02
  Muse effectively does the same from Meta.
• 20:05
  And now that we're in this world where AI is making more and more decisions for people and I actually worry for the general public and their data security, wonderful as I am, because I think people are signing up
• 20:15
  to give it.
• 20:17
  Like for example, you log into Muse, the first thing it wants is access to your entire Gmail history.
• 20:22
  Like that is a big.
• 20:23
  That's a big deal.
• 20:24
  Yeah,.
• 20:25
  That's a big deal.
• 20:26
  Like job applications, health records, finance statements, all sorts of like tax returns.
• 20:31
  What's in there?
• 20:32
  It's, it's, it's crazy.
• 20:34
  And, and these agents are making, have gone from, you know, summarizing or creating bullet points or make my email less angry or whatever it is that people are.
• 20:44
  You know, we have been using AI for, to actually making a choice and making a decision on behalf of a business.
• 20:49
  So it's critical that.
• 20:51
  So kudos to you for being way ahead of the game on that one.
• 20:53
  Oh, I appreciate it.
• 20:54
  And again, the team's done a phenomenal job and being able to kind of identify this shortcoming and realizing that, hey, okay, this does tie.
• 21:02
  And back in those previous entities, we actually built a solution and a product that was around helping companies with cloud adoption.
• 21:09
  And really at the end of the day we're doing the same thing here.
• 21:11
  But it's from AI adoption, right?
• 21:13
  Because at the end, like these requirements are starting to convert everything you just said from, from, you know, government policies.
• 21:19
  And then commercial world is starting to get in it and you're the burden of proof is, is on the company to, to say hey, this is exactly what happened and this is, this is what we do and this
• 21:29
  is what we're, we are doing and this is what is happening in our environment.
• 21:33
  So really for us it came down to like what does full visibility actually mean?
• 21:37
  And we look at it in really four things.
• 21:39
  So the first is inventory.
• 21:40
  You know, what AI tools, models, agents are actually in use and by whom.
• 21:45
  An actual real list.
• 21:46
  It's, it's the approved list but then it's also enforced and it can't go around that second's activity.
• 21:53
  Every request, every tool, call, data that crossed the boundary where it went everything.
• 21:58
  Third is policy.
• 21:59
  What what is being allowed, flagged, blocked, what's against the rules?
• 22:04
  What's you know, what's happening?
• 22:06
  Not, not a policy document but actually like an enforcement record.
• 22:09
  Because really, I mean I think we all know this right at the end of the day, policy without enforcement PDF.
• 22:15
  I like that phrase.
• 22:16
  Fourth is, is evidence.
• 22:17
  Right?
• 22:18
  So the fourth one's evidence isn't.
• 22:19
  This is durable tamper evident trail.
• 22:22
  You can hand it to an auditor, a customer, a diligence team without a crazy three week fire drill of holy crap, what happened.
• 22:29
  We can literally just report a, this is the report that shows everything that, that every environment or every AI that's happened in our environment, it's right here.
• 22:36
  So actually what we learned is the fifth thing you kind of get for free out of this is you find out what functions are actually using it and how they're using it, which is really, that's ties back to
• 22:44
  the operating the company more efficiently and more focused.
• 22:47
  So it's been great.
• 22:48
  And at the end of the day this is what we built.
• 22:50
  We built, we call it watch for it sits between the people and AI models.
• 22:54
  It sits between the models and the tools are reaching out to.
• 22:56
  So every request is inspected, enforced against policy that you said it's recorded.
• 23:01
  Its job is to observe, govern and assure.
• 23:04
  But that's what it's focused on.
• 23:05
  You know, we built it because we needed it.
• 23:07
  We demoed it, you know, to a handful of customers at this point and early AI adopters and validated my earlier comments that you know, organization's hesitation to adopt AI right now is visibility into what's happening.
• 23:18
  Keep going back to cloud.
• 23:19
  Just like hesitant to adopt cloud until they had a solution for controlling what users can do from a spend and an understanding of what they're doing in the cloud.
• 23:27
  Right.
• 23:27
  And especially from our, from our, our do and our intel customers, right, they.
• 23:31
  They start to.
• 23:32
  We're all hearing the stories and seeing the stories of agents running wild, you know, from, you know, the various frontier models.
• 23:38
  And the last thing that anybody in the agency wants to answer upstream to is, what'd you guys do?
• 23:44
  What happened?
• 23:45
  Yeah, there's a couple of things to unpack.
• 23:48
  One, I want to resonate your Cloud point, because this all happened during my time when I was at Capital One.
• 23:53
  Anyone can spin up a server and anyone can expand processing, whatever else, and suddenly you go from like, oh, we're literally spending millions and millions of dollars a month on aws.
• 24:02
  We should find a way to govern and control this.
• 24:04
  And there was actually, without doing a pitch on behalf of my former employer, actually they turned it into a product called Cloud Custodian, which is now available by AWS to keep control of it.
• 24:14
  That came out of Capital One.
• 24:16
  So the very same pattern.
• 24:18
  The interesting thing, we're dealing with large enterprises who had access to this technology to use technology to make decisions.
• 24:25
  Now everyone has got that access.
• 24:26
  And I think there's a couple of things coming up on the rails, the first of which I think is a huge glut of legal liability cases because we're going to get a way down the pipe before the first
• 24:38
  big thing blows up, goes to court and someone tries to say, oh, the AI decided to do it, and then they find out that they're the ones who are responsible for the decision because they gave it the context,
• 24:47
  they gave it the decision matrix and they just, you know, ultimately they decided to do it to draw back to another Capital One example.
• 24:53
  The credit risk models are so carefully change, controlled, guarded and understood so that any credit decision, you can say, this is why this was made across these number of factors, and you can draw everything back to the reason
• 25:06
  why, because this and effectively results, because this is what we told it to do.
• 25:09
  So I think there's a bunch of legal liability things coming up on the rails that we're going to see over the next couple of years where AI is no defense.
• 25:17
  Right.
• 25:17
  And someone needs to be just like data, data, just like technology.
• 25:20
  Someone needs to be in control of the governance and responsible for the decisions that are being made.
• 25:24
  Absolutely.
• 25:25
  This is.
• 25:25
  You're spot on.
• 25:26
  This is going to be a very, very interesting next couple of years and just the various cases and scenarios that we're going to see play out, just like you're saying.
• 25:35
  So again, we keep going back to, we think visibility and the full picture into what's happening is critical.
• 25:39
  It's going to be needed.
• 25:40
  There's no doubt about it.
• 25:42
  Yeah, I won't go off too long on this tangent, but there was a.
• 25:44
  My wife alerted me to this fun thing that Amazon is doing.
• 25:46
  If you go into your Amazon account, apparently it will tell you that.
• 25:49
  It will tell you the things that it thinks about you.
• 25:51
  Like, this is where you live.
• 25:52
  We think you're married, we think you have two teenagers.
• 25:54
  We think there's a toy and it reels off like a list of 30 things, most of which are true.
• 25:57
  Although also had to answer, I had to answer some questions because one of mine said you have a girlfriend.
• 26:01
  I was like,.
• 26:05
  Oh,.
• 26:06
  Here's my defense.
• 26:07
  Here's my legal defense.
• 26:09
  My, my wife kept her name when, when we got married.
• 26:11
  Is one and, and two birthdays and Christmas.
• 26:14
  I am absolutely the one ordering all the gifts.
• 26:17
  The, the, the lady gifts on behalf of all my children, both my children for the, my wife.
• 26:23
  So that's, that's my defense.
• 26:24
  But as I said to her, like, people think it's a neat party trick of like, oh, we worked out you've got a dog because you order dog food.
• 26:30
  I said to her, like, do you realize this is just the very thin top layer of everything that Amazon knows about you?
• 26:36
  Like they are, they are just flicking the light switch on for half a second to a whole room of stuff that they know about you.
• 26:42
  It's, it's kind of incredible.
• 26:44
  Yeah.
• 26:44
  And to your point earlier about Gmail and everything, just imagine what Google knows.
• 26:48
  Oh yeah, all right.
• 26:50
  Yeah, exactly.
• 26:51
  Well, let's hope everyone behaves responsibly with it.
• 26:53
  Governance can be counter to innovation.
• 26:55
  Right.
• 26:55
  People can position it as an enemy of progress.
• 26:58
  Like, they'll go, oh, all this governance.
• 27:00
  And I'm used to this in the data world as well.
• 27:02
  Like, oh, we have to define all our terms and like someone has to be responsible for every data set.
• 27:06
  What a drag.
• 27:07
  But if you want to mandate adoption and establish governance at the same time, like, how do you make that balance and how do you help people understand that one pays for the other?
• 27:17
  Yeah, it's a great question.
• 27:18
  And you are right.
• 27:18
  I mean, governance done incorrectly absolutely can be a hindrance.
• 27:23
  You know, in this case, honestly, I don't think really one paid for for the other.
• 27:28
  It really was a symbiote, symbiotic relationship.
• 27:30
  So adoption gave governance something real to govern.
• 27:34
  Right.
• 27:34
  And governance is, is what led me to keep saying yes to, to the adoption that we're doing.
• 27:38
  If you pull one out, then the other stalls I think it actually was a really good symbiotic relationship between the two.
• 27:45
  You know, a lot of, of what comes from how we think about innovation and synergy.
• 27:49
  Like we don't build hobby shops, we don't fund something because it's an interesting technology exercise.
• 27:54
  You know, we fund it because it solves a real problem.
• 27:56
  So watch floor is a direct result of that.
• 27:59
  It wasn't dreamt up in a lab, it was born because we needed to answer exactly where AI was being used in our company.
• 28:06
  And nothing at the time, nothing off the shelf could, could tell us that.
• 28:08
  And quite frankly, you know, these, these frontier providers and whatnot.
• 28:12
  I mean, making the models as as good as possible is far more of, of their focus than reporting.
• 28:19
  So their own built in reporting isn't going to tell you that.
• 28:22
  So really, I mean, the way I describe our approach to governance is, is guardrails, not speed bumps.
• 28:27
  A speed bump slows everybody down whether they're, they're the problem or not.
• 28:30
  Right.
• 28:31
  A guardrail is invisible most of the time.
• 28:33
  Only matters when you're about to go off the road.
• 28:35
  We never want to hamper production or take away capability from people.
• 28:39
  We just want to make sure it's being used responsibly.
• 28:41
  On sequencing.
• 28:43
  I'd be lying if I said it was a clean phase one, phase two, Right.
• 28:46
  Like they ran together almost from the start, but real usage produced real edge cases and those edge cases became kind of our governance model.
• 28:55
  If anything, usage got ahead because you can't govern something that doesn't exist yet.
• 29:00
  Right.
• 29:01
  So I think sometimes people have it backwards.
• 29:04
  Governance isn't what slows you down.
• 29:06
  What slows you down is a general counsel or a security lead who has to say I don't know what's happening.
• 29:12
  So therefore they default to no or.
• 29:15
  Generic governance.
• 29:15
  That doesn't really pay much heed to what the innovation is trying to do in the first place.
• 29:19
  It just restricts for restriction's sake.
• 29:21
  Exactly.
• 29:21
  So in this case, visibility turns that blanket.
• 29:24
  No.
• 29:24
  Into a very specific bounded.
• 29:26
  Yes.
• 29:27
  Because they know what's happening.
• 29:29
  Part of what we found is it was an interesting actually shadow AI situation.
• 29:32
  I mentioned it earlier.
• 29:33
  Right.
• 29:33
  So when we turned the lights on, we found kind of what everybody finds.
• 29:37
  There's, there's personal accounts, there's browser extensions, you know, tools the team picked up because they were trying to solve their problem.
• 29:43
  There's Excel spreadsheets getting uploaded to public models.
• 29:45
  Yeah,.
• 29:46
  There you go.
• 29:47
  And it's not, you know, for the risk wasn't that people were careless.
• 29:50
  It was that information was moving, you know, through paths and produced with no record.
• 29:54
  So we couldn't prove what happened or what didn't.
• 29:56
  And the fix wasn't, wasn't punishment.
• 29:58
  It was actually, I mean, it was making the sanction path easier than the workaround.
• 30:02
  Because shadow AI is a demand signal.
• 30:04
  Right.
• 30:04
  It's not a discipline problem, it's a demand signal.
• 30:06
  People were telling us exactly where the friction in their job was and they were going out trying to find solutions.
• 30:11
  Yeah,.
• 30:11
  They're trying.
• 30:11
  To.
• 30:12
  Be good employees.
• 30:12
  I get it.
• 30:13
  Yeah, exactly.
• 30:13
  So that's kind of like, and that's really, really what the mandate piggybacked on.
• 30:17
  Right?
• 30:17
  It was like, do that, do that.
• 30:18
  Just let's do it in a way where we're all watching and we can all see what's happening.
• 30:21
  And then from there we realized, all right, this is great.
• 30:24
  However, we need a way to always know what's happening.
• 30:27
  Yeah, yeah, it's a good one.
• 30:30
  And it's interesting that even in enterprise, I think people, you read about the ZDR at the zero data retention promise that you get from the providers, but actually if you dig more carefully into the terms.
• 30:42
  Yeah, I believe that they don't keep identifiable data.
• 30:47
  I believe that they do keep that promise.
• 30:49
  But if you look at the terms and conditions around it, if you put IP up there that they then reuse, then that's fair game.
• 30:57
  Right.
• 30:57
  And I think there was a math problem that got solved.
• 31:00
  I can't remember which of the frontier models it was that solved this math problem.
• 31:04
  But the human who wasn't at the Frontier lab who was trying to do it was working with, I think, GPT to do it.
• 31:12
  And while his data and the fact that it was him doing it wasn't used in the eventual solving of it by the frontier labor, the IP certainly was.
• 31:22
  The idea was, the approach was right.
• 31:25
  So you know, when people talk about the, oh, we're protected because it's got an enterprise account and it's zdr, you need to always understand how they're governing the data at their end as well.
• 31:35
  It's an interesting one.
• 31:36
  Anyway, sorry, I'll get off my, I'll get off my soapbox.
• 31:40
  I want to finish with some advice for someone who's about to embark on the journey that you've already taken.
• 31:46
  So a brand new PE backed CEO, just say that he's just been parachuted and he or she's just been parachuted into the business.
• 31:51
  200 person services firm, they've had no structure or Governance or policy around AI at all.
• 31:56
  Like where should they start if they're starting on Monday next week?
• 32:00
  Oh, great question.
• 32:01
  I mean, I think my advice to that is you pick the one thing that we're already doing manually every week, run it with AI, know what it looked like before, you know, do it where you can see it
• 32:15
  and review the results.
• 32:16
  Right.
• 32:16
  I mean that's, that's really where, that's essentially exactly where we started.
• 32:19
  The original ask was that, right, just use AI's enabler on your, your day to day.
• 32:23
  And in a short time we went from, from a handful of people in our back office using AI to, to just about everybody.
• 32:30
  It stopped being, you know, one tool that everybody uses.
• 32:32
  The same way people started building their own skills and workflows and tailored their usage to best fit their own jobs, not other organizations in the company.
• 32:40
  And that's not unique to us, right?
• 32:42
  Anybody can, can go down this AI adoption model just following the same path and fast forward right?
• 32:49
  Today, our back office usage ranges everywhere.
• 32:52
  From somebody getting a daily recap of their email and calendar, automatically scheduling meetings based on emails all the way to we have agentic workflow agents monitoring our production, corporate firewalls and doing our staffing candidate pipelines, right, like that
• 33:07
  range from personal productivity to autonomous agents running against an infrastructure.
• 33:11
  I mean that's exactly why visibility matters.
• 33:14
  You could not get to that range of usage and stay blind to it.
• 33:17
  It would, it just doesn't work.
• 33:19
  Those are some very important processes.
• 33:20
  You talk about that.
• 33:21
  Exactly.
• 33:22
  So you know, if I'm that, that CEO, you know, where do I start?
• 33:26
  I think it's again, high volume, repetitive, heavy workload tasks.
• 33:30
  Outputs can be reviewed before they go anywhere.
• 33:33
  From a services perspective, it's probably finance operations, you know, invoice and AR reconciliation, timesheet, expense review proposals, recruiting, you know, a human's already checking all of that work.
• 33:45
  So you know, a bad output maybe cost an, you know, a client relationship if you, if you try it somewhere else, you know, on data, I think the bar is probably lower than, than people think, but, but more
• 33:56
  specific than they think, right?
• 33:57
  So you don't need a data lake.
• 33:59
  You don't need to let anybody sell you that as a prerequisite.
• 34:01
  You need, you know, one authoritative place for each kind of data.
• 34:04
  This is what I mentioned earlier.
• 34:05
  If we started to figure out we had data scattered everywhere, right?
• 34:09
  You need permissions that actually mean something, something you can enforce, you know, and more or less your top five processes written down even in a mediocre level of detail.
• 34:18
  If your information is scattered across individual inboxes and, you know, in people's heads, AI is not going to fix that.
• 34:26
  It's just going to make confusion faster.
• 34:28
  Yeah, the same as you've got data quality issues as well.
• 34:31
  Like there's a number of, I think, nascent products out there that claim that AI can solve your data quality issues.
• 34:36
  But without the context of what your columns mean and what your company's trying to do, I don't believe that that can.
• 34:41
  I.
• 34:42
  I don't believe AI can tell you, you know, be more confident about your data than you can.
• 34:48
  I can just tell you what the data is and it'll start making assumptions and correlations.
• 34:53
  And if the data is bad, then the assumptions are going to be bad and the correlations are going to be bad.
• 34:57
  So it's important to have the data in the right spot and be good data.
• 35:01
  Right.
• 35:02
  And then, you know, I think for that, that CEO, and from the governance perspective, it really, it fits on one page, right?
• 35:07
  So an inventory of, of which AI tools are approved and who's using them, a clear line in the sand just for what data never goes into a model and what data sets, models can't access in plain language, something
• 35:18
  that you can actually enforce and apply, you know, and then a sanctioned path that just records and everything that happens from there.
• 35:25
  You need a monthly report to by usage, by function, whatever got flagged, blocked.
• 35:29
  If you do that, I mean, do that before scaling and, you know, you'll be able to answer the question I couldn't answer when we first started this path.
• 35:36
  So I do think it's, it's simpler than it sounds.
• 35:39
  But I mean, especially just AI in general is a, is a daunting problem.
• 35:43
  Not even problem, but it's a daunting topic because it's so expansive and it's changing so fast and it's being used in so many places, used correctly and incorrectly in so many places.
• 35:53
  And I think people that aren't, you know, heads down in it every day have a hard time deciphering what is used new and unknown to many things.
• 36:00
  Correct.
• 36:01
  Yeah.
• 36:01
  And I think I remind people like this, like folks like you and I feel, who spend our days buried in it, I think can sometimes take for granted the level of knowledge that us and our teams might have
• 36:10
  compared to like a layperson on the street or a new person and a new company you're going into.
• 36:15
  It may just be a thing that's shrouded in fear and uncertainty.
• 36:18
  Completely agree.
• 36:19
  Completely agree.
• 36:20
  I mean, I think that's human nature, right?
• 36:21
  Kind of.
• 36:22
  We're just naturally afraid of what we don't know or don't understand.
• 36:25
  So I think that's that applies here.
• 36:27
  Absolutely.
• 36:28
  Well, I wish you luck with Almost Journey.
• 36:30
  It sounds like you couldn't be set up better over a synergy for all the opportunities as well as the the risks to govern that lie ahead of all of us.
• 36:39
  So thanks so much for sharing that story.
• 36:41
  I'm sure it's valuable to a ton of people who are listening.
• 36:43
  Thanks for coming on and hopefully we bump into each other at the next ACG National Capital event and if not, sometime soon.
• 36:51
  It's awesome.
• 36:51
  I really appreciate the time, Graham, and thank you for reaching out and continuing to work the schedule.
• 36:56
  So appreciate the opportunity to talk about what we're doing and couldn't be more proud for what the company and what the team is really embraced AI and what we've done with it to this point.
• 37:06
  So it's been fantastic.
• 37:08
  Thank you.
• 37:09
  Thanks so much.
• 37:09
  Appreciate you coming on.
• 37:10
  Thanks for listening to the PE Data guy.
• 37:13
  The place where Private Equity meets data.
• 37:15
  Please forward this episode to your favorite private equity friend.
• 37:19
  Thanks for listening.
• 37:20
  See you next time.