[00:00:04] Welcome to another episode of Tech
[00:00:05] Unhinged where tech gets human. I’m your
[00:00:07] host Rabia Javeed. Then joining me today
[00:00:09] is Joel Grossman, chief information
[00:00:11] officer at IMG Academy where he oversees
[00:00:14] technology for both the physical campus
[00:00:15] and digital platforms. Joel would spend
[00:00:17] over 25 years in leadership including um
[00:00:20] CTO roles at Legacy.com, I prospect and
[00:00:22] nearly a decade at leapfrog online
[00:00:25] serving Fortune 500 clients. He’s also
[00:00:27] founder and mentor at Tech Stars
[00:00:30] Chicago. Welcome to the show, Joe.
[00:00:32] >> Thank you very much, Rabia. It’s a
[00:00:33] pleasure to be here.
[00:00:33] >> You are the CIO at IMG Academy, right,
[00:00:36] which is one of the world’s leading
[00:00:37] sports training campuses, and you also
[00:00:39] oversee NCSA College Recruiting, which
[00:00:41] is probably the world’s uh the largest
[00:00:43] college recruiting platform in the
[00:00:45] country. For people who work in
[00:00:46] traditional tech roles, what does a
[00:00:48] CIO’s day-to-day actually look like when
[00:00:50] your users are like, you know, elite
[00:00:53] athletes, coaches, and families
[00:00:55] navigating college recruiting? And how
[00:00:57] is it different from leading tech at a
[00:00:59] SAS company or enterprise?
[00:01:01] >> Yeah, it’s certainly very different and
[00:01:03] right now is a is a unique moment. Um, I
[00:01:06] spent most of my career, as you
[00:01:07] mentioned, building technology platforms
[00:01:09] and and digital products for sort of
[00:01:12] internet scale consumerf facing
[00:01:14] businesses. So this was the first time
[00:01:16] uh about halfway through my my tenure
[00:01:18] here at the academy where I had to
[00:01:21] confront not just a a very different
[00:01:23] kind of environment in terms of who’s
[00:01:25] present and and who my stakeholders are
[00:01:27] but a very much 24/7 inerson
[00:01:30] environment. So day-to-day uh the CIO
[00:01:33] role uh is one where I’m operating in
[00:01:36] close partnership with a few different
[00:01:37] folks. uh the CTO who oversees the
[00:01:40] overall academyy’s uh technology
[00:01:42] strategy and is a little bit more upward
[00:01:43] and board facing. My focus is really uh
[00:01:47] in the last 8 months or so has been very
[00:01:49] much on campus. The day-to-day role of
[00:01:51] the CIO really uh in a business that is
[00:01:55] one half roughly remote and kind of
[00:01:58] largely uh organized that way. That’s
[00:02:00] the precursor entity uh that we today
[00:02:03] call the online business in the in IMG
[00:02:06] Academy. uh the core of which was NCSA
[00:02:08] and another business called Sports
[00:02:09] Recruits we acquired uh recently and and
[00:02:12] several other new ones that we built.
[00:02:13] That’s very much a remote business. The
[00:02:15] IT portions of that world are are pretty
[00:02:18] well defined and and well structured.
[00:02:20] They’re kind of the traditional back
[00:02:21] office businesses of SAS or tech enabled
[00:02:23] services business where the focus is
[00:02:25] largely on making sure that the product
[00:02:28] and engineering teams have the tools
[00:02:29] that they need. Um which is a pretty
[00:02:31] straightforward stack for for for
[00:02:33] software engineering. uh a large sales
[00:02:35] and marketing function or commercial
[00:02:36] function has the tools that they need.
[00:02:38] And then somewhat uniquely uh in what we
[00:02:40] do in in recruiting and developmental
[00:02:42] coaching, we have to make sure that a
[00:02:45] different class of coaches have the
[00:02:47] tools they need to interact with student
[00:02:49] athletes seeking to play sports in
[00:02:50] college or develop their specific
[00:02:52] talents. And that’s really about
[00:02:53] internal tools and applications that
[00:02:55] provide context and guidance on what’s
[00:02:57] the next best thing to help this family
[00:02:59] or help this student athlete on their
[00:03:01] journey towards development goals or or
[00:03:03] recruiting. The campus business where my
[00:03:05] focus has really been the last year um
[00:03:07] is is an interesting and entirely
[00:03:09] different beast. I like to describe the
[00:03:11] campus environment um we we call it one
[00:03:13] of one and it truly is. The way that I
[00:03:15] describe it is it’s like if you are
[00:03:18] running Hogwarts at the Olympic Village
[00:03:19] inside Disneyland. It is a a totally
[00:03:22] unique environment. It is several
[00:03:25] hundred acres, many many buildings,
[00:03:27] world-class facilities for training and
[00:03:30] playing sports. There is a hotel, there
[00:03:32] is a golf club, there are events for
[00:03:35] tens of thousands of people there
[00:03:36] several times a year. Professional
[00:03:38] athletes and teams and leagues are there
[00:03:40] frequently to to train and play. It is a
[00:03:43] small town in its own right. And so
[00:03:45] given most of my background in history
[00:03:47] has been in in kind of developing
[00:03:49] software behind the scenes for people
[00:03:51] that I don’t get a chance to see,
[00:03:53] especially at this moment of of frontier
[00:03:55] models kind of re reshaping how
[00:03:57] technology works. Being able to interact
[00:03:59] on a day-to-day basis with a lot of
[00:04:01] people delivering, creating um
[00:04:04] facilitating excellence in young people,
[00:04:06] young athletes. It’s just a wonderful
[00:04:08] thing to be able to do and and frankly
[00:04:10] quite a lot more fulfilling than sort of
[00:04:13] the pure development of software for
[00:04:15] strangers has been for me. Um, not that
[00:04:17] that isn’t fun and I’ve had many many
[00:04:19] years of wonderful time uh building
[00:04:21] teams and and building software and
[00:04:23] products to do that, but at this moment
[00:04:25] with with Frontier Models, uh, what’s
[00:04:27] working on on campus is is it’s very
[00:04:30] wide, right? We’re building enterprise
[00:04:31] systems. We’re building infrastructure.
[00:04:33] We’re building platforms for our our
[00:04:36] customers who are student athletes and
[00:04:38] their parents and for other visitors to
[00:04:41] campus. So, there’s a lot of of
[00:04:43] day-to-day swings back and forth. But I
[00:04:45] think to sort of generalize across the
[00:04:47] use cases, it is about knowledgeable
[00:04:49] practice, right? We have coaches, we
[00:04:52] have athletes, we have people who are
[00:04:53] recruiting coaches and specialists in
[00:04:55] the college college space all with lots
[00:04:57] of deep tacid expertise about how these
[00:05:00] things work. Uh and my job is to really
[00:05:02] instrument them with the tools they need
[00:05:04] to do that work more efficiently and
[00:05:06] make sure that uh their specific
[00:05:09] expertise is is developed uh and
[00:05:11] maintained and delivered as as
[00:05:13] efficiently and effectively as possible.
[00:05:14] Also Joel, you joined in 2020 as their
[00:05:17] first ever CTO when the company was
[00:05:20] already 20 years old. What was the state
[00:05:22] of the technology when you walked in and
[00:05:24] what was the biggest gap you needed to
[00:05:26] close at that time?
[00:05:27] >> Yeah, so Robia, the the history there is
[00:05:30] is interesting. So INMG Academy today,
[00:05:32] like I said, has several different
[00:05:33] businesses and I came in through that
[00:05:35] Greek purser business NCSA which as you
[00:05:37] said was 20 years old at the time. Most
[00:05:39] of the latter part of my career has been
[00:05:42] walking knowingly into situations like
[00:05:44] this. I like to describe myself as a
[00:05:46] restructuring CTO. And restructuring
[00:05:48] being both in the sense of kind of M&A
[00:05:51] or corporate development changes, new
[00:05:53] investors, new ownership, new product
[00:05:55] strategy, new corporate strategy, new
[00:05:56] market dynamics, and then trying to
[00:05:58] reconnect or connect in the first place
[00:06:00] the technology function to achieve a
[00:06:02] higher degree of technical leverage in
[00:06:03] the business. So that means dealing with
[00:06:06] vast amounts of technical debt. It means
[00:06:08] dealing with cultural and organizational
[00:06:09] dynamics throughout the organization,
[00:06:12] senior leadership teams, mid-level
[00:06:14] management, folks in the trenches
[00:06:16] writing writing software or building
[00:06:18] configuration and frankly walking in um
[00:06:20] I think it was a known mess. So it was
[00:06:22] it was very clear that the technology as
[00:06:26] a as an asset needed a great deal of
[00:06:28] work. It was very very functional in the
[00:06:30] context of a business that grew on a
[00:06:32] meteoric basis for for 20 years straight
[00:06:34] um and continued to grow as I arrived.
[00:06:36] But the scalability of that business and
[00:06:38] its ability to develop more and new and
[00:06:40] differentiated products as the
[00:06:42] competition increased and sort of the
[00:06:44] sophistication of parents and student
[00:06:46] athletes increased that was a material
[00:06:47] block. You know the first thing
[00:06:49] obviously is is a a large scale
[00:06:51] assessment. Um and one of the most
[00:06:52] common patterns uh and one that we
[00:06:54] applied we used the term the replay to
[00:06:56] describe basically to say look we have
[00:06:59] repeated ourselves several times we have
[00:07:00] invested in multiple different layers of
[00:07:02] technology or products. Let’s be you
[00:07:04] know ruthless in assessing which of
[00:07:06] those things have scale which of those
[00:07:08] things have legs and frankly which
[00:07:09] increase the optionality of the business
[00:07:11] right to move in particular directions
[00:07:13] whether that’s specific features or
[00:07:16] capabilities or subsets of the market
[00:07:18] segments that we want to tackle in
[00:07:20] different ways make those technical
[00:07:22] plans supportive of the accretion
[00:07:24] accumulation of revenue and and and
[00:07:26] IBIDA in a way that scales um as opposed
[00:07:30] to you know the sort of because frankly
[00:07:32] the NCA They say business is great and
[00:07:34] remains great at generating a lot of
[00:07:36] outcomes from a recruiting performance
[00:07:38] standpoint. There is so much more that
[00:07:41] has emerged especially in the last 5
[00:07:42] years in terms of what we can do to help
[00:07:45] in recruiting and what we can do to help
[00:07:46] in athletic and personal development.
[00:07:48] And so being able to to create a set of
[00:07:51] digital products at a decreasing
[00:07:54] marginal cost was really kind of the the
[00:07:57] mission. And the fact that we were
[00:07:59] acquired twice in that first four years
[00:08:03] made it a fairly interesting exercise to
[00:08:05] kind of rebuild and create a platform
[00:08:08] from what was some fairly disconnected
[00:08:10] parts that were, you know, we’re working
[00:08:12] great in the context of a very
[00:08:13] salesdriven organization to facilitate a
[00:08:15] shift to more of a productled uh growth
[00:08:17] organization. That that was really the
[00:08:19] the mission and um and one that I’m I’m
[00:08:22] pleased we got done in our our ability
[00:08:24] to to release digital products.
[00:08:25] increasing sophistication and speed is
[00:08:27] is kind of where uh I handed things off
[00:08:30] to to our CTO, Gen 4. So, we’re making a
[00:08:32] ton of progress coming out of what was a
[00:08:34] frankly pretty messy uh state of affairs
[00:08:36] back in 2020.
[00:08:38] >> I can’t really imagine, you know, what
[00:08:40] was that mess like? But, you know, Joel,
[00:08:42] AI is everywhere in 2026, but in sports,
[00:08:45] it’s not just probably a buzz word. It’s
[00:08:47] embedded in how athletes they train, get
[00:08:49] scouted, and develop. for someone
[00:08:51] outside the sports world and or
[00:08:54] listeners in particular, can you paint a
[00:08:55] picture of where AI is actually being
[00:08:58] used today at IMG Academy or NCSA and
[00:09:01] what does AI in sports concretely mean
[00:09:04] in your environment?
[00:09:05] >> Yeah, so I think let me answer that more
[00:09:07] broadly across sport. Um, I’ve had the
[00:09:10] the privilege of really understanding a
[00:09:13] great deal the science behind
[00:09:15] performance and athletics, but more
[00:09:18] broadly beyond simply what we do at the
[00:09:19] academy. If you think about AI in sort
[00:09:23] of the the pre-large language model
[00:09:25] broader sense in which we began talking
[00:09:27] about it, there are a number of
[00:09:28] different ways in which we see the use
[00:09:30] of AI. So I think the first wave of
[00:09:33] things or maybe the the first one that I
[00:09:35] really had to confront was the use of
[00:09:37] computer vision and biomechanical
[00:09:39] analysis around movement tools,
[00:09:41] hardware, software, camera, wearables
[00:09:44] that track motion um and look at how uh
[00:09:48] athletes are moving in real time to
[00:09:50] understand what they can do, what where
[00:09:52] they’re going and and kind of how
[00:09:54] they’re performing in field and on
[00:09:56] training. So that’s a big part of of the
[00:09:58] landscape related uh beginning certainly
[00:10:01] in more in the professional world but um
[00:10:03] certainly making its way down to to
[00:10:05] youth athletic development in a big way
[00:10:08] would be looking at load monitoring and
[00:10:10] injury risk and um periodization which I
[00:10:13] can talk about in a little bit more
[00:10:14] detail but basically understanding
[00:10:16] streams of biometric data to look at um
[00:10:19] the level of training to prevent
[00:10:21] overtraining before breakdown injury
[00:10:23] prevention and remediation patterns of
[00:10:26] movement and physical stress that allow
[00:10:28] for optimization of training and
[00:10:30] recovery. Um, and lots of great work u
[00:10:33] being done by startups in that space
[00:10:35] right now. The NFL is doing some some
[00:10:37] great work with their digital athlete uh
[00:10:40] tool in the professional leagues. I
[00:10:42] think there’s also uh a huge amount and
[00:10:44] certainly we see this with our our
[00:10:46] online recruiting business of
[00:10:47] identifying talent and scouting at
[00:10:49] scale. So we have so much data on
[00:10:51] performance available in video feeds
[00:10:53] from you know companies like Huddle or
[00:10:55] Game Changer that are just you know
[00:10:57] sharing data publicly about particularly
[00:10:59] in youth athletics where are the great
[00:11:01] athletes and not just um the the
[00:11:03] standouts but where are the strong
[00:11:05] performers because as young athletes are
[00:11:08] developing in talent their ability to
[00:11:09] play in different contexts and fulfill
[00:11:11] different roles on school teams on club
[00:11:14] teams and on uh college teams and then
[00:11:17] eventually professional teams those are
[00:11:19] all really really um dramatically
[00:11:21] scalable capabilities. There’s also a
[00:11:23] lot of work as it relates to automated
[00:11:26] video. So this is a a huge area of
[00:11:29] utilization on campus in particular, but
[00:11:31] also in our recruiting businesses where,
[00:11:34] you know, we are looking at highlights
[00:11:35] and recaps of of performance and play
[00:11:37] and training and turning that into ways
[00:11:40] for student athletes to to different to
[00:11:42] differentiate themselves and show what
[00:11:43] they’re capable of doing, but also to
[00:11:46] kind of compress the workflow of
[00:11:47] understanding what could be done
[00:11:49] differently. How could performance be
[00:11:50] improved? How are teams playing together
[00:11:53] effectively or not? competing against
[00:11:55] different kinds of of opponents. I think
[00:11:57] those are the the big sort of
[00:11:58] generalized cases. I’m sure I’m
[00:12:00] forgetting several more. But as it
[00:12:02] relates to campus back in 2022 23 as the
[00:12:07] models started to develop and this is
[00:12:09] before, you know, February of this year
[00:12:10] where everything kind of changed with
[00:12:12] Opus 4.6. The way that we try to think
[00:12:15] about AI and the and the AI moment and
[00:12:18] this period of time in in technological
[00:12:19] development and sport is the
[00:12:22] accumulation of training data, right?
[00:12:23] So, while the large language models are
[00:12:26] superb and will continue to grow in
[00:12:28] their ability to process and and
[00:12:30] equipped with the right context, answer
[00:12:31] all sorts of questions at at greater
[00:12:33] speed and scale than ever before. At the
[00:12:35] end of the day, um, and as I think about
[00:12:37] a lot in terms of what’s happening in
[00:12:38] traditional SAS companies, it is those
[00:12:40] companies who have unique and
[00:12:42] proprietary data sets that can inform
[00:12:45] and and make the best use of AI in a
[00:12:48] variety of different contexts that kind
[00:12:50] of come out ahead. And so you know the
[00:12:52] academy uh is a place where we have both
[00:12:55] in the context of recruiting and in the
[00:12:57] context of athletic and personal
[00:12:58] development and you know we have this
[00:13:00] immense amount of data about what works
[00:13:02] and how we can help student athletes
[00:13:05] perform and and win their future and in
[00:13:07] that context I think it boils down to
[00:13:10] you know if you are collecting the data
[00:13:12] properly which we do for broad use cases
[00:13:15] it’s AI for facilitation so that’s the
[00:13:17] direct application of AI in in teaching
[00:13:20] and training and coaching developing
[00:13:22] student athletes. It’s AI for
[00:13:24] augmentation which is kind of the more
[00:13:26] traditional what we used to call machine
[00:13:27] learning but in different ways and and
[00:13:29] at different speeds leveraging AI for
[00:13:31] insights and analysis and automation to
[00:13:34] either generate net new capabilities for
[00:13:35] understanding or greater efficiency
[00:13:37] scale uh for for existing ones. Third
[00:13:40] would be for prediction. So, like I
[00:13:42] said, you know, the companies uh and the
[00:13:44] organizations that accumulate the most
[00:13:46] context and training data are the ones
[00:13:48] that tend to to win in the long term in
[00:13:50] in the current moment. And so, if we’re
[00:13:52] able to use a sum total of our collected
[00:13:54] data to identify potential outcomes or
[00:13:57] directions that might be helpful um for
[00:13:59] our customers and our business, then
[00:14:01] that’s super powerful. And then the last
[00:14:03] which I think is the most unique in our
[00:14:05] campus business where we have a very
[00:14:08] very close loop interaction where we can
[00:14:11] see how did what happened on the field
[00:14:12] today affect how we’re going to coach or
[00:14:14] train tomorrow. We can build AI models
[00:14:17] that move beyond prediction of what
[00:14:19] might happen into suggesting specific
[00:14:21] courses of action to our teammates our
[00:14:23] our coaches of whatever sorts. You know
[00:14:25] we have I think it’s something like 10
[00:14:27] coaches that are non-sport specific for
[00:14:28] every sport specific coach on campus.
[00:14:30] that’s for nutrition and mental
[00:14:32] conditioning, um, for, you know, injury
[00:14:35] recovery or for things like leadership.
[00:14:37] Um, and certainly we have, you know, a
[00:14:39] number of teachers. Those sorts of
[00:14:40] abilities to say, hey, here’s what we
[00:14:42] could do to increase this the the
[00:14:44] likelihood of a positive outcome for
[00:14:46] this particular student athlete. That is
[00:14:49] where I think we are at at the edge of
[00:14:51] what’s possible and um, and growing
[00:14:54] every day. So that’s that’s kind of
[00:14:55] where we are and it’s an exciting time
[00:14:57] because the cycles of developing those
[00:14:59] capabilities are are certainly
[00:15:01] increasing dramatically almost every
[00:15:03] every you know frontier model release.
[00:15:05] >> Oh I hear you and you know some great
[00:15:07] initiatives um that have been taken and
[00:15:10] done. But if you talk about the
[00:15:12] recruiting side there and here we would
[00:15:14] want to analyze a bit you know on the
[00:15:16] agent versus the human scout side. So we
[00:15:20] know that AI is changing the processes.
[00:15:23] Um you can talk a bit more about if that
[00:15:26] is happening at the academy right now or
[00:15:28] not but do you think that we are moving
[00:15:29] toward a world where algorithm
[00:15:31] identifies the next great quarterback
[00:15:33] before a human scout can do?
[00:15:36] >> I think that this is sort of the crux
[00:15:38] question and and one that we’re looking
[00:15:40] at all of us are kind of experiencing
[00:15:42] across a wide range of domains not just
[00:15:45] sports. My personal take is that barring
[00:15:48] really dramatic AGI style you know
[00:15:51] changes in the landscape there is an
[00:15:53] inevitable human dimension um as it
[00:15:55] relates to the use of AI in recruiting
[00:15:58] outcomes um and more broadly in kind of
[00:16:01] the athletic development uh life cycle
[00:16:04] uh in youth sports. I think that has a
[00:16:06] lot to do with the fact that at present
[00:16:08] there are dynamics in terms of um what
[00:16:11] often is called coachability that may
[00:16:13] not be obvious from performance data.
[00:16:15] You know, for individual sports it may
[00:16:17] be a little bit different in that, you
[00:16:18] know, that individual tennis player,
[00:16:20] golf player is a self-contained unit
[00:16:23] that you can look at in a number of
[00:16:24] different dimensions. But when we think
[00:16:26] about team sports, um I think it gets a
[00:16:28] lot different because number one, you
[00:16:31] know, the the tenure of a college coach
[00:16:34] just as one example is probably averages
[00:16:37] about a year. So, we we tend to think
[00:16:39] about in kind of the marquee sports like
[00:16:41] football and basketball, these beloved
[00:16:43] coaches who hang around for for a long
[00:16:46] time, but uh the truth is in in other
[00:16:49] sports, you know, you have a year, you
[00:16:50] come in as a coach, you’re trying to
[00:16:52] figure out what you have, what your
[00:16:53] recruiting capabilities are, what you
[00:16:54] need, who your graduating seniors are.
[00:16:57] There is certainly an immense amount of
[00:16:58] data and and it’s available more more
[00:17:00] rapidly than ever before to help you
[00:17:02] understand what a given student athlete
[00:17:05] might be able to do on the field. But I
[00:17:07] think there is a human judgment element.
[00:17:09] So that human judgment element is about
[00:17:12] the interaction between those players on
[00:17:14] the field or even on a team of
[00:17:16] individual in an individual sport. So I
[00:17:18] think AI is great at speeding up the
[00:17:20] process of identifying the pool of
[00:17:22] potential people to be interested in.
[00:17:25] But the element that we bring in our
[00:17:27] online recruiting businesses of people
[00:17:29] who have played those sports, who have
[00:17:31] traveled the road of what is involved
[00:17:33] for you to show up every day at 5:30 in
[00:17:35] the morning while carrying a full course
[00:17:37] load and you know training and competing
[00:17:40] whether at a a D3, a jo
[00:17:44] whatever level of effort and whatever
[00:17:46] sport that is still a fairly human
[00:17:49] judgment kind of domain and the
[00:17:52] recruiting coaches in in our business
[00:17:54] have that human experience. So I think
[00:17:56] it is, you know, what what at the very
[00:17:58] beginning of um the birth of large
[00:18:00] language models used to call sort of the
[00:18:01] the centaur mode where it’s a human but
[00:18:04] augmented with AI capabilities. That to
[00:18:06] me seems like where we are at present.
[00:18:08] Um the speed of the I think the the AI
[00:18:11] is the horse parts. The speed of the
[00:18:13] horse parts is probably increasing
[00:18:14] pretty dramatically. But I do think
[00:18:17] there is a human judgment in a specific
[00:18:19] domain. And more broadly, if if you were
[00:18:22] to assess the extent to which judgment
[00:18:24] and consideration is present in large
[00:18:26] language models outside of deterministic
[00:18:29] cases like say code bases that are well
[00:18:31] understood. I don’t think that we’re
[00:18:32] there yet. It doesn’t mean that we we
[00:18:34] might not get there, but I still think
[00:18:35] there is a tremendous amount of context
[00:18:37] and judgment that is experientially
[00:18:39] accumulated by people who have gone
[00:18:41] through the process that is unavoidably
[00:18:43] and and materially useful for for
[00:18:45] parents and student athletes navigating
[00:18:47] the recruiting process.
[00:18:48] >> Joel, how does leading technology for a
[00:18:50] mission-driven organization like IMG
[00:18:53] changes your priorities or how do you
[00:18:56] measure that kind of tech success? I
[00:18:58] spent uh the earlier parts of my career
[00:19:01] in and and sort of a core chunk of time
[00:19:04] you mentioned earlier my time at leaprog
[00:19:06] online work we’re building a technology
[00:19:08] platform and building a business in
[00:19:09] customer acquisition for for the better
[00:19:11] part of a decade where the mission was
[00:19:13] really the craft right I’ve had the
[00:19:16] privilege of working with a number of
[00:19:18] people who’ve traveled with me from
[00:19:20] organization to organization over you
[00:19:22] know three four five different employers
[00:19:25] where our focus on craft and and the way
[00:19:28] that we built software and the way that
[00:19:29] we operated a team was kind of the the
[00:19:31] mission and we were a tool given to a
[00:19:34] business to kind of create better
[00:19:35] outcomes for for fairly sizable clients.
[00:19:38] After that experience for me it became
[00:19:40] obvious that the actual thing to be done
[00:19:43] the meat of the work needed to play a
[00:19:45] more central role for me and um that was
[00:19:48] behind my decision to go work at
[00:19:50] legacy.com um where though in a a
[00:19:53] somewhat difficult period of time in
[00:19:54] people’s lives um at end of life the
[00:19:57] mission was was really about helping
[00:19:59] people in in their time of need and that
[00:20:01] was very very fulfilling though in the
[00:20:03] middle of co um certainly not the most
[00:20:05] upbeat job to have. All of this is to
[00:20:07] say I am extremely grateful every single
[00:20:11] day to wake up and particularly days
[00:20:12] that I am on campus to be working not
[00:20:15] only in a missiondriven organization but
[00:20:17] to be working in an organization whose
[00:20:20] focus is to help young people develop
[00:20:22] and particularly young people whether it
[00:20:24] is in what tend to be uh more elite
[00:20:27] athletes on campus or those who simply
[00:20:30] want the experience uh the very
[00:20:32] rewarding and developmentally like
[00:20:34] massively important uh experience of
[00:20:36] playing sports in college to achieve
[00:20:38] their futures. You know, we think a lot
[00:20:41] about certainly the financial
[00:20:43] performance of a business. We are owned
[00:20:44] by a large private equity fund and so
[00:20:46] like any other business, the efficiency
[00:20:48] of that business and its profitability
[00:20:50] are are certainly top of mind. But the
[00:20:52] beautiful thing about being in a
[00:20:53] missiondriven business like IMG Academy
[00:20:55] is that the better job we do supporting
[00:20:58] our teammates with technology, the
[00:21:00] better job we do helping them support
[00:21:01] our student athletes and our parents and
[00:21:03] families and our visitors to campus and
[00:21:05] anyone else that that is making use of
[00:21:06] our facilities and capabilities. We’re
[00:21:08] driving not only a social good in
[00:21:11] creating people who are better prepared
[00:21:13] to contribute to the organizations they
[00:21:14] become a part of and the society more
[00:21:16] broadly because of that experience, that
[00:21:18] focusing experience that sport has. I
[00:21:20] didn’t grow up playing competitive
[00:21:22] sports. I I wrestled in high school.
[00:21:24] I’ve kind of done Olympic sports to keep
[00:21:26] myself healthy. Big cyclist, do martial
[00:21:28] arts, but I did not have that team
[00:21:30] sports experience beyond wrestling um in
[00:21:32] high school. And I think in observing
[00:21:35] and interacting with student athletes
[00:21:36] both in our recruiting business and on
[00:21:38] campus and our our development uh
[00:21:40] digital product businesses, it is
[00:21:42] tangible. It is visible. It is
[00:21:45] recognizable watching people and young
[00:21:48] people in particular authentically
[00:21:50] commit to developing themselves and you
[00:21:54] see the excellence all around you. It is
[00:21:56] it is a thrilling thing to watch. So
[00:21:59] from a motivational standpoint,
[00:22:00] obviously, you know, waking up every day
[00:22:02] and knowing, hey, I’m helping this kid
[00:22:04] who is giving it their all at age 14, at
[00:22:07] age 12, sometimes at even younger in in
[00:22:10] sports like tennis and golf to become
[00:22:12] great, to try their hardest every day,
[00:22:14] and to have the honor of helping coaches
[00:22:18] who are allin every single day in
[00:22:21] helping, whether that’s in a sport,
[00:22:23] whether that’s in mental conditioning or
[00:22:25] strength training, or our dorm mentors
[00:22:28] who are providing leadership skills and
[00:22:30] kind of helping young people who are
[00:22:32] navigating the challenge of being in
[00:22:35] school of playing sports in a very very
[00:22:37] competitive environment and developing
[00:22:39] themselves that is a a privilege. So
[00:22:41] from an outcome standpoint I think being
[00:22:44] in a technology leadership role is about
[00:22:47] access and tools and you know do you
[00:22:49] have the software or hardware that you
[00:22:51] need to get the job done or increasingly
[00:22:53] like the data assets and and the AI
[00:22:56] capabilities too. We’re here to help
[00:22:58] produce an outcome. And that outcome is
[00:23:01] a social good. It is an economic good,
[00:23:03] but at the end of the day, it’s a human
[00:23:04] being that ends up, you know, with a
[00:23:07] level of preparedness to contribute to
[00:23:09] whatever they choose to do. And that’s
[00:23:11] that’s a beautiful thing to be able to
[00:23:13] facilitate. It keeps us honest and keeps
[00:23:15] us focused in a way that few other
[00:23:17] businesses I’ve I’ve worked in uh are
[00:23:19] able to to produce.
[00:23:21] >> No, no, I think that’s that’s a great
[00:23:22] way to um put it. Also Joel, when when
[00:23:25] you are trying to introduce AIdriven
[00:23:27] features on top of legacy
[00:23:29] infrastructure, what really breaks?
[00:23:31] What’s the biggest technical challenge
[00:23:33] in modernizing a 20-year-old platform
[00:23:36] without disrupting millions of users?
[00:23:39] Now, you can probably quote an example.
[00:23:41] >> Yeah, that’s a great a great question. I
[00:23:43] think the first wave of of really
[00:23:46] materially powerful um frontier models
[00:23:49] altered the dynamic a bit. in the
[00:23:51] earlier days and in the earlier models
[00:23:54] um there still was a widespread notion
[00:23:56] and certainly I think in the broader
[00:23:58] sort of senior leadership COOs chief
[00:24:01] commercial officers uh CEOs that we
[00:24:04] would still have to do the kinds of
[00:24:06] largecale data warehousing kinds of you
[00:24:10] know projects in order to achieve
[00:24:12] utility uh out of large language models
[00:24:14] and introducing AIdriven features and I
[00:24:17] think that remains partially true but
[00:24:19] the speed at which we’re able to
[00:24:21] overcome what would have been multi-year
[00:24:24] largecale you know assembly of data
[00:24:26] refining of data you know structuring of
[00:24:29] data problems is it’s it’s just it’s
[00:24:31] been a sea change in in a very very
[00:24:33] short amount of time so while I do think
[00:24:35] it remains the case that applying large
[00:24:38] language models either for analytics use
[00:24:40] cases or even feature building whether
[00:24:42] that’s kind of chatbased agents or
[00:24:45] different kinds of recommendation
[00:24:46] engines that are more large language
[00:24:47] model part the actual technical
[00:24:49] implementation are not particularly
[00:24:51] difficult. I think that for me the the
[00:24:55] hardest part is sort of a semantic
[00:24:58] abstraction issue, right? So we don’t
[00:25:01] really I mean for the scale of business
[00:25:02] that we’re in if if this was a startup
[00:25:04] and we’re starting from scratch would be
[00:25:05] a different thing. But as you mentioned
[00:25:06] our recruiting business was 20 years old
[00:25:09] um the campuses has been around for for
[00:25:11] even longer. So it’s not like a rip and
[00:25:13] replace situation. What we generally
[00:25:16] have done um as has been the case with a
[00:25:18] lot of technical debt uh scenarios I’ve
[00:25:20] dealt with is to sort of build a
[00:25:21] scaffolding on top of what exists and
[00:25:23] slowly replace pieces in order to make
[00:25:25] it more efficient to use in this case AI
[00:25:28] either to refactor or as sort of adding
[00:25:31] feature capabilities. I think the hard
[00:25:33] part is largely organizational not
[00:25:34] technical. So you have to sequence that
[00:25:36] change so the platform is not you know
[00:25:38] debilitated for any given set of users
[00:25:41] in the middle of adding those those
[00:25:44] features. We have found that the tasks
[00:25:47] that used to be more challenging in
[00:25:49] terms of articulating complicated data
[00:25:52] models that we could use or data
[00:25:54] structures and databases to enable
[00:25:57] answering questions or enabling
[00:25:59] particular kinds of features be they for
[00:26:01] recommendation or what we call in our
[00:26:03] recruiting business matching. You know,
[00:26:04] here’s a set of schools that you might
[00:26:06] want to consider based on your set of
[00:26:08] attributes as a student athlete or hey
[00:26:10] coach, here’s a set of of student
[00:26:11] athletes that you might want to look at
[00:26:13] and talk to. those sorts of features
[00:26:14] existed with older style of AI and kind
[00:26:18] of traditional machine learning
[00:26:19] techniques. So it wasn’t a huge shift to
[00:26:22] bring in large language models into that
[00:26:23] context. But what we have found in
[00:26:25] particularly on campus, this is this is
[00:26:27] relevant as we think about applying a AI
[00:26:29] to performance science and sport science
[00:26:32] is there are ways to integrate sets of
[00:26:35] data particularly kind of hardwarebased
[00:26:37] biomechanical measurements and you know
[00:26:39] where is this player on a field or
[00:26:41] combinations of hey here’s the GPS
[00:26:43] location of a set of players on fields
[00:26:45] in competition plus specific performance
[00:26:47] measurements that are you know literally
[00:26:49] on their body. Those are things that
[00:26:51] would have taken a long long time just
[00:26:53] to do the integration work before you
[00:26:55] could assemble a training data set that
[00:26:57] is now you know measured in weeks. So
[00:27:00] the large heavy lifts of assembling data
[00:27:03] in a warehouse or star schema those
[00:27:05] sorts of things that we did even 5 years
[00:27:06] ago we can kind of do reconnaissance by
[00:27:08] fire. We can find out where the
[00:27:10] mismatches are, where the missing pieces
[00:27:12] of data, where you know the set of
[00:27:14] things that we need to join together in
[00:27:16] order to answer questions are more by
[00:27:17] asking the questions of the data and
[00:27:19] then doing a lot of adversarial review
[00:27:22] of both the structure of the data, the
[00:27:23] way that we’re assembling the features
[00:27:25] and the outcomes uh in order to get um
[00:27:28] increasingly accurate and useful models.
[00:27:31] So it’s this it’s kind of this pushpull
[00:27:33] where just to use campus examples you
[00:27:35] know some of our leading data scientists
[00:27:37] in terms of performance and sport
[00:27:38] science are building new models every
[00:27:41] couple weeks we’re kind of figuring out
[00:27:43] how to assemble the data in ways that
[00:27:45] help them do that with greater
[00:27:46] efficiency which then results in them
[00:27:48] saying hey I can bring this new data set
[00:27:50] in or incorporate it in different ways
[00:27:52] and so we’re kind of going through this
[00:27:53] almost like I’m a cyclist so so bicycle
[00:27:56] you know the bicycle crank is going
[00:27:58] where their needs kind of generate
[00:28:00] assembly of data, preparedness of the
[00:28:02] model, training cycles which then result
[00:28:04] in in further advances their ability to
[00:28:06] model the data and kind of do analysis
[00:28:08] and that’s been the case in earnest for
[00:28:10] about you know the last 8 months at a
[00:28:12] really really fast pace. So it’s it’s an
[00:28:14] exciting time to be doing this
[00:28:16] >> and and recently you’ve also been
[00:28:17] transitioned from CTO to CIO. So you
[00:28:20] know expanding your scope to include
[00:28:22] both campus and digital operations. your
[00:28:24] focus has been on accelerating um
[00:28:26] technical leverage on the campus as well
[00:28:28] and that we’ve discussed at at length
[00:28:30] too, but what does that actually look
[00:28:32] like when you’re dealing you know with a
[00:28:34] physical facility serving 200,000 annual
[00:28:36] visitors versus you know taking care of
[00:28:38] a digital platform?
[00:28:40] >> Yeah, that’s been um I I would describe
[00:28:42] it as an unexpected pleasure um and
[00:28:45] challenge right. So I’m like we said
[00:28:47] earlier I tend to to do restructuring.
[00:28:49] So I think the campus environment like I
[00:28:52] said is is a very very unique one. Um
[00:28:54] and the earlier parts of my career I
[00:28:57] would never have imagined kind of being
[00:28:59] the CIO of what is effectively you know
[00:29:01] a small town in in Florida filled with
[00:29:05] some very very special people um all
[00:29:08] trying to um and delivering succeeding
[00:29:10] and delivering excellence. So there are
[00:29:13] dimensions of technology um around
[00:29:16] operations, facility, logistics, campus
[00:29:19] safety. You know, we have a fleet of
[00:29:22] golf carts running around campus. We
[00:29:24] have a hotel, we have, you know, a
[00:29:26] number of different dormitories and very
[00:29:29] unique sports specific facilities and 40
[00:29:31] fields and, you know, a number of
[00:29:33] different different things in place. I
[00:29:35] think the most interesting part of
[00:29:37] working in a facility like that which
[00:29:39] combines, you know, the most unique
[00:29:41] aspects of hospitality
[00:29:44] um and education and athletic training
[00:29:47] facilities and event space, right?
[00:29:50] There’s there’s so many things that
[00:29:51] happen on campus. So, some of it is, you
[00:29:53] know, maintaining a degree of visibility
[00:29:55] across an extraordinarily uh wide set of
[00:29:59] responsibilities. And so being able to
[00:30:01] keep tabs on what’s happening on campus
[00:30:04] in and of itself so that we can be most
[00:30:06] effective. Um that was really the first
[00:30:09] year’s goal when my team took on full
[00:30:11] responsibility for campus and I’ve dived
[00:30:13] in in greater detail. I think the most
[00:30:16] important thing at the moment is because
[00:30:18] of the existence of the frontier models
[00:30:20] and what large language models can do.
[00:30:22] Um we have seen a tremendous growth in
[00:30:25] the ability of the very credentialed and
[00:30:29] experienced and capable experts on
[00:30:31] campus whether that’s in leadership or
[00:30:34] in strength and conditioning or in
[00:30:36] injury recovery or teaching math or you
[00:30:39] know leadership or whatever the case may
[00:30:41] be to use the large language models to
[00:30:45] build their own software. So it’s timely
[00:30:47] to have this conversation now. Yesterday
[00:30:49] we launched what we’re calling the
[00:30:50] campus apps garden. Campus Apps Garden
[00:30:53] is a place where nontechnical users or
[00:30:56] non-technology staff who are developing
[00:30:59] um solutions in the context of what’s
[00:31:01] usually a browserbased session with
[00:31:03] claude or or chatpt can develop their
[00:31:07] ideas locally for as long and whatever
[00:31:10] depth is necessary and then hand that
[00:31:12] over to our team add uh some things that
[00:31:16] may not be obvious for people who are
[00:31:17] not traditional software developers.
[00:31:19] Right? So cloud and chat GPT are great
[00:31:21] at building you know kind of single page
[00:31:24] applications that are extremely comp you
[00:31:26] know complicated and specific very
[00:31:29] individualized use cases but um once
[00:31:31] they get out of the browser tend to tend
[00:31:33] to break down right they don’t have
[00:31:34] persistent data storage or they don’t
[00:31:36] have you know different kinds of access
[00:31:38] control or u don’t consider cyber
[00:31:42] security issues or you know don’t have
[00:31:45] workflow engines that would scale beyond
[00:31:48] one person operating operating it in in
[00:31:50] a browser. So what we are building is a
[00:31:52] set of services and an infrastructure
[00:31:54] environment that allows any user on
[00:31:56] campus with a compelling idea to take
[00:31:59] what they’re doing in conversations with
[00:32:01] an agent and then bring that to
[00:32:04] technology staff to kind of kick the
[00:32:06] tires a little bit, try and understand
[00:32:08] what they’re doing and then equip that
[00:32:09] with a set of services that we can then
[00:32:11] hand over as prompts to the agent that
[00:32:13] they’re working with to say, “Hey, layer
[00:32:15] on these APIs that you can use for
[00:32:17] persistent data storage. Here’s an API
[00:32:19] that you can use for managing role-based
[00:32:21] authentication and access control.
[00:32:23] Here’s an API that will give you access
[00:32:25] to these source systems on campus that
[00:32:26] we have to be pretty careful about, you
[00:32:28] know, who gets access to what, but more
[00:32:30] importantly, who has canonical or access
[00:32:32] to canonical data. There are many many
[00:32:34] different kinds of of flavors of of
[00:32:36] software on campus. And so getting
[00:32:38] access to all that information for a
[00:32:40] single agent who doesn’t have that
[00:32:42] context because they’re just operating
[00:32:43] in a browser session pretty tough. We’re
[00:32:45] making it easy for us to inform agents
[00:32:47] and nontechnical users on how to layer
[00:32:49] those things on and build really
[00:32:51] immensely powerful applications and put
[00:32:53] them in a safe scalable production
[00:32:55] environment that allow anyone on on
[00:32:58] campus or in our in our digital
[00:33:00] businesses to make use of them. Um, and
[00:33:02] that has been an amazing amazing thing.
[00:33:04] It’s a very different style of of
[00:33:06] development and a style of developing it
[00:33:08] than I think has existed previously
[00:33:10] because it democratizes the process of
[00:33:12] leveraging technology. It removes some
[00:33:15] of the constraints that exist around
[00:33:16] resources or staffing. I don’t I don’t
[00:33:18] have Infinity, you know, we’re not a
[00:33:20] we’re not a Silicon Valley company. I
[00:33:21] don’t have Infinity developers uh to
[00:33:24] work with and I don’t, you know, have a
[00:33:26] a huge number of senior software
[00:33:28] engineers. But what I do have are a
[00:33:31] small team of really skilled people who
[00:33:32] are great at building almost like the
[00:33:34] giant robot suit that basketball coach
[00:33:37] can put on and go hey I know a ton about
[00:33:39] how to drill developing u basketball
[00:33:42] players and I’ve worked with chatpt to
[00:33:44] think about how I want to do that but I
[00:33:46] don’t know how to share that with my
[00:33:47] fellow colleagues or maybe even take
[00:33:49] that capability and apply it to
[00:33:51] volleyball coaching and so we’re
[00:33:53] creating tools that allow that kind of
[00:33:54] local development to have all of the
[00:33:57] rigor and robustness and scalability and
[00:33:59] resilience and you know security safety
[00:34:01] that production consumer grade internet
[00:34:03] applications would have you know knock
[00:34:05] on wood that they’re all that secure put
[00:34:06] those well behind the corporate firewall
[00:34:08] and make them available to anyone who
[00:34:10] would want to use them and that’s that’s
[00:34:11] going to be a lot of development over
[00:34:13] the next couple years. know you know
[00:34:14] these were these were some really
[00:34:16] interesting insights that how you know
[00:34:18] um orgs are thinking about going into
[00:34:21] this particular direction and we’re also
[00:34:23] seeing um you know AI shift from
[00:34:25] individual tools to full platforms as
[00:34:28] you also talked about AI as service in
[00:34:31] sports that means um things like dynamic
[00:34:34] ticket pricing um AI agents that
[00:34:36] summarize sponsorship calls and generate
[00:34:39] ROI reports automated highlight reels so
[00:34:42] is that where you know the academy is
[00:34:45] headed where there they would be
[00:34:47] treating AI like infrastructure rather
[00:34:49] than a feature. What does that
[00:34:51] transition look like?
[00:34:52] >> I think it’s different in in in each of
[00:34:54] our businesses. So in in our digital
[00:34:56] products businesses, there is certainly
[00:34:59] active and ongoing utilization of of AI
[00:35:02] capabilities as features simply because
[00:35:04] the recruiting process, one-on-one
[00:35:07] coaching, um, our IMG Academy essentials
[00:35:11] product, those all really make good use
[00:35:14] of AI in the actual delivery of content
[00:35:16] and and coaching and recommendations.
[00:35:19] And so that’s always going to be part of
[00:35:21] of the exercise. And again, you know,
[00:35:23] recruiting is very much a two-sided
[00:35:24] marketplace. So, we’re as much trying to
[00:35:26] help the college and university and that
[00:35:28] coach figure out, you know, how to most
[00:35:31] efficiently and effectively find the
[00:35:33] talent they need for their team as much
[00:35:35] as we’re trying to create a great
[00:35:36] outcome. So, that student athlete, you
[00:35:38] know, not only is playing their sport in
[00:35:41] college, but succeeding in playing that
[00:35:43] sport. And on senior day, they’re
[00:35:44] showing up having not just, you know,
[00:35:47] produced great outcomes for their team
[00:35:48] and for themselves, but really a fully
[00:35:51] formed human that that is ready to take
[00:35:53] on whatever the next challenge will be.
[00:35:54] So that there is an AI part to that. As
[00:35:56] it relates to our campus business, I
[00:35:58] think it is much more in that
[00:35:59] infrastructure, right? So when you are
[00:36:01] delivering middle school, high school
[00:36:03] education in a boarding school
[00:36:04] environment, when you are delivering,
[00:36:06] you know, campus safety for a wide
[00:36:08] ranging event space of of really unique
[00:36:11] capabilities and proportions, when you
[00:36:13] are delivering, you know, extraordinary
[00:36:15] levels of of sophistication and quality
[00:36:19] in athletic training across, you know,
[00:36:22] 14, 15, 16. and we keep adding sports
[00:36:25] different sports with their own unique
[00:36:27] sets of capabilities as well as our
[00:36:28] proprietary uh athletic and personal
[00:36:30] development methodology those have to
[00:36:33] make use of infrastructure. It is not I
[00:36:35] think economically feasible to expect
[00:36:37] moving forward that it is only in
[00:36:39] technology organizations where the
[00:36:42] ability to use technology is is going to
[00:36:44] scale. We have to create just in the
[00:36:46] same way we’re creating aic harnesses to
[00:36:48] do our software development, we’re
[00:36:50] creating AI harnesses as technologists
[00:36:52] to let people who have really specific
[00:36:55] and detailed expertise, performance
[00:36:57] scientists and recovery specialists and
[00:36:58] physical therapists and, you know,
[00:37:00] people who help high school students
[00:37:02] navigate the challenges emotional and
[00:37:04] physical of being, you know, an elite
[00:37:08] athlete and just being a high school
[00:37:09] student, just being a teenager. they,
[00:37:11] you know, benefit from having AI tooling
[00:37:14] um and capabilities. And so that that is
[00:37:17] a pretty sacred mission for us to give
[00:37:19] them what they need and allow them to
[00:37:20] self-service as much as possible, not
[00:37:22] just to reduce the the cost of
[00:37:24] technology, but to actually create
[00:37:27] better outcomes for student athletes to
[00:37:29] in their future. There’s also a lot of
[00:37:30] hype around AI for injury prevention and
[00:37:33] performance optimization, variables
[00:37:35] tracking, biometrics, computer vision
[00:37:37] analyzing movements, but you’re actually
[00:37:40] implementing this at IMG as well. So
[00:37:42] with your experience, what you’ve seen,
[00:37:45] how much of it is real and what is still
[00:37:47] experimental?
[00:37:48] >> Yes. So you mentioned uh earlier in my
[00:37:51] career, I had the privilege of uh
[00:37:53] working with with Troy Hanikoff. was was
[00:37:56] the the founding of Techstar Chicago and
[00:37:58] so spent a fair amount of time but as a
[00:38:00] earlier as a startup founder myself and
[00:38:02] and then later as a mentor to to
[00:38:04] startups and still doing that today. The
[00:38:06] academy and particularly campus is a
[00:38:07] place where a great number of different
[00:38:11] uh startups in that space show up with a
[00:38:14] range of things from promises to
[00:38:17] extraordinary um products. So we do see
[00:38:21] and a lot that works. I think some of
[00:38:24] the more interesting things that we see
[00:38:25] right now are um the intersections
[00:38:27] between the wearables technology and
[00:38:30] sort of the machine vision. So
[00:38:32] intersections between you know cameras
[00:38:34] on the field or cameras in the weight
[00:38:36] room devices that that the student
[00:38:38] athletes are wearing be that for
[00:38:40] biomechanical measurements but
[00:38:41] increasingly kind of neurological uh
[00:38:44] assessments. We’re starting to get to a
[00:38:46] place where there is a a holistic
[00:38:48] monitoring capability very much tied to
[00:38:51] or kind of it’s almost feels like it’s
[00:38:53] been waiting to catch up to the sport
[00:38:55] science. So the sports science science
[00:38:57] and you know I mentioned earlier
[00:38:58] periodization I think periodization is a
[00:39:00] a big part of what we do at the academy
[00:39:02] which is trying to understand how you
[00:39:04] know to create sort of these macro messo
[00:39:06] and micro cycles of development that
[00:39:08] match both um the events taking place in
[00:39:11] a regular school year right for our
[00:39:13] student athletes but also at the team
[00:39:15] level what are the competition events
[00:39:17] that are happening and match that to
[00:39:20] okay you know what’s happening in the
[00:39:21] weight room on any given day or what are
[00:39:23] the series of things that we’re working
[00:39:24] on in a given for this particular
[00:39:26] student athlete and this particular
[00:39:27] team. That is a complex execution and
[00:39:30] the ability to instrument that entire
[00:39:33] cycle of development which we’ve been
[00:39:34] doing for a long long time um without AI
[00:39:37] tools. I’m starting to see again that
[00:39:39] that sort of bicycle cranking happening
[00:39:42] where the amount of data that we have
[00:39:44] and our ability to process it is is
[00:39:46] increasing. So for us the startups are
[00:39:48] bringing really unique hardware software
[00:39:50] solutions to that measurement cycle and
[00:39:52] performance assessment. One of the
[00:39:54] startups that we’ve been working with
[00:39:56] most recently is a company called Athos
[00:39:58] who has done a great job in taking what
[00:40:01] is a very complex and sport specific set
[00:40:03] of developmental capabilities and
[00:40:06] guidelines for how we’re you know
[00:40:08] training softball players and golf
[00:40:10] players and giving coaches really
[00:40:14] thoughtfully curated data sets and using
[00:40:16] AI within the product to generate
[00:40:19] insights about kind of where that that
[00:40:20] student athlete may be in the context of
[00:40:24] developmental cycles or injury
[00:40:26] prevention or you know in in very
[00:40:28] exciting moments sort of where they’re
[00:40:30] coming to local maxima of their
[00:40:32] performance capability and re-shifting
[00:40:34] the training plan to say you know what
[00:40:36] you are you’re at a place where you are
[00:40:39] performing at peak you have a game
[00:40:40] coming up let’s take it down a notch to
[00:40:42] prepare or you know what we’ve been your
[00:40:45] your weight training has been here we’re
[00:40:47] seeing that your performance over the
[00:40:48] last couple weeks suggests that we we
[00:40:50] probably could go to the next level that
[00:40:53] is happening every day now, ongoing
[00:40:56] conversation. And part of what what what
[00:40:59] is really unique to the academy is the
[00:41:00] ability to to have these world-class
[00:41:02] sport and performance scientists and the
[00:41:04] coaches who have all the context of how
[00:41:06] to develop um student athletes in those
[00:41:08] given positions and and teams in sports.
[00:41:11] that cycle is really really datadriven
[00:41:13] in ways that are are not sort of you
[00:41:16] know here here’s some useful suggestions
[00:41:18] which is great and I think a a great
[00:41:20] deal of of past history but now our you
[00:41:23] know fast cycles of hypothesis
[00:41:26] experiment recommendation result and
[00:41:28] adaptation in a day I mean that’s you
[00:41:31] know we we’ve been operating at the day
[00:41:32] cycle for a long time on the at the
[00:41:34] academy but now the level of insight
[00:41:36] that we’re able to deploy whether that’s
[00:41:38] from sort of you know the tools we have
[00:41:41] for measurement. The player datas and
[00:41:42] purchase of the world or tools like I
[00:41:45] mentioned with Athos. Um those are those
[00:41:47] are creating a very very rich
[00:41:49] environment for analysis assessment and
[00:41:53] making it easier for coaches be they
[00:41:55] sport coaches or or development
[00:41:58] performance science coaches to make very
[00:42:01] specific recommendations interventions
[00:42:03] in a way that not only helps them
[00:42:05] understand um kind of that cycle of
[00:42:07] development but also helps a student
[00:42:08] athlete understand what’s happening to
[00:42:10] them and to their body and to their
[00:42:12] capability which is very exciting. You
[00:42:14] know, I think it’s easy to look at this
[00:42:16] as technology that’s happening to
[00:42:18] people, but what I think is very
[00:42:20] interesting because so many of the
[00:42:22] student athletes on campus are, you
[00:42:23] know, we used to use the term digital
[00:42:25] native. I think we’re beyond that. They
[00:42:27] are assumptive that there is a lot of
[00:42:28] data being assembled about them that
[00:42:30] they’re creating and they are conversant
[00:42:32] with it. They understand, you know,
[00:42:34] their measurements. They understand
[00:42:35] what’s what happened yesterday and are
[00:42:37] looking at it and participating in that
[00:42:38] conversation with their coaches and with
[00:42:40] their teachers as well. So it’s it’s
[00:42:42] it’s an exciting time to see how this is
[00:42:44] applied on campus.
[00:42:45] >> We we know that you know this is
[00:42:47] something which is inevitable and we
[00:42:48] cannot now ignore it anymore to make it
[00:42:50] part of the processes. But how do you
[00:42:53] think um about fairness when we talk
[00:42:56] about AI you know AI and its its
[00:42:59] fairness and it notating or giving
[00:43:01] decisions that might lead towards you
[00:43:04] know existing um inequities that we
[00:43:07] already have. So what safeguards are
[00:43:09] there in place to make sure that the
[00:43:11] technology is not reinforcing those
[00:43:13] existing iniquities?
[00:43:14] >> As we began to develop policy around AI,
[00:43:17] you know, obviously the first generation
[00:43:18] of that was a little bit more about
[00:43:20] security and auditability and kind of
[00:43:23] data providence. I think the
[00:43:25] conversation shifted fairly quickly to
[00:43:27] to these questions. The academy has
[00:43:29] always been a place where a focus on
[00:43:32] fairness is sort of bred into our DNA.
[00:43:35] uh you know I think that there is the
[00:43:37] perception that every single student
[00:43:39] athlete at the academy and every single
[00:43:41] you know recruit in our online
[00:43:42] businesses is is a top tier elite
[00:43:44] athlete but that is is not the case.
[00:43:46] There are people at all different levels
[00:43:48] and again for us sport is a lens through
[00:43:51] which we’re we’re we’re helping to make
[00:43:53] better people as much as we are making
[00:43:55] better better athletes. And so for that
[00:43:57] reason, I think the biggest tool that we
[00:43:59] wield and has been a big part of the
[00:44:02] development of policy and training for
[00:44:04] our staff is about this issue of
[00:44:06] fairness. And the the tool that we use
[00:44:07] for that is adversarial review. Right?
[00:44:10] So adversary review, which really is a
[00:44:12] kind of a term of our coming out of more
[00:44:15] technology fields, is a way that when we
[00:44:18] are getting specific recommendations
[00:44:20] that are coming out of a large language
[00:44:21] model, we’re subjecting that
[00:44:23] recommendation to especially in
[00:44:26] strategic decision-m to assessments from
[00:44:29] not only other humans um but from other
[00:44:32] agents to really pick apart and ask
[00:44:34] questions about fairness and about bias.
[00:44:37] Again, you know, the models are
[00:44:40] extraordinary in their capability. Um,
[00:44:42] but they’re also extraordinary in their
[00:44:44] ability for those who are not not paying
[00:44:47] attention to kind of uh assume perfect
[00:44:51] context, which just doesn’t exist. So by
[00:44:54] forcing adversarial review, you know, by
[00:44:57] sticking to kind of the values that are
[00:45:00] kind of, you know, ever present on
[00:45:02] campus, which include ones about
[00:45:04] fairness, we’re ensuring that reflex
[00:45:07] exists both at the level of how are we
[00:45:10] developing this software using AI
[00:45:12] features or how are we equipping our
[00:45:14] coaches um and our teachers and our dorm
[00:45:16] mentors and our other staff with these
[00:45:19] tools in a way that we’re not
[00:45:20] reinforcing those biases. I think
[00:45:22] there’s a widespread uh awareness and I
[00:45:24] think you know my team has really tried
[00:45:27] to be not so much a backs stop but just
[00:45:29] kind of a a guide in saying hey these
[00:45:31] models were built on training data that
[00:45:33] has essentially you know started with
[00:45:36] the public internet of the last 5 6
[00:45:38] years that has lots of built-in biases
[00:45:41] in it um certainly the models themselves
[00:45:43] and the people that develop these models
[00:45:45] um carry with them a variety of
[00:45:47] different biases and I think it’s fair
[00:45:49] to say I’m part of the tribe of that
[00:45:50] people that set of people who bring with
[00:45:53] them kind of engineering thought
[00:45:54] processes and software thought processes
[00:45:56] and kind of nerd thought processes that
[00:45:59] make their way into the model. Um and
[00:46:02] and so you know I don’t think that
[00:46:03] there’s a magical panacea for this in
[00:46:06] the same way though I wish there was a
[00:46:07] magical panacea for fairness across the
[00:46:09] society as a whole. what we have to do
[00:46:11] is to to put these guard rails in place
[00:46:13] and constant reminders and training to
[00:46:15] say especially when we’re dealing with
[00:46:17] issues of of fairness and also accuracy
[00:46:20] and also pragmatism right in in much the
[00:46:24] same way you know I think fairness is
[00:46:25] certainly of ethical importance than
[00:46:27] than any of these other criteria but the
[00:46:30] models themselves are reasonably
[00:46:32] accomplished when contextualized
[00:46:34] properly to say you know I have this
[00:46:36] built into all of my agentic harnesses
[00:46:38] right like please apply ly, you know, a
[00:46:41] significant degree of doubt of your on
[00:46:43] recency bias, right? Or on a number of
[00:46:45] other biases. And, you know, we try and
[00:46:47] build those into the technology that
[00:46:48] we’re using. Um, and we’ve built those
[00:46:50] certainly into our policy and training.
[00:46:51] >> Joel, give us one real moment, something
[00:46:53] that, you know, went hilariously wrong
[00:46:55] or surprisingly right with AI in the
[00:46:58] sports context.
[00:46:59] >> Let’s see. I think the I think the uh
[00:47:02] the best example I could give you would
[00:47:04] be around uh periodization. So, one one
[00:47:08] of the joys of of my job is I’m
[00:47:10] constantly learning, right? So, I I
[00:47:12] don’t have a background in training in
[00:47:13] in performance science and sport
[00:47:14] science, though I’m proud to work with
[00:47:17] leaders in that space. Dr. Taran Morgan,
[00:47:19] who leads our athletic and and personal
[00:47:21] development team, is probably chief
[00:47:22] among them. And, you know, I’ve learned
[00:47:25] so much from her about how
[00:47:27] periodization, as I mentioned, work,
[00:47:28] right, works and and kind of the thought
[00:47:30] process behind that. And so as a
[00:47:32] software engineer and a technologist,
[00:47:35] when it came to trying to figure out how
[00:47:37] we would build automation and large
[00:47:39] language model capabilities into our
[00:47:41] approach to periodization and its
[00:47:43] implementation on campus, I looked at
[00:47:45] that problem as a fairly complicated and
[00:47:48] unfamiliar one. But in the last probably
[00:47:52] 6 months, a lot of my effort has been in
[00:47:54] building tooling that allows for the
[00:47:56] decomposition of complicated problems in
[00:47:59] performance science into more manageable
[00:48:02] chunks that teams of agents can kind of
[00:48:04] address. So lit literally yesterday a
[00:48:06] couple of my colleagues presented me
[00:48:08] with a fairly tight time constraint need
[00:48:11] for a specific application around
[00:48:13] periodization which needed to be very
[00:48:16] very thoughtful about sports specific
[00:48:18] differences in you know how we want to
[00:48:22] create training cycles and how those
[00:48:23] relate to a variety of different
[00:48:25] constraints around how campus works
[00:48:27] right there’s a lot of capacity to to
[00:48:30] train in the weight room a lot of
[00:48:32] capacity on fields but we have 1600 kids
[00:48:35] here plus you know thousands tens of
[00:48:37] thousands of campers the logistics of
[00:48:39] all of that plus the desired kind of
[00:48:42] performance science outcomes like that’s
[00:48:43] a fairly complex data set you know in a
[00:48:46] very short amount of time I was able to
[00:48:48] build a first iteration of of a solution
[00:48:50] to that problem that was uh commensurate
[00:48:53] with published research that exists
[00:48:57] around periodization in sport science
[00:48:59] and delivered the functional
[00:49:00] capabilities to automate some very very
[00:49:03] timeconuming in difficult problems. So I
[00:49:05] did not expect that first iteration to
[00:49:08] go well and I spent probably more time
[00:49:10] attempting to falsify the hypothesis and
[00:49:12] and sort of assess biases in the code
[00:49:15] but it it works uh and was able to kind
[00:49:18] of put that in our AI application
[00:49:19] infrastructure in in pretty time
[00:49:22] yesterday. So, it’s just a lot of fun
[00:49:24] right now, especially because in that
[00:49:26] context in developing AI, I don’t right
[00:49:29] now I have the privilege of being able
[00:49:31] to work closely with my collaborators in
[00:49:32] a way that building building software
[00:49:34] for the public internet just doesn’t
[00:49:36] afford. And that’s not that I I don’t
[00:49:38] enjoy that and that’s not a wonderful
[00:49:40] calling. However, being able to go, you
[00:49:43] know, literally toe-to-toe and and
[00:49:45] momentto moment with really skilled um
[00:49:48] experts in in sport uh and sport
[00:49:50] science, develop software for them to
[00:49:52] solve complex problems and have them
[00:49:54] give me immediate feedback and fix in in
[00:49:56] fast cycles because my ability to
[00:49:58] iterate and prototype and and then build
[00:50:00] for scale is so much faster. It’s joyous
[00:50:02] and I can’t I can’t describe it any
[00:50:04] other way. It’s helping really really
[00:50:05] skilled people do hard things. um which
[00:50:08] is why I got into doing technology in
[00:50:10] the first place.
[00:50:10] >> No, that’s great. And I think we’ve
[00:50:12] we’ve come a long way in the
[00:50:13] conversation as well and you know these
[00:50:15] are incredible initiatives that you’re
[00:50:17] taking at IMG um academy plus. Um here’s
[00:50:21] a oneliner that I’m going to throw and I
[00:50:23] want you to tell me your thoughts on
[00:50:24] that. AI is now most heavily for game
[00:50:27] analytics, fan engagement, and
[00:50:29] predicting player performance and
[00:50:31] injuries. Knowing what you know about
[00:50:34] where this is all headed, would the
[00:50:36] world of sports be better without AI? Or
[00:50:39] is the genie out of the bottle in a way
[00:50:41] that’s ultimately positive?
[00:50:43] >> Yeah, you know, I think I would reject
[00:50:45] the the binary structure of that
[00:50:48] question. It’s easy to sort of objectify
[00:50:52] the technology as good or bad. Uh I
[00:50:56] think you know we’re using these tools.
[00:50:58] Yes, there is a a material maybe one
[00:51:02] could describe it as exponential change
[00:51:06] uh in the capacity of this particular
[00:51:08] technology. So it’s not a step change,
[00:51:10] better software, uh more access to data.
[00:51:13] This is something different. I’m a big
[00:51:14] believer in the idea um that technology
[00:51:19] is in of itself a human construct. It’s
[00:51:22] not something that exists outside of the
[00:51:24] judgment that we as as human beings
[00:51:27] applied in its development and it and in
[00:51:30] its application. I think the particular
[00:51:32] risk of large language models and
[00:51:35] generative AI is that they carry with
[00:51:38] them the technology carries with it
[00:51:40] rather the a fairly easy mechanism to
[00:51:45] skip that judgment part because it is so
[00:51:47] fast and because it is so helpful. And
[00:51:50] certainly uh there is an extent to which
[00:51:52] anthropic and and open AI have built
[00:51:55] features in the way that the models work
[00:51:57] that want you to keep going down the
[00:51:59] path of reasoning without applying
[00:52:02] critical judgment and without applying
[00:52:04] things like adversarial review and
[00:52:06] checking for bias. But at the end of the
[00:52:08] day, keeping human judgment central,
[00:52:10] being transparent in the extent to which
[00:52:13] that judgment plays a central role in
[00:52:15] the application, I think puts us in a
[00:52:17] better place. If you are a writer trying
[00:52:19] to distill human experience and emotion
[00:52:22] into a piece of literature, the fact
[00:52:25] that you may have used a large language
[00:52:27] model to help facilitate that process to
[00:52:30] me is less relevant than the human
[00:52:32] experience of consideration of context
[00:52:35] of trying to to take what happened to
[00:52:38] you and your experience in living in the
[00:52:39] world and bottling that up in art that
[00:52:41] you share with others. I think the same
[00:52:43] is true of the software and tools that
[00:52:45] we build with AI. If our context is
[00:52:48] good, if our goals are worthy and if our
[00:52:51] thought process is critical and
[00:52:53] pragmatic and applies judgment, there’s
[00:52:56] a whole sub field of of uh academic
[00:52:59] subfield called computational
[00:53:00] hermeneutics which is fascinating topic
[00:53:02] for another conversation that really
[00:53:04] addresses these issues of how you bring
[00:53:06] human judgment and interpretation into
[00:53:08] the use of large language models and
[00:53:10] other computer models. If we do all
[00:53:11] those things, I think we end up with a
[00:53:13] mirror on ourselves that allows us to be
[00:53:15] more honest about our own biases and our
[00:53:17] own capabilities and we’re where we’re
[00:53:19] being honest with ourselves more than
[00:53:21] we’ve ever been able to before and how
[00:53:23] we accumulate knowledge. So I I think um
[00:53:26] this is a moment that we’re all going
[00:53:27] through. I think if we are able to
[00:53:29] appeal to our better instincts um in the
[00:53:31] use of this technology, we will achieve
[00:53:33] major things. But we have to be very
[00:53:35] very thoughtful about preventing the
[00:53:37] opposite from happening as well. Moving
[00:53:39] towards the last question of our
[00:53:40] conversation. Joel, you’ve led um
[00:53:43] technology teams for over 25 years from
[00:53:45] the early internet days at playboy.com
[00:53:47] and Britannica to digital marketing to
[00:53:50] now sports tech. For CTOs or IT
[00:53:53] directors thinking about moving into
[00:53:55] missiondriven spaces like sports,
[00:53:57] education or nonprofits, what’s one
[00:54:00] thing you wish someone had told you
[00:54:02] before you made that leap in your life?
[00:54:04] >> Oh, that’s a fantastic question. the
[00:54:06] first part of my career, I wish I had
[00:54:08] known how much deep satisfaction there
[00:54:11] is or maybe even more deep satisfaction
[00:54:13] there is in being able to provide an
[00:54:15] opportunity for other people to develop
[00:54:18] their talents. So, as a technologist and
[00:54:20] as a a young kind of uh engineer and
[00:54:23] system administrator, my focus was on
[00:54:26] accumulating knowledge and accumulating
[00:54:28] skill and being able to deliver a better
[00:54:31] hack and solve a problem more
[00:54:32] efficiently. And those are all wonderful
[00:54:34] things. But at a point in time today
[00:54:36] especially where large language models
[00:54:38] begin to make you know the possibility
[00:54:40] of personal software and you know sort
[00:54:43] of the mutability of programming
[00:54:44] languages and frameworks like that that
[00:54:46] sort of set of skills is uh a little bit
[00:54:49] more ephemeral. I think whether it is
[00:54:52] finding purpose in the mission of the
[00:54:53] organization or in the mission of
[00:54:55] developing or creating opportunities for
[00:54:57] success uh for the people around you
[00:55:00] where I find really the the deepest
[00:55:02] satisfaction. So, you know, yeah, I’
[00:55:05] I’ve been around since uh encyclopedias
[00:55:07] were an actual paid product, right? Um
[00:55:10] the things that I look back on, you
[00:55:12] know, with with great and deep
[00:55:14] appreciation and gratitude are, you
[00:55:16] know, the people who I started working
[00:55:18] with when I was a very young person who
[00:55:20] now have families of their own and have
[00:55:22] built businesses of their own and
[00:55:23] careers of their own largely because we
[00:55:26] were able to take the adverse and
[00:55:28] challenging conditions that we found
[00:55:30] ourselves in, whether those were
[00:55:31] management teams or business challenges
[00:55:33] or market challenges or technical
[00:55:35] challenges, and learn from them. and
[00:55:39] figure out how to create environments in
[00:55:41] which you know we looked at solving
[00:55:43] problems not people at which we created
[00:55:46] uh an environment in which you know
[00:55:48] trying to do things better um for one
[00:55:50] another as people trying to increase our
[00:55:53] knowledge trying to increase our
[00:55:54] adaptability trying to increase our
[00:55:56] optionality in the businesses that we
[00:55:57] were responsible for those were were
[00:55:59] skills that applied wherever anyone went
[00:56:02] and I think you know looking back you
[00:56:04] know I’m picturing the faces in my mind
[00:56:05] of people some of wh many of whom I’m
[00:56:07] still working with today at IMG Academy,
[00:56:09] but are, you know, in businesses across
[00:56:11] the planet who all share this idea that
[00:56:13] we’re here to help each other be more
[00:56:15] successful. And in helping each other be
[00:56:17] more successful, we’re building better
[00:56:18] software and technology. And in building
[00:56:19] better software and technology, we’re
[00:56:21] helping build uh better businesses and
[00:56:23] organizations and and overall kind of
[00:56:26] raising all the boats in the society.
[00:56:27] That’s that’s something I wish I had
[00:56:29] known would be so rewarding. As much as
[00:56:31] I’m proud of the first generation of
[00:56:33] technology that we built for for.com and
[00:56:35] successive ones, I think much more I
[00:56:38] feel much more gratitude for having been
[00:56:40] able to to help other people achieve
[00:56:43] their goals and and achieve what they
[00:56:45] can be, which is why uh my job at the
[00:56:47] academy resonates so deeply because
[00:56:49] that’s what I’m doing every day for for
[00:56:50] student athletes and their families.
[00:56:52] >> Well, Joel, thank you so much for this
[00:56:53] rounded and insightful conversation. We
[00:56:56] are really glad to have you with us
[00:56:57] today. Oh,
[00:56:58] >> it’s my pleasure, Robia. Thank you so
[00:56:59] much for for an excellent conversation
[00:57:01] and uh appreciate the time.
[00:57:02] >> Well, and to everyone listening, thanks
[00:57:04] for joining us on Tech Unhinged. Until
[00:57:06] next time.
