The Real Reason AI Projects Fail with Karina Arteaga

AI feels like it should be a shortcut to productivity, yet most teams still struggle to turn pilots into real business value. 

We sit down with Karina Arteaga, CEO and founder of Visible Global and former operations leader at Meta Reality Labs, to unpack the uncomfortable reason: AI projects fail less because the tech is weak and more because the organisation is unclear, messy, or misaligned. 

If your objectives, decision-making, and workflows are broken, automation just scales the chaos.

We explore what it takes to build a human AI operating model that actually works in the real world, where people have fears, incentives, and habits.

Karina shares lessons from building an operating model from scratch in a high-growth, AR/VR and AI environment, and she explains why leaders should start with basics: map the workflow, clarify ownership, and decide what outcomes matter before choosing tools. 

The conversation moves into agentic AI and autonomous workflows, and why that shift makes human judgement more important, not less. 

Finally, we get practical about change management: listen to employees, solve the most painful parts of the job first, and use internal champions to drive adoption without creating a Big Brother culture. 

If you want AI transformation, organisational design, and leadership culture to pull in the same direction, subscribe, share this with a colleague, and leave us a review.

We would love you to follow us on LinkedIn! 

https://www.linkedin.com/company/amplified-group/

Transcript

Welcome And Guest Introduction

Sam 0:09

Welcome to Get Amplified, the podcast about the people that power the tech industry. Morning, Vicky. Glorious Sunshine, where you are. It is. Lovely here.

Vic 0:18

Yeah, yeah, it feels like.

Sam 0:20

The last thing I want is to be stuck inside on a podcast with you. I'm joking, I'm joking. I'm delighted to be here as always.

Vic 0:32

That's put me in my place completely, hasn't it, Sam? Thank you very much for that. Well, I'm very pleased to see you. Thank you.

Sam 0:41

Glorious Sunshine when you are too.

Vic 0:43

Yes, it is. It's absolutely yeah, spring has arrived, hasn't it?

Sam 0:47

Yeah, it's wonderful, wonderful to see. So who have we got on the pod today?

Vic 0:52

So today, Sam, we have with us Karina Arteaga. And sorry, Karina, if I've not pronounced that very well. But Karina and I worked together when VMware acquired AirWatch. And so Karina has now, gosh, she's had such an interesting career since we last worked together. I thought it would be a fabulous opportunity to get a really wonderful inspirational women leader on the podcast. Because we don't have many of those, only because I just don't have the connections that I should have. So Karina is CEO and founder of Visible Global, and she's doing some very, very cool stuff. So uh Karina, it's great to have you. Thank you.

Karina 1:40

Thank you for the invite, uh Vicky, and very nice to meet you, Sam.

Sam 1:45

Nice to meet you too. You're very welcome. We're delighted to have you here, Sam. Sounds like uh we're in for a fascinating half hour or so. So maybe you could start by giving us a run through your career and how you got to where you are today, please, if you don't mind.

Karina’s Path Into Tech Leadership

Karina 1:59

Yeah, sure. I started my career in tech 13 years ago. And I I currently have my own consultancy firm that I started last year, where I help leadership teams actually uh build new operating models, human AI operating models around all of different uh changes that are actually undergoing in this new uh market landscape that is very different with AI in the scene. Before that, I spent several uh years in large organizations like VMware, where I met Vicky, uh, but also I would say most of my formative years at Meta, more recently, where I was the first person they hired in the operation side on the reality labs business of Meta, which is basically the business unit dedicated to AI, AR, and VR products. It was very fun to actually build our operating model from scratch. Uh, we we were successful, and and actually the model that blueprint was rolled out to North America and to APAC. It's not it's not the usual motion, actually. But this this time uh was the other way around.

Sam 3:17

Sounds like you prototyped it in Europe almost in startup mode and then found what worked and then spat it back over to the American head office.

Karina 3:26

That that's right, that's right.

Sam 3:28

That sounds fascinating. So first operations are globally at Meta Reality Labs. You must have learned a huge amount in that period. Indeed. Tell us about that.

Karina 3:40

It's a lot because when you enter a large organization like Meta, you normally think, you know, they they have everything sorted, you know, they they they have everything figured out. It's a large company. I I started at Meta in 2019, and there were a lot of um, you know, different scenarios and businesses going on. At that time, I I would tend to think I started actually in the reality labs team that at that time was called ARVR uh business. It was not really developed, it was almost like a mini startup within the big world of beta, if that makes sense. And um, and although they they did have the resources in terms of funds and financial resources, uh it was a tiny team that that had to kind of build from the ground up entirely new categories. And I think that's uh that was the most interesting part, uh Sam of the job. Because you're not talking software, you're not talking, I don't know, accommodatized type of product. It's just a set of products that is very, very unique and uh and entirely um unknown for customers, for partners. Like it's it's so that was the really exciting part. And when you enter the organization and you realize, okay, they have a basic um CRM going on, but everything is in spreadsheets. There's no operating systems, there's there's no processes, but we can start this actually from the learnings of what we haven't done uh right somewhere else. Um it's actually an exciting opportunity that that you can't you know not be thrilled to jump into.

Building Reality Labs Ops From Scratch

Sam 5:27

But you've taken that experience and turned it um or your other experience into this consultancy business that you started. So, how does the stuff that you've got from Meta Meta um help you to advise leadership teams today?

Karina 5:43

I think mostly, you know, that this connection that you see between um a technical project, the thing is that engineers are brilliant actually, and they think uh, you know, in processes, but they tend to think very heavily and and go deep down into the technical aspect of an implementation. Um, but then there's this connection with actually where do you stand in terms of impact on the on the real business outcomes, um, impacting culture, you know, it's people that are actually gonna adopt AI. That's that's kind of completely disregarded. How you're gonna enable those people to actually uh leverage the tools the best way. When you see that disconnection, it's where you see that gap actually grow and impact negatively. So uh most of the times people start leadership teams, especially, uh, wanting just to tick the AI box, right? Because it's what they're supposed to do now. There's a lot of comparison with competitors, but at the end of the day, there's no one size fits all. And I think this is something you know very well. Uh it applies not only to AI, but to many other uh solutions uh in terms of uh building team culture, productivity, and efficiencies around processes and systems. So ultimately, it's just about you know really sitting down, understanding the plan, what do I really want to achieve uh and accomplish? And then uh starting starting from from there.

Sam 7:23

So you mentioned 2022, so yeah, this this latest sort of wave of AI, the generative stuff has been kicking around for three years ish now. Are people starting to see returns? Are boards happy with what they've done with AI, or is it all still hype? Where are we on the hype cycle? Are we at the bottom of the trough of disillusionment? Are we on the way up and out of the what is it, the the slope of enlightenment, whatever it's called?

Karina 7:53

No, there's actually an interesting study that it was made by MIT just towards the end of last year, but they they actually brought up a very interesting uh stud, which is basically only 20% of AI projects right now or or intents to improve processes are actually successful. And the reason why that is the case is is mostly not because there's any technical constraint. I mean, we know you know how powerful this technology. Exactly. I mean, no no doubt about it. Um, but the problem is that normally what is broken is your own organization internally, like your structure is not clear, your objectives is not clear, probably your coverage in terms of people um versus uh goals uh that that you want to achieve are not uh necessarily correct. So basically, you are automating this function, and and you need to start sitting down and first redesigning a little bit so you don't end up just automating something that is broken from the very beginning. Um like that.

Sam 9:00

Don't don't automate dysfunction. That's uh yes, brilliant. Yeah, that's a good tagline.

Turning Meta Lessons Into Consulting

Karina 9:07

Yeah, but you know, ultimately it's it's really true because what what we are seeing is that the the adoption is fast, sometimes it's chaotic, but because uh a little pilot has been rolled out, uh, but users are still going to use their preferred um use case, they're gonna they're gonna actually start um resolving their own problems on their own terms, there's no governance, so there's like the system is broken because only the technical part of the project has been taken into account. Yeah, a lot of technical teams are just measuring um, you know, rate of adoption or how many active users do we have every day. But sometimes you have uh the early adopters that are in every team, you know, the very um uh hyped uh enthusiasts that are gonna heavily use a tool and that perhaps actually are skewing that number for you because they are making up out of your 80 percent of usage of your AI tool internally, is just used by 20% of your employees because those are the ones that really want to use it, right? But disregarding a huge bunch of people that actually can can really impact negatively, whereas it's because they're scared of using the tool, they're scared of uh you know so many uh layoffs or or the fact they're giving away their IP or their language, right?

Vic 10:41

Yeah, yeah, I can so relate to that, Karina. Um, so one of the studies that we quote every time we start with a new leadership team is McKinsey did a study and they looked back at the transformations that they had worked on over the last 20 years, and they said successful transformations only happen when you spend an equal amount of time on the people piece as well as the truth actually, what are you trying to change, the work piece that you're trying to change. They said if if you only focus on the work bit that you're trying to change, you have a 30% success rate. If you focus equally, and this isn't about HR doing it, this is about the leadership team, if you focus equally on the people bit, your success rate goes up from 30% to 80%.

Sam 11:38

Isn't that always the story? Yeah, why we do what we do? This this is not specifically an AI problem. No, it's not, it's it's a tech problem.

Vic 11:47

Yeah, but I can see that because as as Karina's just said, the rate that AI is being expected to be adopted, that pressure is really having a massive impact for people to do it. And this whole point that you've made, Karina, of don't automate dysfunction. God, there's so much dysfunction out there.

Karina 12:13

It is, it is because ultimately real AI transformation is an organizational design problem, right? It's not a technology problem. And and and if you don't start fixing that, of course, you need to like people want to see results um in the short term. Sometimes they they take longer as well. I don't want to call it a bet, but you know, somehow it's an investment. You're doing your front loaning to actually see um uh the results later on. But but the reason why largely I I think most companies are are seeing these problems is that because despite they have some signals that this is the case, that they're you know about to to kind of automate some kind of dysfunction, unfortunately, no one's to own that responsibility.

Vic 13:04

Yeah.

AI ROI And The Hype Cycle

Karina 13:05

So at the end of the day, when you know AI transformation is everyone's responsibilities, it ends up being no one's responsibility. Nobody's responsibility. And I think that's that's uh what I try to normally start with, uh, because if you have that little foundation solved and ready, and it doesn't need to take ages to actually solve, it's just being clear in where you're going and that path that's gonna take you there is a lot easier. But that acknowledgement, uh owning that responsibility is important, and and and that's why I think AI leadership teams need to understand that that's that's something they can't disregard. It's not it's not a nice add-on at the end of the uh of the foundational. It it needs to be considered from the very beginning, yeah.

Sam 14:03

Yeah, it's the old, and you know, it's become a bit almost a bit of a cliche, hasn't it? We it we seem to mention it every other podcast, but people process technology in that order. Yeah, so you know, I I think I first heard Joe Bragley talk about that um maybe 15 years ago or something like that. And you know, I'm I'm not sure he would lay claim to having invented that, but that's the way we have to deal with these challenges, isn't it? Um we start with the tech because it's cool and shiny, and you know, I don't I'm I'm guilty of loving a new whiz bang, as we used to call them with subcats, as much as anybody else, but it's not the place to start.

Karina 14:44

Yeah, yeah, yeah, yeah. I think if you don't have clarity, you know, how your workflows run, how you make decisions, how like each team are operating independently and are not becoming probably a bottleneck for another one, you know, in in a in a in a big process, is like how can you sit down and actually um figure out if AI is the solution or just a simple automation, right? I mean, the thing is that the reality is that most uh companies, I would say, even enterprises, are not necessarily the ones that are gonna be building AI fantastic bespoke solutions for themselves. They're just gonna be actually using an AI embedded capability in a tool they already use today. So if you don't start by understanding what you have already and you're already paying for, by the way, and the best way to actually leverage what you got, how can you actually understand what you actually what you need on top of everything to really boost and accelerate that process?

Vic 15:53

So, what you're basically saying is back to basics to start with.

Karina 15:56

But yeah, back to basics. I know, as boring as it sounds, Vicky.

Sam 16:00

Yeah, yeah, yeah, yeah. So you've got a system for helping people sort this stuff out, a human AI operating model. Yeah. Tell us about that.

Vic 16:12

That's very cool, Karina.

Sam 16:14

Yeah. No, it's it's just precisely cyborg geek.

Don’t Automate Dysfunction

Karina 16:19

So it's precisely this framework that helps you understand, you know, all of all of all of that, those workflows, the SOPs that you have, the you know, KPIs, OKRs, like really understanding where you're going. It normally can help you if you kind of apply that that uh model. Um it helps you just uh plan then your AI automation or uh transformation because sometimes it's a it's a combination of different uh types of tools and different levels of uh that are at uh sit at different levels of maturity and and complexity as well. Uh it just helps you organize all of that body of work uh nicely and with a mindset in actually humans that are the ones that are gonna be using that. They can make the whole difference in in you know between breaking the system for you or actually you know making you just skyrocket and fuel your your company. So it really makes the the entire um difference right now, I think. Um so that's basically it's a model that obviously it depends on each case, uh what each company needs to accomplish. Uh but but yeah, it's it it basically uh helps you put everything in in a more clear view uh more strategically, if that makes sense.

Vic 17:48

So Karina, because the the conversation that I've been having in my head whilst listening to you is is it is it to help the creation of the AI system or is it to help with adoption? But then I was thinking, and I need to I need you to get me straight on this. Well, actually, if you haven't got the right system created in the first place, it's never going to be adopted. So it helps both. So just give give me the clarity around those two bits.

Karina 18:18

Yeah, if I'm brought in into the scene, you know, in the right time, you can always fix later and you know, kind of patch it. But uh ideally, if I'm at the very beginning of a project, basically that creation is is a lot seamless and um and and and a lot smoother, is what I wanted to say, right? Because it has better chances to actually being adopted, yes, it has better chances of not impacting negatively your actually team morale, uh building a better culture as well. I think sometimes these times of transformation, what I've seen is that it's the perfect time to actually revamp your culture. If you already have a grade one, fantastic, you just kind of uh build up on top of those. But if not, is the best time to actually sit and rethink? How can you actually uh cultivate a more healthy um productive uh culture, right? Because it needs to be somehow built in systems, like you leadership team, the leadership team needs to operate at that level and the rest of your team as well. Otherwise, you know, you wouldn't feel inspired, motivated to follow anyone on something you don't believe in. So you really need to again, I think is is the perfect time for leadership teams to demonstrate where where leadership really stands, and it's not just because of your job title, right?

Vic 19:54

Yeah, yeah. Actually, we had a really good example of that, Sam, didn't we? With Darren Theron talking about Google and how Google had adopted AI and become an AI organization because of the way that leadership was set out in the first place and the parameters that were put in place for it. But you're obviously talking about it on a very, very different scale, Karina.

People First Or Transformation Fails

Karina 20:16

Yeah, and somehow it's a learning I brought from Meta, right? I mean, if you think about it, it's a it's a strong culture, um, really with very clear pillars, uh, and and and where everybody operates on that mindset, despite how big or small your unit division could be, or how big or small the business is. Like, I mean, meta is a is a is obviously a massive business, uh, you know, probably close, uh bigger, bigger actually, of of many countries, uh GDP, but um but and that's the reality is is is massive, but despite being so big, the mentality, the culture inside is always operating as it as if you're just starting a project. And that applies to everything, even when you finish and deliver like an amazing project. In my case, I was um you know responsible for the go-to-market uh planning and execution uh piece. Whenever we were launching already um a product, uh amazing product like the smart glasses, the meta Raven smart glasses, or or the Oculus headsets. When it was Oculus, then it it passed to be Meta VR headsets. The it, you know, all of those fantastic products, you just launched it and you were like, oh, okay, we're back to square one. We're starting to build again. And that's the mentality because you grow when when when because it's so well settled uh in in the in the mindset not only of the leadership team, but every employee since you enter, you're really well trained to think that despite you've done your best. work or you feel you've delivered a strong a strong outcome you can still do better because uh you know

Vic 22:08

I really like that mindset that's Adam Grant that yeah and and this is no such thing as best practice always better

Karina 22:14

yeah and this is something um by the way Zuckerberg always uh repeats sock said in the in the first i i still remember being in the first uh uh all hands in person after the pandemic which was obviously very exciting because we haven't seen each other for a very long time I was lucky enough that actually had a leadership meeting uh that day and that day was the old hands for the entire company um led by Mark and he said like remember we were always just one percent at one percent of the journey so we're not finished and and it's that mentality what actually keeps people innovative bringing great ideas um trying to to hack things that don't work uh it's very much ingrained in the culture and I think that that's very important as well when you're rethinking um you know your entire company because you really want to leverage this um incredible uh technology to to grow but you it it's not only the business that needs to grow it it grows with your people like if you're not you know actually looking to grow your people how can you

Vic 23:29

grow your business right things I love that yeah yeah grow your people to grow your business yeah

Sam 23:36

yeah

Karina 23:37

as AI get smarter by the way human human judgment actually becomes more and more important it's it's that's line of defense I would say yeah yeah yeah

Sam 23:48

that's a that's a yeah I think that's the point yeah yeah

Human AI Operating Model Explained

Karina 23:53

actually I can't remember to be honest I don't want to quote it like exactly but I I remember have read recently another paper but um it it it actually like shows in in stats how more impactful the the outcomes of a team that includes humans in the loop versus entirely uh automated agentic workflows is actually incredibly impactful when you have the right humans at the right points of any workflow to be there to apply that judgment um that ultimately you know gets not only your business results better but um you know your customers happier because they feel they they're they're you know they're they're human as well as your employees so it's um it's impacting everything

Sam 24:49

so we're in a oh sorry Karina

Karina 24:52

no sorry i i was just going to say not only hallucination which is obviously the technical term like engineering teams use but it's also um there's a lot of potential bias that you need to be very careful with so um and and aware of um i i used to back at meta used to be very active in in what they call the red teams the red teams uh are actually the ones putting under pressure the models and like you know yeah so almost to destruction yeah exactly and and we used to test it right to kind of catch on time before things uh were were shipped and and released to the public um you know and any potential bias so i think it's um uh that that's very important whereas you do that at a big scale or a or a smaller scale uh for any enterprise or even large company but it it really needs to be tackled on time and that and hence why it's so important that the humans are in the loop somehow um serving almost as your quality control and and yes and and also factual checkers like you know they they hold knowledge that not not necessarily is documented anywhere in your um uh company so that this is this is a time to actually uh make that knowledge even more special

Sam 26:19

yeah makes sense so it feels like the narrative has moved on it was generative AI and now everybody's talking about a genetic AI and using tech to run workflows autonomously um does that change anything from a leadership point of view is are humans still critical or are they even more critical in that that shift

Karina 26:47

I think they're even more critical both when you're actually mapping out identifying what you want to do because it's very hard uh to build um agentic support if you don't really know you know what what you're trying to do so normally I think the best way and the most efficient way to start those processes sometimes is reverse engineer yeah you you kind of see what you're doing how you're doing it right now you find the flaws um identify and map out that um that process and then you bring on the the the agent so actually the humans needs to need to be part of that first phase and then of course they they you will need to know exactly when you're mapping out that process where they need to be intervening because it's it's it's actually giving that extra uh security layer uh in your whole um workflow um and then even for lessons learned like the system of course can automatically learn um and and keep training on itself but the human that human input in in the post mortem as I like to is so key to really see your um uh your system exponentially improving especially on on agentic um uh deployments

Vic 28:20

yeah that makes sense

Sam 28:22

makes a lot of sense have you seen anybody get this stuff right how do people go about getting it right

Karina 28:29

getting it right is is it sounds easier than

Sam 28:32

obviously obviously employing you in your consultancy company that that's a good start but

Humans In The Loop And Bias

Karina 28:37

yeah yeah um no i think um what i see companies doing really right the ones that are that are more successful is really um trying to to um instead of just uh running uh crazy with a pilot that perhaps was boiled in a in a in a in a small um you know leadership room is just going out of you know out of your comfort zone perhaps but you know asking really your employees especially in those areas where you see more more critical uh impact um how they want to you know what's what's the worst part of of of their job uh what they would really need it need help with and and starting there once you have kind of proved that it's also very smart uh and i've i've seen this uh done really right when you get uh you know those those uh early adopters those enthusiasts that i was uh speaking of at the very beginning and embed those in your actual enablement plan for the rest of the organization because they are going to convince beta you know the rest of their colleagues of of the beauty of this system of the benefits of of these new tools uh more than just a mandate that comes hierarchically from above so um that kind of horizontal approach i've seen it done really really well and um and i think is is is very key to success

Sam 30:15

yeah that makes sense almost having colleague advocates but some sort of horrible phrase like that to yeah to get people on board rather than it being a a management mandate yeah champions yeah yeah

Karina 30:29

yeah champions ambassadors call it i mean you can call it whatever but ultimately it's not the same when actually a colleague is saying to you hey i i do my job faster doing this than you know than some something that sometimes uh feel feel imposed i actually have have seen other not great cases uh as well um especially now in big tech uh without revealing any any details but it's more for instance like uh ai used in middle as almost as middle management layer and trying to do check-ins and saying hey I don't know if you're an engineer you haven't spent you you've spent only 70% of your time this week uh coding what what's going on like you need to be spending this percentage

Vic 31:19

yeah sounds like a big brother type approach isn't it

Champions Beat Mandates For Adoption

Karina 31:22

precisely and and people are hating it and it's really hurting the morale of teens so you need to be like super strategic and very careful in how you really deploy things because I'm I'm seeing a lot of um uh of these and and people don't like it because they're people so they don't want precisely that that big brothery feeling um on them even though I mean we know um big tech companies are very automated I remember like uh at meta we even our thank yous are counted like thank yous we give thank yous we receive but like yeah it's really it's really automated and it and it's great like they're there because they're meant initially the intent of those tools actually is very good is is promoting um a culture of kindness of saying you know um yeah thank you for helping me uh you know to another colleague all of that by the way it goes into into or roll up to into performing cycles etc like when when people are assessed and that's that's really great but I think you know at the moment you start detaching that from a human element entirely you you're really risking you know the and compromising I I think um you know the the culture of your company so I I think you know it's it's it's mostly about sitting down being smart and that doesn't mean you can't try something sometimes we try we implement we see it doesn't like um it doesn't work and we move on from that we we take out those learnings right that brings and repeat process uh is is super important that's how we grow that's how we build great things but I think it's really being being strategic about it

Vic 33:18

yeah yeah makes complete sense

Sam 33:21

interesting interesting we should probably uh bring the the podcast towards a conclusion any last words of advice sort of summaries takeaways for our our listeners

Karina 33:35

yeah I think ultimately understanding that ai is not coming for your company is not coming for your job I think it's it's actually come coming for for your operating model is coming for the way you work if you don't want to learn anything new probably as an employee you're gonna struggle and and as a as a team lead uh as a leader in in your organization if you're not thinking that you need to actually redesign uh your model your org uh probably uh you know the technology is is gonna come for you so leaders who redesign first are the ones that will be still standing i'm not gonna attempt to to sing uh uh elton john song but they're gonna keep still they're gonna keep keep there and and and still stand so yeah so yeah that's that's what I think

Sam 34:32

that brilliant thank you we got any book recommendations for our listeners

Final Advice And Book Recommendation

Karina 34:37

yeah well I love actually Grit from Angela Dockworth um I think you know she she presents a a a very interesting model of of this work it's not a new book is to I think actually by the time I was starting to work with Vicky 2016 or around there um is is when that work uh that book uh uh was was actually um released but I think it it brings a great uh concept I would say which is uh the concept of greed and it basically the core idea is that talent is overrated that at the end of the day it's all about um you know work perseverance perseverance yeah and passion for what you do that you never surrender even you know it doesn't matter how many times you go down you always you know stand up yeah and keep learning and that that that kind of determination is really yeah that resilience but that that determination that um you know it doesn't mean that you're not gonna feel bad when something goes wrong but it means that next day you're gonna brush that off learn from what you're doing and continue that makes sense

Sam 35:59

brilliant thank you for that anything to add Vicky

Vic 36:02

no thank you i i've really really enjoyed yeah fascinating uh listening to you karina thank you i particularly like your don't automate dysfunction i think that is flipping brilliant

Sam 36:15

yeah yeah that's almost the title of the podcast isn't it yeah there you go

Karina 36:20

there you go yeah yeah amazing

Sam 36:23

no all good thanks Karina we really appreciate you being on the pod I think our listeners will absolutely love it and it just remains for me to say thank you for listening to get amplified from the amplified group your comments and your subscriptions are always gratefully appreciated