Latest / Agentic Conversations (formally mlops.community) / Collaboration and Strategy // Vin Vashishta // #176

Transcript

Collaboration and Strategy // Vin Vashishta // #176 · Agentic Conversations (formally mlops.community)

• 0:00
  So my name is Vin Vashishta. I am the founder and CEO of v ^2
• 0:07
  and I take my coffee Black double Nitro, little bit of
• 0:12
  sugar. Welcome, welcome, welcome
• 0:15
  everyone to another edition of the ML OPS Community Podcast.
• 0:19
  I am your host, Dimitrios, and today we're talking to none
• 0:22
  other than my man, Vin Vashista. He is an Internet influencer
• 0:28
  extraordinaire when it comes to data.
• 0:31
  He's been talking about the value of data and data strategy
• 0:36
  since before I even knew what machine learning was.
• 0:42
  Let's be honest, he's been in the game.
• 0:44
  For ages and he has hit a nerve with the greater Community
• 0:50
  because of the value that he brings and how he cuts through
• 0:56
  the hype. And he also reminds people to go
• 0:59
  back to square one, which is, and I think you will find this
• 1:03
  as a theme throughout this conversation.
• 1:06
  The TLDR, if you get one thing from this, it is that you cannot
• 1:11
  have a valuable machine learning.
• 1:15
  Or data product without first figuring out what the strategy
• 1:22
  is around that. And so we go into the depths of
• 1:26
  what he means by strategy. And I challenged him a little
• 1:28
  bit on strategy because I think it gets a bit of a bad rap.
• 1:32
  You know, anytime I hear people talking about strategy, I think
• 1:36
  about PowerPoints and basically it's code word for people saying
• 1:42
  that they do not want to actually do anything, they just
• 1:46
  want to talk about. Doing it and then let somebody
• 1:48
  else do it, so he pushes back as rightly so.
• 1:52
  And he tells me what good strategy and good execution is.
• 1:57
  And I hope you enjoy this edition of the Milos Community
• 2:01
  Podcast. For all those people that are
• 2:03
  listening on Spotify, We are dangerously close to getting
• 2:08
  over 100 reviews, which for me is a small little milestone that
• 2:12
  I would love to hit. So if you are listening, hit us
• 2:14
  with a review, give us some feedback and we will see you all
• 2:18
  on the other side. Take care.
• 2:20
  Let's get into it with Vin. All right, so for those who did
• 2:30
  not know, Vin, you were one of the first people to come on
• 2:32
  here, one of the first people that I would consider a big name
• 2:36
  to come on to this podcast and grace us with your presence.
• 2:38
  And since then, I've probably quote you at least once a month.
• 2:43
  And there are some things that you have said that have made
• 2:47
  their way onto the ML OPS community.
• 2:50
  Merch one being that Kubernetes is a gateway.
• 2:54
  Drug and I mentioned it last LM's in production conference.
• 2:59
  I think that we may have to update that to ChatGPT is now
• 3:02
  the gateway drug for everyone, but it's great to have you back
• 3:05
  here. I'm so excited for this
• 3:06
  conversation and I think we can start, dude, I just want to know
• 3:10
  what have you been thinking about recently?
• 3:12
  Like strapping me back? Yeah, I think Chad should be.
• 3:15
  That's really interesting, that generative AI.
• 3:17
  It's not really a gateway drug, It's more, I don't know, heroin.
• 3:19
  It's first hit, it's addictive, and then you realise it's kind
• 3:23
  of taking over your life. And some people right now are
• 3:27
  sort of going through that, that rehab stage where they're
• 3:31
  thinking, well, OK, well, maybe I'll, I took this a little too
• 3:33
  far. But that's, I mean one of the
• 3:36
  big things I've been thinking about is like big picture, you
• 3:40
  know, when I was heads down building, it was a completely
• 3:46
  different view and I got sort of 20/15/2016.
• 3:51
  I got a little taste of what it took to get a big initiative,
• 3:56
  one of those that was expensive but delivered ridiculous ROI.
• 4:01
  What did it take to get that through?
• 4:03
  And that's been percolating. I mean, the last time we talked
• 4:07
  was before the book, before fully pivoting into teaching and
• 4:13
  explaining strategy, product management.
• 4:16
  So that's my head's really been what are the implications here
• 4:21
  and what's coming next? Because that's where if you're
• 4:24
  talking about strategy, strategy isn't about where we are today,
• 4:27
  where we were yesterday. It's forward-looking.
• 4:30
  It's prescriptive. So I'm trying to think more
• 4:33
  about where do we go next and how do we position data teams,
• 4:38
  how do we position careers. You know what, what's a data
• 4:41
  engineer going to be in two to three years?
• 4:44
  What's a data scientist going to be in two to three years?
• 4:48
  But also what are customer preferences going to look like
• 4:51
  across domains in two to three years?
• 4:54
  What are businesses where we talk about generative AI, big
• 4:57
  productivity boom and improvement, What's that going
• 5:01
  to do to our businesses? What a, what does that mean?
• 5:04
  It's so it's been it's different, really different
• 5:08
  headspace. Yes, because I remember when we
• 5:11
  talked last time one thing that. Constantly stays in my mind is
• 5:15
  something that I think you just basically went all in on and
• 5:18
  it's. You have to understand the
• 5:23
  business value of any project that you undertake.
• 5:26
  You made it so that I understand that and so when you talk about.
• 5:32
  Where we're going and kind of peeking around the corner,
• 5:36
  especially when it comes to data engineers and data scientists.
• 5:39
  I would love to hear what your thoughts are and what hypothesis
• 5:42
  you have on how those roles will change.
• 5:45
  Your libraries, generative AI, all of this stuff.
• 5:48
  It makes things faster so I can do the same thing that I did two
• 5:55
  years ago. Today, with a generative AI
• 5:58
  assisted tool, I can do it faster.
• 6:01
  It's not easier. Well, that's really the
• 6:04
  distinction that we have to make.
• 6:05
  It didn't get any easier. The tough parts of data science
• 6:08
  are still tough. Data science still needs data
• 6:12
  generation, didn't change any of that.
• 6:14
  So we still need to build models.
• 6:17
  Generative models will not do everything.
• 6:20
  Yes, you can make generative AI do some of these things, but not
• 6:25
  the GPT that it doesn't work out-of-the-box.
• 6:29
  And so a lot of what data scientists are doing hasn't
• 6:32
  changed. We're still working in multiple
• 6:36
  directions. It feels like, you know,
• 6:37
  generative AI is everything. It's not.
• 6:40
  There are already different research avenues that are
• 6:44
  proposing replacements and improvements for Transformers
• 6:48
  because they're ridiculously expensive to build and they
• 6:53
  don't really work with the data efficiently or effectively.
• 6:57
  They don't get great signal. We just throw Hons of data at
• 7:03
  them. That's not sustainable.
• 7:04
  We need to start really fleshing out and exploring causal machine
• 7:08
  learning and creating that connection between causal
• 7:12
  inference and machine learning and our models and our ways of
• 7:17
  generative plays into that. We can do causal discovery with
• 7:20
  generative models, and we can begin to use them to build
• 7:23
  hypotheses about causal relationships from these massive
• 7:27
  data sets. O we've got a lot of work to do.
• 7:29
  There's robotics, there's autonomous, there's pretty soon
• 7:33
  IoT is going to give U.S. data the likes of which we've never
• 7:37
  had access to. We have to figure out, you know,
• 7:40
  again, optimization. That's a lot of data.
• 7:43
  We have to figure out how to train models on device or run
• 7:46
  smaller versions of models on these lower resource devices.
• 7:50
  We've got so much work to do. Data science hasn't changed.
• 7:56
  We're still doing things. So if you get rid of your data
• 7:59
  science team, I mean you're really hobbling yourself.
• 8:02
  But at the same time, I think generative AI has brought some
• 8:05
  realism. You don't need like these, you
• 8:08
  know, high end advanced researchers with eight pH D's
• 8:13
  and 15 conference presentations. And you don't, most businesses
• 8:17
  don't need that person. That's and that's the realism.
• 8:21
  I think generative AI is brought to the table is the paradigm
• 8:25
  will be these big foundational models and breakthroughs.
• 8:30
  Those will be provided by third parties by truly AI first
• 8:34
  companies. Cause back to that ROI question
• 8:38
  from the beginning and I think the narrative that you harp on
• 8:43
  is very much like Google can afford that because point O 1%
• 8:48
  increase is still millions of dollars for them.
• 8:51
  And so they're able to see the ROI very clearly, but for the
• 8:58
  rest of the companies out there. Maybe it's not as clear and
• 9:02
  they're not at a scale where that .0001 increase in accuracy
• 9:06
  or whatever the metric is, is going to make such a difference.
• 9:09
  So I 100% see where you're coming from on that.
• 9:12
  And I think you also shared a great meme.
• 9:16
  And by the way, this is like a side note for anybody that does
• 9:19
  not follow you on LinkedIn or Twitter, go follow Vin because
• 9:22
  it's just quality all day long. And thank you, the meme was.
• 9:27
  Where it's the guy that's jumping over all the steps,
• 9:31
  right? He's got 1 foot at the bottom
• 9:33
  and then one foot at the top of the stairs.
• 9:35
  And each stair is like data collection, data cleaning, data
• 9:41
  access or like data engineering type things.
• 9:43
  And then at the top of the stairs where the guys trying to
• 9:46
  get to it just says generative AI.
• 9:49
  And so you're skipping over all of that and boom, you go to the
• 9:53
  generative AI use case and it comes back to this idea of it.
• 9:58
  Didn't change much, it just made things faster.
• 10:01
  O. What exactly does that mean?
• 10:04
  That it's faster now because we can just hit an API.
• 10:09
  It's a different paradigm. And I this is something that I
• 10:11
  try to teach in all of my classes.
• 10:13
  I spent the whole chapter on my book's talking about this
• 10:17
  because it's important to understand it really in a lot of
• 10:21
  the ways that you're you touched on 1st, when you talk about that
• 10:25
  one percent, 110th of a percent, and being able to monetize that,
• 10:30
  your initiatives returns must scale faster than costs.
• 10:36
  And as you begin to create more advanced models, your costs
• 10:41
  begin to scale faster than returns for almost every use
• 10:46
  case. So we have to be very
• 10:48
  intentional and think about is this a place where returns,
• 10:54
  Returns for the business will scale faster than costs.
• 10:59
  If you look at the reason why a company like Google or Amazon or
• 11:03
  Meta or really Walmart or a company like Airbus or Merricks
• 11:09
  who does the shipping companies like FedEx, UPS.
• 11:14
  The reason why they can operationalize and make tonnes
• 11:18
  of money on very high reliability models is because
• 11:23
  they have scale. They have problems at scale.
• 11:28
  So for them the the expense of building out these more complex,
• 11:32
  more reliable models is it makes the unit economics works, makes
• 11:39
  sense. We have to think about our roles
• 11:42
  that way. But we can't ask data
• 11:44
  professionals to become strategists and product.
• 11:49
  They've got other work, so we need to talk about bringing on
• 11:54
  new roles. We need value centric data
• 11:57
  professionals. They need to understand the
• 11:59
  business. We need to have a level of
• 12:00
  business acumen. Leaders need an even higher
• 12:04
  level of business acumen and awareness of.
• 12:08
  But really how to get other parts of the business involved
• 12:12
  and engaged, how to lead strategy and lead strategically,
• 12:17
  These are skills that we need to develop in our data
• 12:21
  professionals. So we don't need management, We
• 12:23
  need leadership. We said we don't need strategy.
• 12:26
  No, we don't need another PowerPoint presentation.
• 12:28
  That's what we don't need. We do need strategy.
• 12:31
  We need business acumen. So there is this, you know,
• 12:35
  there's this separation that we need to make where some roles
• 12:39
  are technical facing but still have that business acumen so
• 12:43
  that they can do the job and bring these concepts of the
• 12:48
  businesses value and business returns and customer needs to
• 12:53
  all of the decisions that we make every day that influence
• 12:56
  the downstream product that we deliver.
• 12:58
  But we have to support all these other roles because you know,
• 13:01
  it's the reason why C level leaders want to step over all of
• 13:05
  that stuff is because they don't understand any of it.
• 13:08
  They have no idea what any of that means or how data connects
• 13:14
  to the business and business value.
• 13:17
  And as we now have more and more CDO's coming into the business,
• 13:23
  we need to support them so that they understand how to get a
• 13:26
  seat at the sea level strategy planning table because that's
• 13:31
  where these perceptions get influenced.
• 13:33
  O there's kind of two sides to this.
• 13:35
  I want data professionals to have Better Business acumen,
• 13:38
  Better Business skills because that results in better products.
• 13:42
  I want these new roles, data, product managers, AI product
• 13:46
  managers, data and AI strategists to come into the
• 13:49
  fold because that frees up the data team and it also helps keep
• 13:53
  the connection between strategy and execution, keep it really
• 13:57
  close together. The reason why I wrote from data
• 14:01
  to profit is because you need a single playbook.
• 14:03
  And if the business isn't operating from a single
• 14:06
  playbook, or if they're asking the data team to sort of
• 14:09
  overextend itself and do things that it's just not not
• 14:13
  structured, not staffed, not empowered to do, there's going
• 14:18
  to be, you know, kind of a no win situation.
• 14:21
  We'll keep having C level leaders want to step straight
• 14:24
  over 5 steps to get to generative AI because that's
• 14:27
  what they're being told they need.
• 14:29
  Whenever I hear the word strategy, it always makes.
• 14:34
  A little red light go off in my head or the red flag go around
• 14:37
  and it's basically like I equate it to someone that doesn't want
• 14:41
  to actually do this thing, they just wanna tell the person to do
• 14:44
  the thing. So what does this actually?
• 14:46
  What does this look like in practise?
• 14:49
  Well, that's one of the challenges we have is the
• 14:51
  perception of strategy sometimes meets reality.
• 14:54
  There are a tonne of strategists who are hand waivers and
• 14:58
  PowerPoint jockeys. Those are really, they've given
• 15:03
  strategy a bad name and it's it's well earned as a bad name
• 15:08
  where they show up, ask a tonne of questions and then drop off a
• 15:11
  book and they say, yeah, go do this.
• 15:14
  And you look at the book and you're like, man, I come on,
• 15:17
  come on man, we knew that. Come on, what are you doing?
• 15:21
  Or look at my book. What are the examples that I
• 15:24
  bring up is CNN Plus where they brought in the company, brought
• 15:28
  in a strategy Big 5 strategy consulting company to rubber
• 15:32
  stamp a really bad decision and that's that's why strategies
• 15:36
  have a bad reputation because. They Then nobody gets fired.
• 15:39
  Yeah, yeah, they earned it. This is kind of the reality that
• 15:43
  if you are looking at it, the technical strategy standpoint,
• 15:47
  we're cleaning up a lot of this and I call that strategic debt
• 15:50
  and businesses have a tonne of strategic debt because we've
• 15:54
  been making these decisions about strategy that are
• 15:57
  forward-looking, that are prescriptive.
• 15:59
  We've been making them through best guests and whoever yells
• 16:03
  the loudest, whoever says, yeah, I've got 8 degrees from these
• 16:08
  business schools. I've been doing this for 38
• 16:10
  years, need to listen to me. You know, that's what strength.
• 16:14
  And if you're a data professional, you're looking at
• 16:16
  that like, right, That's why it comes off that way.
• 16:21
  It is fluffy. But it's there's another reason
• 16:24
  for it too. Scientists and engineers
• 16:25
  struggle with strategy because it takes these away from us.
• 16:29
  Strategy has nothing to do with your hands.
• 16:32
  And we are so hands focused. We are so tactical.
• 16:35
  We are so execution do do, do that.
• 16:40
  Even when we plan, we are planning to do something
• 16:42
  immediately. And so strategy is a challenge
• 16:45
  for us. We want to quickly move away
• 16:48
  from strategy because it's uncomfortable and get two let's
• 16:52
  do something because that's where we live.
• 16:54
  That's why our comfort zone is. But strategy is different.
• 16:58
  Strategy needs frameworks. Because when it comes to
• 17:02
  strategy, we're not explaining what to do.
• 17:05
  I'm not telling you do this, do that.
• 17:08
  That doesn't work. Why?
• 17:09
  Because conditions are different at every business.
• 17:12
  Conditions are different from year to year.
• 17:15
  Conditions aren't static. So the advice that works for one
• 17:20
  company won't work at another one.
• 17:22
  We have to provide something that's more flexible.
• 17:25
  Strategy isn't be plant strategies.
• 17:28
  That framework that allows you to come up with multiple plans
• 17:33
  that lead you to success, to that goal that you've decided
• 17:38
  you want to pursue strategies, even a framework to decide what
• 17:42
  goal you should pursue. So it's a different mindset.
• 17:48
  The reason why we need strategy is because without it, there's
• 17:52
  no connection back to value. And even if your business is a
• 17:57
  not-for-profit, I mean if value means something totally
• 18:00
  intangible where it is societal good, you can see that no matter
• 18:04
  what value is defined as by the business, data science has a
• 18:09
  role. We can help the business
• 18:12
  understand its options. In some cases, we're helping the
• 18:17
  business understand itself for the first time, understand its
• 18:20
  customers in a new way for the first time, understand the
• 18:25
  marketplace. These are all complex business
• 18:28
  systems, so strategy can be supported by data science
• 18:33
  because we can create models that explain these complex
• 18:38
  business systems and make it easier for that strategy
• 18:42
  planning process to be forward-looking and prescriptive
• 18:46
  about what should happen. But also measure those feedback
• 18:49
  loops and when things are not working, when the plan doesn't
• 18:54
  seem to be going in the right direction because we're not
• 18:57
  seeing the outcomes we expected, we can help reveal options.
• 19:01
  Well, what types of changes will be most effective?
• 19:04
  O you can see data science is almost strategic by nature.
• 19:08
  We have a deep connection to real strategy, not fluffy
• 19:12
  strategy. And when you begin to introduce
• 19:15
  data analytics and advanced models into the strategy
• 19:18
  planning process, it stops being fuzzy.
• 19:21
  It's not this abstract, you know, little thing where you
• 19:26
  know, you drop it off the book and you're running, You're
• 19:28
  dropping off data models. Really concrete support for this
• 19:37
  particular opportunity and that's never been possible
• 19:39
  before my book. I talk about CNN Plus and that's
• 19:42
  what strategy used to look like. That's what it's always been.
• 19:46
  So strategy now, and I say strategists in many circles, are
• 19:52
  terrified of data scientists because as soon as you realise
• 19:55
  what you can do, you can replace many strategists, but you have
• 20:01
  to change the way that you operate and that's why so many
• 20:06
  data scientists have a hard time.
• 20:08
  Making the jump, even though what we do is so interlinked
• 20:14
  with strategy as a function of a business.
• 20:18
  O What are some things that you go from that's maybe just walk
• 20:22
  through As someone who is a data scientist or even a data
• 20:25
  engineer, what are the questions you are asking yourself as you?
• 20:30
  Level U. How could I become a multiplier?
• 20:33
  That's the biggest question that engineers, data scientists,
• 20:38
  analysts, data engineers doesn't matter.
• 20:40
  If you're technical, you need to be asking yourself, how can I
• 20:43
  become a multiplier? Because that's the mindset shift
• 20:47
  we think about. Here's how much value I can
• 20:51
  deliver, here's how fast I can code that.
• 20:53
  Here's how complex of a project I can take on.
• 20:57
  Here's how many different frameworks I can learn.
• 21:00
  But think about it, when you scale that and that's what it
• 21:04
  takes to get a promotion, get a raise, you have to deliver more
• 21:08
  value. So we're scaling, and we think
• 21:11
  that our technical skills are going to be the things that
• 21:13
  scale us. But there's a ceiling, because
• 21:17
  you could only work so many hours O eventually, no matter
• 21:21
  how smart, no matter how technically capable you become,
• 21:27
  right around the 10 or 15 year mark, you hit the plateau.
• 21:30
  And it's a hard ceiling. So year five, you need to be
• 21:35
  thinking about what should I learn to make myself a
• 21:38
  multiplier where I am not the only source of improvement.
• 21:44
  So when you get better, the team gets better.
• 21:48
  When you get better, the projects you deliver begin to
• 21:52
  deliver higher returns. That's where you begin to see
• 21:57
  why some of the people that are at the staff level get to be
• 22:00
  where they are. Because they are amazing
• 22:02
  teachers. They don't just take on complex
• 22:06
  work, they make the entire team more capable.
• 22:10
  They train and mentor and get people who are at the mid level
• 22:15
  to perform at just a, you know, a half a level shy of where they
• 22:19
  are. So they don't just make
• 22:21
  contributions individually, they make the entire team better.
• 22:26
  Great leaders create opportunities for their team.
• 22:30
  They create teams that are more effective, more productive,
• 22:35
  deliver more value as a unit than they would as just a group
• 22:40
  of individual contributors. Folks, real fast.
• 22:44
  I want to mention that we are having an LLM in production part
• 22:47
  three. That's right, Uno Dose Trace.
• 22:49
  This is the third time around and I'm pretty sure we're
• 22:52
  getting better each time. You know what?
• 22:54
  You can expect top level talks from people that are actually
• 22:59
  doing this. No hype, none of this, this
• 23:01
  Twitter demo bullshit. It's all real deal,
• 23:06
  productionized using LLM's. All right?
• 23:08
  So I'm going to tell you a few of the speaker names to try and
• 23:12
  get you excited and signing up. For the event.
• 23:16
  You can click on the link in the description or you can just go
• 23:19
  to LUMS introduction com or go to home Milos community and you
• 23:25
  will see the event there. Let's talk about some of these
• 23:29
  speakers because wow, chef's kiss.
• 23:31
  All right, we've got Ben Harvey from AI Squared.
• 23:34
  You know who he is. He's doing all kinds of wild
• 23:36
  stuff with government LM, security, privacy.
• 23:41
  Oh, it's so good. We've got Timothy from Mistral.
• 23:44
  Might as well be Timothy Chalamet, because Mistral.
• 23:47
  Is blowing up. After that gigantic round they
• 23:50
  raised, we've got Azul from Nichola also blowing up because
• 23:53
  of time. GPT, a time series open source
• 23:56
  foundational model. Mind blown.
• 23:58
  Right there we have everyone's favourite influencers, Shreya
• 24:02
  Shankar. She's going to be on a panel
• 24:04
  talking about how to debug LMS. What do you do?
• 24:06
  You just prompted a little more. Is it prompt tuning?
• 24:09
  Do I need to fix my RAG? Do I need to fix my vector
• 24:11
  database? Sparse vectors?
• 24:12
  Dense vectors? I don't know.
• 24:14
  My mind is exploding. It's a common theme in this
• 24:18
  conference. We also have Lily Chan from
• 24:20
  contender Bill Salek, CTO of Brainly.
• 24:24
  And if you're into agents, you know who we've got.
• 24:26
  We've got Silent, the R&D lead from Auto GPT, talking to us all
• 24:30
  about the best practises that he and the auto LGBT community has
• 24:35
  found since releasing that a few months ago.
• 24:38
  We also have Harrison Chase. He's going to be coming on
• 24:42
  giving a talk. How could we not?
• 24:44
  And this is not even mentioning the hard hitters that we have
• 24:48
  from the gigantic corporations like NVIDIA and Google and Meta.
• 24:52
  We've got all these people. I'll leave you with one of my
• 24:55
  favourite talks that I am most excited for that is from Dmitro,
• 24:58
  the creator of Onyx. He is currently leading and the
• 25:02
  CTO of Fireworks AI. For that he was lead maintainer
• 25:07
  of Π Torch. Guys got a solid track record.
• 25:10
  Shout out to our main sponsors process, they are the reason
• 25:14
  that this is happening. And that you all can benefit
• 25:18
  from it. And last but not least, before
• 25:20
  we go, let's just talk about what some of the different
• 25:25
  topics are going to be, what the talks are.
• 25:28
  We've got deploying open source models in production, what your
• 25:32
  different options are, regulatory aspects of MLM's, how
• 25:36
  to build economical LM solutions, how to select
• 25:39
  prioritise LM use cases. We've got tonnes of product
• 25:42
  owners and product engineers coming to give talks like
• 25:45
  Philip. From Honeycomb, he was on the
• 25:48
  podcast a few pods ago. Go and check that.
• 25:50
  Talk out if you haven't enjoyed this podcast.
• 25:53
  We'll see you all there. And you know what else is gonna
• 25:56
  come in the breaks of these events?
• 25:58
  It's gonna be some bad musical improvisations from yours truly.
• 26:04
  You see the power of leadership. Leadership removes roadblocks.
• 26:07
  Leadership creates opportunities for growth.
• 26:11
  And so instead of the leader just being sort of that centre
• 26:14
  point, the leader is making their team more valuable, more
• 26:20
  effective, more productive and focusing them on higher value
• 26:25
  activities. So they're multiplying those
• 26:28
  hours that they work. They work one hour and instead
• 26:31
  of getting that one hour's worth of value, they're getting a
• 26:34
  improvement out of every part of their team strategists.
• 26:39
  The same way, if you look at a product manager, the product
• 26:42
  manager is often the difference between a product that gets
• 26:45
  delivered, that doesn't get adopted by customers and that
• 26:48
  home run. There are multiplier.
• 26:51
  So they make sure that all that money that gets spent on the
• 26:55
  initiative isn't wasted. And you think about that.
• 27:00
  That's a huge multiplier and that's where strategy is so
• 27:04
  powerful, so valuable. That's why strategists and
• 27:07
  product managers get paid ridiculous amounts of money.
• 27:11
  Because if you think about all the investments that the
• 27:13
  business makes, strategists make sure that those investments
• 27:19
  return more with them in that strategist role than they would
• 27:26
  without them in that strategist role.
• 27:29
  And as you move U sort of the strategist ladder or leadership
• 27:33
  ladder from manager to CDO or from individual contributor to
• 27:37
  product manager to a a strategy technical strategist or data and
• 27:41
  AI strategist role, you begin to have a greater and greater
• 27:46
  impact. I like this.
• 27:49
  Idea of becoming a multiplier because there was so many ways
• 27:52
  that my mind started thinking about like how you can become a
• 27:56
  multiplier, right? And it felt like there's almost
• 27:58
  these sophomoric ways of looking at it.
• 28:01
  Like, I'll just use more copilot when my coding and that'll make
• 28:06
  me a multiplier because I'll be able to work more and get more
• 28:10
  done faster, whatever. But then there's also ways of
• 28:14
  looking at it that are much more mature and wise, I would say,
• 28:18
  and like you were saying. Bringing on the team, creating
• 28:22
  situations where the team can thrive and then not only that,
• 28:25
  but there's these other ways of becoming a multiplier that I
• 28:29
  wasn't even thinking about, which is?
• 28:32
  Becoming a teacher so that the people around you are raised up
• 28:36
  and whether that is you doing it in by creating incredible docs
• 28:41
  that people can read. Or it's just you pair
• 28:45
  programming with someone and making sure that they understand
• 28:48
  and so then the next time they do that, they're doing it 10
• 28:51
  times faster, whatever it may be.
• 28:54
  The code reviews are another excellent, like code reviews,
• 28:57
  model reviews. Unreal opportunities too.
• 28:59
  Same thing. Exactly.
• 29:02
  And so I love this. I also really want to dive into
• 29:07
  what you're talking about here. Where?
• 29:09
  It seems like there's this new job that.
• 29:15
  You're talking about which is like the AI product manager and
• 29:18
  what that looks like. So it's not necessarily new.
• 29:22
  It's becoming more ubiquitous. I think it was primarily big
• 29:27
  tech and companies like you know IBM, SAP had them as well, some
• 29:32
  of the more traditional technical companies.
• 29:35
  But it's now gaining a lot of traction at more more diverse
• 29:43
  businesses. MasterCard now has not only
• 29:46
  product managers but also strategists and they'll embed a
• 29:50
  data or an AI strategist in every organisation to help them
• 29:54
  figure out how they're going to monetize.
• 29:56
  And so there's these new roles that are beginning to permeate
• 30:00
  into businesses. The main I guess cause behind a
• 30:06
  data and AI product manager versus just take your digital
• 30:09
  product manager is digital is monetized differently than data
• 30:16
  and AI it's built and used by customers differently.
• 30:21
  So all of these different types of of really paradigms that are
• 30:27
  associated with data and AI, you have to understand that there's
• 30:31
  a difference if you try to do and this is what most companies
• 30:34
  have figured out, it's the reason why they're hiring data
• 30:36
  and AI product managers. If you try to fit data and AI
• 30:39
  products into a digital box, it doesn't work.
• 30:43
  We live in now this sort of multi technology marketplace
• 30:48
  where you have digital, you have cloud, you have IoT, 5G, we've
• 30:53
  got web, we've got mobile, we have data analytics, advanced
• 30:57
  machine learning platforms are here and emerging.
• 31:03
  And I mean think about it, five years from now we're going to
• 31:05
  have quantum computing that is going to be a wave that hits us
• 31:09
  in five to 10 years. So we have all of these
• 31:11
  different technology waves. They're all monetized
• 31:15
  differently and they all have to coexist.
• 31:18
  A product that starts its life as a digital product may become
• 31:24
  a data product, or may integrate data features into it to make it
• 31:30
  more valuable to find new ways to monetize customers or that
• 31:34
  data. It may then incorporate
• 31:37
  analytics. It may use more advanced, more
• 31:41
  reliable models to support functionality.
• 31:45
  So we think of products today as sort of the static.
• 31:49
  It's either this or that. And technology isn't like an
• 31:53
  iPhone. You don't just go from the
• 31:55
  iPhone 12 to 13 and I've got to go from digital to AI and toss
• 31:59
  digital out the door. No, it's not going to work that
• 32:02
  way. That's what we need.
• 32:03
  Data and AI product managers, because they understand this new
• 32:07
  paradigm, they realise that data and AI are monetized in
• 32:11
  different ways. If you think about it, data set.
• 32:14
  If I gather data, it's high quality, I can use that for
• 32:19
  several reports, I can use it to train multiple models.
• 32:23
  It's still good. Like I haven't deleted it, I
• 32:26
  haven't used it, and it could be repurposed for something
• 32:30
  completely different, but I don't have to change the
• 32:32
  underlying asset itself. It's still that original
• 32:37
  artefact, and there's been no change to it.
• 32:40
  If you look at something like a digital piece of software, the
• 32:44
  amount of work that you have to do to get a piece of software
• 32:47
  that was used for one use case to support another use case is
• 32:51
  typically pretty heavy. So it can't be monetized in the
• 32:56
  same way. Machine learning models very
• 32:58
  much the same. And Microsoft's teaching a
• 33:00
  master class in this. They're treating Copilot and you
• 33:04
  know GPT like Frank's red hot sauce.
• 33:07
  They're putting it on it everything.
• 33:09
  And it is, it's literally just being monetized.
• 33:13
  It's the same thing, and it's being monetized over and over
• 33:17
  and over again. And that's the only way that
• 33:19
  they could justify putting 10 billion, I mean, seriously, $10
• 33:24
  billion into Open AI. That's that takes some spine
• 33:29
  right there, but it also takes vision.
• 33:31
  You have to see that it's not just an interesting, you know
• 33:37
  chatbot, you have to look at it and you say, I see ways that we
• 33:41
  can monetize Bing like we've never done before.
• 33:45
  We might be able to revive Bing with this thing.
• 33:48
  With generative AI, we could integrate generative features
• 33:52
  into Office 365 and charge more every month for it.
• 33:57
  We could integrate it into VS Code and support new features
• 34:00
  and functionality that we might be able to charge 20, thirty,
• 34:03
  $40 a month for. That's the product paradigm.
• 34:07
  The reason why data and AI product managers are so critical
• 34:10
  because just that one paradigm is one of dozens that you need
• 34:15
  to bring to the role to successfully monetize data.
• 34:20
  Machine learning, even analytics, even reporting.
• 34:23
  You have to think about it completely differently.
• 34:27
  And if you don't, if you're trying to stuff it into a
• 34:28
  digital box, it doesn't end up delivering the kind of value
• 34:33
  that it could. That's why so many businesses
• 34:36
  fail when it comes to IS they try the digital way.
• 34:41
  But they need someone who thinks in the data and AI paradigms,
• 34:46
  and who can not only teach the data team how to deliver value,
• 34:50
  but also teach the business how to find it, how to integrate it,
• 34:55
  how it interacts with customers. It's another paradigm.
• 34:59
  Customers don't use this the same way.
• 35:01
  Users don't interact with it the same way as they do digital
• 35:04
  technology. So there's all of these levels
• 35:07
  of frameworks and paradigms that you need a new role to take on.
• 35:11
  I mean just one part of that. Made me think so much of how.
• 35:17
  You were saying a clean data set, a solid data set is worth
• 35:21
  its weight in gold. And you just think about like,
• 35:23
  wow that. Nice data pipeline that gets set
• 35:27
  up is so valuable and then if you can figure out ways to
• 35:33
  utilise that for many different use cases, it just becomes.
• 35:38
  Like profit on top of profit. Yep.
• 35:42
  And it's better than gold. I mean, you think about it like
• 35:44
  gold or oil or something like that.
• 35:46
  No. If I, you know, mine a barrel of
• 35:48
  oil, I pull it out of the ground.
• 35:50
  I did all that work to get it into the barrel.
• 35:52
  I sell it. That's it.
• 35:53
  It's gone. It's used, but that data set
• 35:56
  over and it's like that barrel fills itself back up and the
• 36:01
  only limitation is the number of use cases you can find for that
• 36:07
  particular data set. That's the only ceiling on its
• 36:11
  value. And that changes the way you
• 36:15
  think about data engineering, doesn't it?
• 36:17
  But now you're not just thinking what you know, Let's gather
• 36:19
  everything. No, Let's gather the data set
• 36:23
  that we can monetize the most. And so we need to stop saying
• 36:28
  it's the new oil. It's not.
• 36:29
  It's different. It's not gold, it's different.
• 36:32
  It's a novel asset class. Models are the same way.
• 36:37
  The more the model generalises, the more use cases it can
• 36:41
  support. And so we need to think about it
• 36:44
  that way. So what should our data science
• 36:46
  teams be doing? Not that stupid simple stuff.
• 36:50
  You want something that the business can monetize over and
• 36:53
  over and over and over again across internal and external use
• 36:57
  cases. And so you think about it
• 37:00
  differently because it is a different paradigm.
• 37:03
  And what's holding so many businesses back is they don't
• 37:05
  have anyone in the strategy role.
• 37:08
  Like you think about what's the purpose of a strategist aside
• 37:10
  from whoo PowerPoint. It's someone who can explain a
• 37:15
  what should our data teams be doing?
• 37:18
  Here's the framework to figure that out because it's different.
• 37:21
  It's monetized differently. So what does that mean?
• 37:24
  That means we need to be looking at opportunity discovery totally
• 37:28
  differently. Yes, We're going to start with
• 37:30
  one main opportunity, but we're going to decide which is the
• 37:35
  most lucrative opportunity based on all the other stuff we can
• 37:38
  do. It's not just the primary
• 37:41
  driver, it's everything. And so you say you're hearing
• 37:44
  like strategy. Wait a minute.
• 37:46
  Ohh. So that's what strategy does.
• 37:47
  I think you're the perfect person to ask this question to
• 37:50
  because there have been so many people.
• 37:54
  That are struggling with the idea of how to.
• 38:00
  Justify the ROI of large language models A because.
• 38:06
  It's really hard to get them working if you're trying to
• 38:09
  bring them in house, right? There's not really the
• 38:11
  infrastructure set up yet for such gigantic models, although
• 38:15
  that's quickly changing. And so maybe by the time this
• 38:18
  podcast comes out, that will be a solved issue.
• 38:21
  But I doubt it. But be it's even if you're just
• 38:25
  hitting an API. One big thing that I saw in the
• 38:29
  survey that we did and then we created a report on it, was.
• 38:34
  Let's imagine that you're adding.
• 38:36
  A new feature to your product and you are now making a call
• 38:42
  to. Chachi Abt.
• 38:45
  And. You aren't able to charge extra
• 38:50
  for that new feature. So you basically you just took
• 38:54
  away some of your revenue. However, I've heard on the other
• 38:57
  side of things is like well, yeah, but that.
• 39:01
  New feature, it affects a metric that we know will get people
• 39:05
  using our product more O in that case they have looked at it and
• 39:09
  they say, OK, it is worth it because we're spending whatever
• 39:12
  100 grand a year on ChatGPT calls.
• 39:15
  But we know that if we sign up to three new clients because of
• 39:20
  that new feature, it pays for itself.
• 39:22
  And so I imagine that's what you like to see someone who has
• 39:26
  thought through it that way. What are some other ways that
• 39:30
  you justify this ROI, and specifically like looking at
• 39:35
  large language models and thinking?
• 39:37
  How do I know if this is actually worth it?
• 39:40
  For me, that's the big deal. And I think it's important,
• 39:44
  which you just said that survey where people are saying, I don't
• 39:47
  know how to justify the cash, I can't, you know, get the ROI.
• 39:51
  It's expensive. The most important words I
• 39:54
  think, that have ever been said in the generative AI era is we
• 39:59
  have no mode those we need to internalise that because think
• 40:03
  about it, if I all I have to do. Is connect U to, you know, a
• 40:09
  standalone instance on Azure of GPT, whatever.
• 40:14
  If all I have to do is that to deliver that feature and
• 40:18
  functionality and get those two new customers.
• 40:21
  As soon as I get those two new customers, every competitor is
• 40:25
  going to say, oh, thank you for proving the market for me.
• 40:28
  Now I'm going to do exactly the same thing because there's no
• 40:31
  mote, there's no barrier. So you have to think about this
• 40:35
  from a strategy and a product approach.
• 40:38
  This is why both are so critical.
• 40:40
  And I love the thought process. We're spending more money.
• 40:44
  We are not charging for it, but we can monetize it in a
• 40:48
  different way. We can monetize it by bringing 2
• 40:51
  new customers on board. I love that thinking.
• 40:53
  Absolutely amazing. But we have to think even more
• 40:58
  strategically than that, which involves looking at who else can
• 41:03
  do the exact same thing, how long will it take them?
• 41:07
  And if we're spending a tonne of money on those two customers, on
• 41:12
  acquiring the customers, on proving the market for it,
• 41:16
  what's going to stop our competitors from taking
• 41:19
  advantage of our investment of creating the market, of proving
• 41:23
  it and just entering and saying hey, guess what, we have an
• 41:27
  alternative, same thing. You know first two doesn't mean
• 41:30
  anything. It's it's not, you know, saying
• 41:33
  you're the first one there doesn't help very much because
• 41:37
  if someone else could offer the exact same thing and be third,
• 41:41
  then you have no Moat, you have no advantage.
• 41:44
  There's no way to protect that investment you made in getting
• 41:48
  those customers. So the first thing I always say
• 41:51
  is if you're introducing the feature and no one's willing to
• 41:55
  pay for it, why are you doing it?
• 41:57
  There has to be something else. We look at the opportunities
• 42:01
  that are right in front of us, but the opportunities that are
• 42:05
  right in front of a business may not be the opportunities that
• 42:07
  are right for the business. There are so many different
• 42:12
  things that you can do with generative AI.
• 42:16
  If you understand the paradigm, you can do opportunity discovery
• 42:19
  like I said, in a completely different way.
• 42:22
  You have tonnes of different opportunities that could deliver
• 42:26
  value to the business, but you have to look at the ones that
• 42:30
  have the highest potential for returns.
• 42:34
  Those should be the ones you're doing first, because if you can
• 42:37
  fine tune a model or if you can embed business expertise, domain
• 42:43
  expertise into your prompts, or you can use some combination of
• 42:47
  fine tuning and prompt engineering, you can create
• 42:51
  something that your competitor can't just swoop in and do
• 42:54
  themselves. They will not be able to
• 42:57
  successfully follow you and that's your Moat in the next
• 43:02
  probably 18 months. It's the data, the domain
• 43:05
  expertise, and it is really that strategic component of it as
• 43:09
  well. So you have the talent, the data
• 43:10
  engineering, the data quality, the data scientist.
• 43:14
  Yeah, there's some prompt engineering, you know, few
• 43:16
  buzzwords on the other side of that.
• 43:17
  But there's also that construct of domain expertise.
• 43:21
  There's also that construct of the business, understanding the
• 43:25
  product paradigm and the strategic paradigm because that
• 43:29
  will be a barrier for the majority of businesses until
• 43:33
  they start bringing these new capabilities in house.
• 43:36
  It's almost like trying to do data science without data
• 43:38
  scientists. It's the same thing.
• 43:40
  You can't have an AI product without someone that understands
• 43:43
  AI products. It's not going to work.
• 43:46
  You can't build out an AI strategy.
• 43:48
  Well, unless you have somebody that understands it.
• 43:51
  So there's these barriers to entry that most companies are
• 43:55
  six months, 12 months, 18 months away from even being able to
• 44:00
  deliver a copycat. And that's how we should be
• 44:03
  thinking first. It's that prioritisation, making
• 44:06
  sure that we're really going after the big winners first, not
• 44:10
  the obvious winners, because the obvious winners are going to be
• 44:12
  obvious to everyone else. That doesn't lean on your
• 44:15
  competitive advantage. So bring that knowledge that you
• 44:19
  have as a business to bear your AI product sense, your AI
• 44:22
  strategy sense, quality data, internal domain expertise.
• 44:26
  You can see this is enterprise wide.
• 44:28
  That's the change here. You could do digital as a silo
• 44:32
  project. You can't do that with data and
• 44:35
  AI. It's bigger.
• 44:36
  So that's the advantage of having some of these components
• 44:40
  in place. Even if your data is not
• 44:42
  perfect, you know at least exactly how to target
• 44:45
  initiatives at the data you need in order to be successful with
• 44:51
  those opportunities. You see it's an advantage and
• 44:55
  and companies need to start going in that direction.
• 44:58
  And so you mentioned before how? Different.
• 45:01
  Data product officers or data product managers are being
• 45:05
  embedded into teams. Is that how you've seen it being
• 45:08
  most successful? And if I am in the enterprise
• 45:12
  and I'm a data scientist or a machine learning engineer or
• 45:14
  data engineer? And I'm looking around what are
• 45:18
  some things, Can you give us some examples of how you would
• 45:21
  go about? Adding that extra value OI
• 45:25
  would, I mean first evangelise that you need new roles.
• 45:28
  Half the battle is just talking about it inside of the business.
• 45:32
  If no one understands that there are jobs to be done that are not
• 45:37
  being done, then you're going to come up against this brick wall
• 45:40
  where you say, hey, we need these new roles, OK, nice.
• 45:43
  Next, anybody else? Is there.
• 45:45
  Anyone else has something that they would like to talk about?
• 45:48
  So the the first message really is for data professionals.
• 45:51
  Hey, we need new roles and you can say some fairly obvious
• 45:54
  stuff. You know right now technology is
• 45:57
  leading strategy that's going to end in technology, but when you
• 46:02
  have strategy leading technology, you end U with
• 46:04
  value. So let's fix that.
• 46:07
  And now you're talking about the new role, the strategist and
• 46:10
  talking about a technical or AI strategist.
• 46:13
  We have plans for data and AI products.
• 46:16
  Shouldn't we have someone who's a product manager that
• 46:20
  understands data and AI products, you know, and it's
• 46:23
  these obvious just one liners and two liners that you can
• 46:27
  deliver that'll change the way everyone around you thinks?
• 46:30
  It'll change the way that other leaders outside of the business
• 46:33
  think about this from an organisational development
• 46:36
  standpoint. I've got like 2 chapters in my
• 46:39
  book that are really long about organisational development.
• 46:42
  So it's kind of a complicated question to answer quickly, but
• 46:46
  the best way to think about it is you have product managers.
• 46:50
  If you have a product management organisation, that's awesome.
• 46:53
  As long as they're open and collaborating across the
• 46:56
  business, you're in good shape. You can put your people there.
• 46:59
  You could put data product managers in the data
• 47:01
  organisation, but again only if there's that open collaborative
• 47:06
  relationship. If not, the data product manager
• 47:08
  is going to have to create that and from within the data
• 47:11
  organisation maybe some challenges that may be harder.
• 47:15
  Your strategist should sit at whatever strategy, organisation,
• 47:20
  product organisation that you have.
• 47:24
  Some companies don't though, so again, they can live in the data
• 47:27
  organisation. You need a CDO or a C level data
• 47:31
  leader or executive at least. Level data leader who sits has a
• 47:36
  seat at the table at the C-Suite and can help with strategy
• 47:39
  planning. That can become that technical
• 47:42
  pillar of strategy where they're not planning strategy.
• 47:46
  That's the strategist role. They're going to lead it.
• 47:49
  They're going to be the ones who take it and implement it.
• 47:52
  And so they need a seat there. They're also part of the
• 47:55
  education process. You can hear all of these roles
• 47:59
  will continue. You know those one liners that I
• 48:01
  gave you at the very beginning. They're going to continue that
• 48:04
  process of education where we are bringing data and AI
• 48:09
  literacy to the rest of the business.
• 48:12
  But there's two sides to this. They're also bringing business
• 48:16
  literacy into the data organisation.
• 48:20
  O every time they're teaching the C-Suite about data and AI
• 48:24
  paradigms and how it impacts strategy and how it can create
• 48:27
  opportunities and how it can align with core goals.
• 48:32
  They're also bringing some of that strategic acumen into the
• 48:36
  data organisation and it's a multiway sort of exchange of
• 48:42
  knowledge where the business becomes data and AI literate but
• 48:47
  the data team becomes business literate and that's so important
• 48:51
  for that two way communication to happen.
• 48:54
  So it's not just you doing 1 liners for the rest of time, you
• 48:58
  know, if you want to be a data product manager, I'll teach you.
• 49:01
  I love it. Great.
• 49:02
  Thanks. You want to be a strategist?
• 49:03
  Yeah. Come on in.
• 49:05
  I'll help you. I'll teach you.
• 49:06
  I'll give you classes, read my book, whatever.
• 49:07
  But if you don't, this is also really valuable because it lets
• 49:11
  you get back to doing what you love doing.
• 49:13
  If all you want to do is data engineering, advocating for that
• 49:15
  role, bring someone from the outside in so you don't have to
• 49:19
  do this anymore. And that's one of the big things
• 49:22
  that I tell you know, data professionals is if you don't
• 49:25
  want to do this, advocate for the roles so that you don't have
• 49:29
  to. You're often bent into these
• 49:32
  roles that you shouldn't be bent into.
• 49:34
  And it's so much more effective if you can get someone who's not
• 49:40
  only got the capability set, but also the desire to be doing
• 49:44
  this, to take on those roles and responsibilities.
• 49:47
  And now you're freeing up the data team to do what it does
• 49:50
  best. Damn.
• 49:51
  All right then. I love this man.
• 49:53
  We've chatted for long enough for anybody out there that has
• 49:56
  not read Vince book. We will put a link to that in
• 49:59
  the description. Maybe we can give away a few for
• 50:01
  the first commenters in. Our YouTube channel or a couple
• 50:06
  of coffees, I can send this. Yeah, we go.
• 50:08
  Here, that's what I'm talking about.
• 50:10
  Yeah, alright, I might be able to bring a couple out.
• 50:13
  Or send me a slack message or an e-mail and we can give some
• 50:17
  discount codes to imagine. Then is kind like that.
• 50:20
  I'm giving away his whole handed out stuff.
• 50:24
  That you want to talk strategy, This is the way to not do
• 50:27
  strategy. Give it all away.
• 50:29
  And actually no that is actually that's a great strategy, give it
• 50:33
  all away, it'll come back in 10 fold, but.
• 50:36
  That's a whole another podcast man.
• 50:37
  This has been super enlightening.
• 50:39
  I really appreciate talking to you.
• 50:41
  Every time I talk to you I have a blast and as I said before you
• 50:45
  created one of my favourite shirts that we got in the merch
• 50:48
  shop. It is Kubernetes is a gateway
• 50:50
  drug and we still have that. It's one of our best sellers.
• 50:55
  Besides now I I also. I don't know if you saw the
• 50:58
  other one. I think you did, it says.
• 51:02
  Yeah, I whose name more than ChatGPT.
• 51:04
  That was the other one that we created that I got some one
• 51:06
  liners. There we go man.
• 51:08
  I need more of those so I can make more shirts.
• 51:11
  But anyway, that's again, for our next conversation.
• 51:14
  We can just have it all be 1 liners and dude.
• 51:20
  All right, I'm going to try and steal some of your time now so
• 51:22
  you can give me some strategy on community building and all that
• 51:26
  fun stuff too. Oops.
• 51:31
  You have to immerse yourself in the ML OPS content.
• 51:35
  The best way to do it is to subscribe to the Inner OPS
• 51:39
  Community Podcast, so. Good luck and keep learning.