Publicis Sapient CEO Nigel Vaz: 93% of enterprise AI pilots fail to scale — here's why and what the fix looks like

is the leading AI trust management platform. You can go check them out. And our next guest, Nigel Vase, is already in the Reream waiting room. Let's bring him into the TVP Ultradome. Nigel, good to meet you. How are you doing?

How's your new year going?

Off to a good start.

Thanks for having me. Great to be on. Yeah, it's great.

Thanks so much. Uh well, we'd love to kick this off since it's the first time on the show uh with a little bit of an introduction on yourself and uh and how you introduce yourself to everyone who's watching. Yeah, I I run Sapient, which is a enterprise AI software and technology business. Sapient was founded about 30 years ago in the early days of the internet, building some of the world's first online banks, the first time you could trade equities, the first time you could pick a seat on an airplane. So we've had about 30 years of building digital systems to really build uh intelligent enterprises in the first wave of digital and now leveraging AI. We're all about how we can actually unlock real transformation beyond pilots uh for some of the largest enterprises in the world. So what we really do is bring together deep industry expertise across sectors from financial services, retail, healthcare and then combine it with our digital DNA um alongside a a few AI products that stem from agentic orchestration through to accelerated software development and then finally managing uh IT systems using AI. How did you process that data point that kind of shook the internet a little bit for a couple days last year where it it suggested that a lot of the AI demos at the enterprise level the experimentation those those experimental budgets uh that they weren't getting renewed that they weren't seeing the level of adoption

they weren't getting renewed it's just that they weren't like

they weren't converting or they weren't working and it was some crazy high number like 93% of enterprise AI tests

yeah yeah so how do you process MIT study. Yeah, that's right.

Yeah, it was an MIT study that came out and actually what we did is we had a bunch of professors evaluate from MIT our work almost as a counterpoint to that data. Um because what we were seeing is of course there was a ton of experimentation in the organizations around AI. But trying something out and then scaling it across the enterprise was one of the biggest stumbling blocks largely because building a proof of concept in a small organization and scaling it is a lot easier than when you are one of the largest automotive manufacturers in the world or when you're one of the largest banks in the world because you have to actually think about the tool chaining of the organization and how these systems deploy in the context of that right most of these organizations I mean Davos is coming up in a couple weeks and Every time I go I over the last couple years AI has been the dominant conversation but more and more clients are now saying hey we understand the concepts around AI we understand all of the conversation is around compute and all the conversation is around models but how am I as an enterprise leader getting value from this in the context of what I do dayto-day

and that is exactly why that stat is so critical because I think essentially the bridge from you experiment experiment with a little bit of AI to you can actually start to see meaningful gains outside of the obvious use cases of you're deploying AI in the call center to route some calls more efficiently or you're deploying AI in uh the context of enabling your sales teams to be more efficient, right? We're talking about largecale uh agentic mesh deployments that are allowing businesses to reimagine a car build that used to take 18 months in 18 weeks. Yeah.

Or allowing an insurance company to monitor worldwide events and on a dynamic basis allow you to react to the fact that I heard one of you guys talking about your house getting burned down in California. And this is an actual piece of work we did for a client where um they were monitoring the air quality post fires

and alerting people who have significant issues with asthma to actually, you know, make sure they had uh uh inhalers or maybe even perhaps moved out of those zones proactively, saving themselves hundreds of millions of dollars in health care and better outcomes for these folks who otherwise might have gone through a traumatic health event, right?

There's a white fil that's the way

completely different orchestration uh than I think a pilot or two.

Could you share uh some more as as much as you can like who your actual like clients are because I think that'd be helpful to ground the conversation because it's some of the biggest companies in the world.

Yeah, it is. It's it's it's some of the biggest banks in the world going all the way from retail banks to investment banks like the likes of a Goldman Sachs for example in the investment space in the in the um retail space we work across the um uh uh the retail space across essentially grocery retail and then uh fashion retail as well as uh general product retail. So we work with the likes of a car 4 in France or a Walmart in the US uh Tesco in the UK. So a number of big businesses there. Uh in the context of the automotive space, the likes of uh a big a number of the big German automotive companies, a number of the big US automotive companies, Asian automotive companies. So we we serve multiple sectors uh across uh countries around the world. And in many cases that's one of the complexities there, right? Because these are global organizations. There's different data governance rules. There's different legacy enterprise journeys that they've gone through. Almost every one of these enterprises is sitting on, you know, a few hundred million dollars of tech debt which you have to kind of work your way through before you can actually deploy any one of these AI uh, you know, ideas. How are how are some of these like uh management teams that you work with uh approaching messaging uh AI internally? Because I feel like the broad fear in the economy right now is related to job loss. And so one of the reasons you'd bring in an external partner is so that they can take just kind of a pragmatic view at the business and really understand where you can get leverage out of AI. Whereas uh if you're running a thousand person organization in a company, there's maybe some more tension around implementing AI, you know, you'll probably lose your job as an executive if you don't implement AI well. But if you uh but uh depending on how you implement it, there's going to be some job loss and you're and people are looking out for their team and maybe they like having a bigger headcount, etc. So, how um yeah, how how is that even uh being messaged at the moment? I look I think the conversation is definitely um across organizations right I mean people in most enterprises are very tuned to the fact that this is a moment much like the internet was and we saw this in the early days of the internet I mentioned some of the things we did back then right um a moment of re-imagination of business in a very fundamental way this is not one of those incremental you know progressive overload type moments this is a significant transformation so I don't think anybody body in your organization sitting there basically saying, "Hey, I think nothing in my world is going to change." I think people expect the change. I think what most folks aren't clear about is what is the journey from here where we have existing systems, existing processes, existing ways of working that the business is running on and you know making sure that you can fundamentally rebuild this plane while you still have targets and commitments and shareholders. uh and you recognize that if you don't do it soon enough to the point you're making, you're going to start to give up margin or you're going to start to give up speed or you're going to start to give up market share. Almost every conversation we're in is about how can I actually drive more growth? How can I actually create a better experience for my customers and this isn't coming just from leaders in the organization. This is coming from people across the organization. What we are seeing though is that at the level of enabling these people to do um work in this new context, organizations are very much separated by execution. So you know so many clients will talk about you know an AI strategy right but do you have a data strategy? Are all your data sets connected? Or actually, are you doing AI pilots on sales data on this ERP system over here and doing, you know, uh, you know, marketing pilots on that ERP system over there, which will eventually evolve to agents on this ERP system for sales, not talking to agents to that ERP system on marketing. And those agents not talking to each other across sales and marketing resembles a traditional baton passing organization of today, right? So you're mimicking a lot of what you do

larger enterprises. How it felt like uh last year was the year uh of uh was all about just like coding, right? Uh every single founder that came on our show would talk about the leverage that they're they're getting. We're getting that leverage internally. We've uh built a bunch of software to run the show that we never would have built historically because it just would have been too time inensive. But now we can have one great engineer Tyler on our team who can uh build an entire product end to end.

Uh please don't try to poach him because he said that. But uh but yeah, how how is uh how have like how have coding agents and uh products been adopted in some of these companies? Uh is there any type of restrictions? Part of the reason that it got adopted so quickly is it was somewhat permissionless. Engineers could just sign up for a product

uh and go bottoms up. But I don't even know how how adoption would work at someone like a McDonald's or someone like a Mercedes or someone at a bigger bank where I don't know that an engineer can necessarily just say like I'm going to start using this uh product.

No. And and No. And you and you can't, right? Um you can't because um the fact of the matter is it's not just Tyler or one engineer. It's one engineer with another engineer with a third engineer with a fourth engineer where a lot of them share context around a bigger application. Whereas if you start using one tool chain and the next person starts using the next tool chain and the third which is by the way what happened in the early days right which is why you got those pilots up and running really fast and people were starting to see real leverage but then very quickly that started to devolve to basically but we're not seeing the net output actually move the needle across big things. So I'll give you an example of a you know we um we did a piece of work for a large healthcare company which was on a 10-year modernization journey migrating millions.

Wow. Thinking in decades. Thinking in decades. People say you should think in decades. These people are actually doing it. I love it.

It's amazing.

Yeah. And it's crazy, right? It's millions of lines of cobalt code written in the 60s and 70s by dudes that are not not only in the company, they're not probably on the planet, right? and and and and you have individual developers going line by line by line trying to interpret what the intent behind the code was. But more importantly, because this is HIPPA compliant data and has serious healthcare outcomes tied to it, um you have to be really thoughtful about what of this code you need to retain, what logic it holds, uh and how much of this you can actually uh you know um do away with, right? That journey u going from like a 10 years to a few years, 2 years or 3 years requires you to be able to have all of the current business context of the organization. It requires you to have a future state that you understand really well and you need to have a group of people working on a very similar tool chain so that the code that you are ingesting then gets created into a spec that a bunch of humans can look at and say is this the spec that we actually want and then that spec can actually get migrated into new code. So we built this platform called Slingshot which is our um software modernization platform and how it differs from a lot of the consumer uh grade you know platforms that have got crazy valuations right now is they were all built for uh individual users writing apps like Tyler for small organizations or for themselves. But the minute you start to deploy them into large enterprises, what you find is that enterprise context graph that is missing or permissions like what part of the app are you allowed to see versus me or what is my relationship to you on a dev team, right? Um how do you actually start to manage authentication permissions? This is the kind of stuff uh that really makes those pilots go ary in the context of these big scale changes. um a lot of work.

What what can we expect uh in uh AI and advertising uh this year specifically? I feel like we went from large companies making an ad and saying we're going to serve this to a million people or 10 million people or 100 million people, right? A Super Bowl ad is is an example of that. And then as you have, you know, platforms like Meta really really really specific targeting. You can basically make an ad and serve it to like a tiny group of people. Very, very specific. You can imagine a world in the future where an advertiser even as big as McDonald's would effectively make an advertisement that's meant for like one individual person just because you could generate it on the fly.

Jordy, you're going to love this.

This big for you.

Yeah. Effectively. So, uh, what what are you expecting on that front? because the models have gotten a lot better and and it does feel like companies should generate maybe at least 50 times as much ad creative this year than last year.

Yeah,

I I think the output's going up. The question is whether the quality and the targeting are going up simultaneously, right? Because Meta sees incredible stuff in their wall garden. Google sees incredible stuff in their wall garden. But how are companies leveraging their first party data to understand that you're Jordy, this is your context, this is the car you drive, or this is the burger you eat or this is the hotel room you like to stay in.

I tell them everything about myself. I want them to target me as specifically as they possibly can. Personalized ad.

You'll be you'll be so surprised though, right? I'm sure you do that all the time. But how often do people actually use what you give them about yourself to actually give you something back that's relevant? The the the hard part is not getting consumers to give data. You know, there's been a lot of conversation about data privacy. And of course, I think consumers are all about like, if you give me value back, I'm going to give you my data because I get something back in return. So, if I if I'm watching Netflix and you know what I like to watch, you're going to show me something back that I care about. Great. I'll give you my data. But most companies can't do that. They can't actually serve up what you want. I mean I'll I'll you know we work with a very large hotel brand Marriott globally and we uh we're basically working with them to basically say how do we capture intent in the context of what people are searching for. So when you actually go to their villas platform, you know, historically what you might say is, "Hey," or a hotel platform, you might say, "These are the dates I've got free. This is the city that I'm searching for and this is the kind of product that I'm looking for." Right? And now you flip that around to say, how can you ingest all of the intent and context that they're willing to give you? So give them a white box and all of a sudden you say, "Hey, I'm I've got young kids. I've got a pet. Uh, one of my children is autistic and really sensitive to noise." blah blah blah goes up people who have the ability to leverage firstparty data combine it with second and third party data from a targeting perspective and then building production tool chains that allow them to use that to really personalize content is really powerful like to give you an example large pharmaceutical company doing a vaccine launch in 115 countries. How this works normally is the marketeteers would have to come up with a bunch of concepts then talk to the lawyers to basically say what are the regs in each of these 115 countries about vaccines. And you'll find crazy stuff like in New Zealand you're not allowed to smile in the ad because a smile implies an outcome. So they want like straightfaced people because you don't actually know if the if the vaccine is actually going to work. So hard to So hard to advertise if people can't smile.

Yeah. So

just like you're going to if you if you take this jab, you're going to have a bad time.

Yeah. So So you've got this crazy process that goes for like 12 months where the marketeers and the lawyers are having a conversation about what can and can't be done content-wise and then the content gets tweaked and produced, right? And what we built using both the Argentic platform, we ingest all of the regs into our platform. We then basically look at the creative concepts um that were originally created by the team and we'll test them against the regs almost on a real-time basis and then regenerate to your point using whatever models uh are more efficient for the kind of content regenerate versions of that content appropriate to every one of these countries and then allow that work to be deployed at significant scale. More importantly, if we discover an issue once it's actually live in the context of that, our monitoring agents are monitoring the content to ensure the the reactions to them and then can actually start to do AB testing about what's actually working better. But all of this is in one synchronous orchestration in the same way that like these orchestrations might connect supply chain data to store inventory to advertising. So we're basically, you know, Black Friday this product's already gone off the shelf. the ad isn't still hitting the same product because those people in the store are only going to get more frustrated and you start to close the loop.

Awesome. Well, thank you so much. Thank you for coming on, breaking it down. What an exciting time to be in this particular business

to be modernizing advertising.

I mean, a decade will happen in years.

I love it. Nobody likes advertising or modernizing more than us and maybe you. So, rest of your day. We'll talk to you soon, Nigel.

Take care.

Great to hang.

Hi. Before we bring in our next guest, let me tell you about Reream. One live stream, 30 plus destinations. If you