#322 Amanda Luther: The Widening AI Value Gap (Inside BCG's AI Research)

0:00We have about 60 % that are either laggards or emerging. They're not getting a lot of value out of AI or maybe not even trying to. Where do you see agents being applied? What part of the business? Surprised myself at how slow the adoption has been. And I'll just take an example that we're all familiar with is customer service. I mean, I'm still pressing one for this or two for that on my phone. What is truly new and agentic versus what is something that a company could have done two years ago and it's just defining as agentic because that's the the nice term du jour uh amanda luther i'm a senior partner with bcg i've been with the firm for 18 years now i lead our ai transformation practice globally and get to see a lot of what's going on across industries and as part of that role one of the fun things that i get to do is uh co-lead our study that we do every year on the impact of AI across somewhere around 1 ,000 to 1 ,500 companies every year.

1:03I've been leading that for the last four years. I mean, the study's been going for the last eight or nine. And so we've got really great longitudinal data on what it takes to be successful in AI and how that has changed over the course of time as well. And this year's survey then and the report that followed is about the widening value gap between early adopters and laggards in the AI space. Is that right? That's exactly right. And that's what we ended up titling the report this year because it was one of the most interesting findings as we got into the data. One of the hypotheses that we had coming in was that maybe this gap would be widening because you've started to create the base of capabilities to build from.

1:50And what we've seen is that's really borne out in the data. The companies that over time were investing to begin with got value from that. That value flowed through into the P &L. They reinvested part of that value into additional tech and AI investments. And now they're getting more value from doing that. And so they're really in this virtuous circle of value generation from AI. Yeah, I have some of the metrics here to talk about, but generally, the companies that are getting value, I mean, we'll talk about the metrics, but what areas are they getting value in and how significant is the value on the bottom line?

2:41It's pretty impressive values. Maybe I'll start with where are they getting it. Yeah. Where they're getting it is primarily from the core business functions. And that looks a little different by every sector and industry, but it's really around sales, marketing, procurement, supply chain. Your core business value drivers tends to be where about 70 % of the value is coming from. The other 30 % is coming from largely efficiency and some of the corporate functions, like finance, like HR, but 70 % really coming from those core functions that are part of the company's business. You might look at like a CPG and that oftentimes, what is a consumer packaged good company?

3:28What matters to them? And largely it's marketing and branding and research and development. And those are the big value drivers for, you know, for a CPG. For a retailer, it might look a little different, you know, for a retailer, it might be around your commerce function, like how are you dealing with agentic commerce and really thinking that through? It may be around marketing as well. And then, you know, store operations and labor being a big value driver and an increasing value driver for retailers. Are these companies generally large enterprises, small and medium enterprises? I mean, what's the range?

4:05Is there a concentration of value among the really big players, maybe because of scale? Yeah, that's a great question. And one of the things that we've done in the report is actually kind of put companies into different maturity levels in terms of how they're dealing with AI. And, you know, to give you kind of the overall sense, we have about 60 % that are either laggards or emerging. They're not getting a lot of value out of AI or maybe not even trying to. There's then a 35%, which we call scaling. And so those are companies that are starting to see value, starting to see value in pockets, but not fully at scale in the P &L.

4:46And then that gets you to like the 5 % that we call future built that really are seeing value at scale across multiple areas of their P &L. And you can start to see this in their EBIT margins and their revenues and their total shareholder return. Basically, any financial metric you look at, these companies are ahead. That 5 % is actually quite a mix of companies. it skews towards more digital or AI native, you know, companies that have been founded in the last 15 to 20 years. Yes, it skews in that direction, but it includes companies that are 100 plus years old. And it includes companies that are really reinventing themselves in the age of AI.

5:26And so you do see a mix of companies that are in there. Is there, are they generally large employers? I mean, you know, oftentimes, and, you know, I've spoken to people at BCG about this, the value is incremental at the employee level and doesn't necessarily accumulate or accrue across the enterprise. I mean, if an employee saves an hour, they're likely to take a longer lunch. Yeah, longer lunch break, more time, you know, gaming or whatever. as they do for fun. But the larger the enterprise, maybe the more those incremental gains add up. I mean, can you see any pattern? Yeah, it's a great question.

6:20So I think there's the larger the enterprise, maybe there's the more chance for some of that to accrue because you can add up an hour and hour and hour and it starts to add up. I think the bigger thing we see on large enterprises is actually is frankly the ability to put real dollars and scale and investment behind AI and recognizing that there are fixed costs around building an AI platform and building a team within the organization that's focused on AI. And so creating some scale around that, I think, is helpful for the biggest enterprises. There's also, I think, some really interesting things happening on the smaller side of the equation.

7:02And this is where you're seeing AI native startups get to$100 million in revenue with like 30 people. And so I think there is this dichotomy that you're starting to see of the largest enterprises that can really invest behind this, but also some really small enterprises that start AI native from the get-go and actually don't even have to create such sizable teams because they're being so thoughtful about the process from the beginning. Yeah, there's been a lot written in the newsletters that I get about people who, you know, are running a company on their own with 10 agents or something. Exactly.

7:42I'm always a little skeptical. I haven't really done a deep dive on any of those companies. Are those generally in new business areas that have opened up because of the digital transformation in the economy? Or are they, and this is something I've wondered, are they entering legacy industries and because they're so streamlined, because of their AI adoption, they're able to quickly penetrate the market that otherwise is dominated by legacy players? So far, I think it is largely a new industries that we are seeing it. So if you look at like, you know, the cursors or the lovables of the Mercores of the world, I think those are the ones that get cited as, oh, look, the, you know, dollar revenue for employees off the charts at these types of companies.

8:39They are largely AI native actually going after AI development, or low-code, no-code type solutions. And so I think that is largely new capabilities that the world is creating. Now, you say that, and also app development is not a new industry. And so I think it probably is disruptive to some of the incumbent players within that space starting to get maybe concerned about that. Yeah. And how do you track the impact of AI adoption down to the bottom line? because the bottom line is a pretty, you know, gross in terms of like large and unrefined measure. It's hard to pull apart what contributes to the bottom line.

9:30Totally agreed. And I can speak to how we do that in our client work, but I think more importantly maybe is how we have done that in the report that we just released and how we got to that widening AI value gap. So we actually looked at three different kind of forms of data to do this work. One is we asked companies maturity level across 41 different capability dimensions. And this is everything from, you know, what does your data governance look like? What does your tech platform look like? You know, what type of talent do you have? So 41 different capabilities that we have shown over the years matter for AI.

10:10That's one set of data. that is not tracked to the bottom line, but you can kind of start to tell AI capabilities. The second set of data is also asking these companies, what types of functions or value pools are you going after in AI? And what have you been able to achieve? And so we asked them for, depends on the industry, but anywhere from 10 to 20 different value drivers that they could be getting out of AI and what P &L impacted that have, whether it's on costs or whether it's on revenue, and then kind of flow that through the bottom line. That's second set of stuff. And then the third set of data is actually external data on their financials.

10:50So what's the reported EBIT margin? What's the reported revenue growth? What's the reported shareholder return? And what you're then able to do, which is exciting, and as we correlate all of that data, is to say, the companies that have the highest maturity scores on those 41 capabilities are also reporting the highest amount of impact from their AI programs. And we're seeing that correlated in the overall financial returns and the TSR and the EBIT and the revenue growth. And so you put all those pieces of the puzzle together, you're really able to start to get a picture of what's being driven by AI.

11:29Maybe last thing I'll say there, and then I'll get off my hobby horse on this. But the other question that I often get asked, which is an insightful question, is, is this correlation or causation? You know, are companies that are really successful? Maybe this is what you're about to ask because that's often what comes next. So are companies that just happen to be very successful, they're able to invest more in AI. And so it's actually going the other way. And this is where the longitudinal data really comes into play because you can start to tease out, you know, which is the chicken and which is the egg.

12:01I think the answer is that it's partially both of these things. Obviously, companies that have been historically successful were able to invest earlier. But then what we're seeing is those that did invest earlier versus those that didn't, even looking at the same success five years ago, you're starting to see that widening value gap because companies that are getting the value are able to take that and they're spending about 2x the amount in investment on AI as the laggard companies. And then that reinvests, they're getting the return from it. And again, you're back to that, you know, that nice virtuous cycle.

12:39Yeah, that's interesting. The, the, and the investment, they're investing, the leaders are investing as much as double what the laggards are investing. Is that right? Is there, are there any absolute numbers to go with that? Yeah, so it's, it's as much as double. and we actually look at it in a few different ways. So we looked at it at the, how much are you investing in IT overall? And then what portion of your tech investments are going towards AI? And what we're seeing is, as I said, about double of that investment for future built companies being in AI relative to others. What it turns out to be is the average global share of IT budget spent on ai is still only around five percent right so it's not a massive number that we're talking about um but it is a you know a growing and meaningful number yeah and um in the report you say that the the leaders are expecting as much as a 40 uh or i guess it's 40 greater cost reduction, not a 40 % cost reduction, but even 40 % greater cost reduction than a laggard is significant.

14:08Where is that? Is that in time savings? Yeah, where is that cost reduction coming from? Yeah, it's really coming from multiple places. So one of the big ones will be time savings and efficiency on the work that's getting done. You often also see it in reduced vendor costs, right? There's sometimes you've got something that you're outsourcing or paying somebody else to do, and all of a sudden you can either bring that back in-house, you can automate it, you kind of have an agentic system run against some of that. And so you're seeing, I think, both vendor costs as as well as some of the internal headcount costs reducing.

14:50I think you also start to see, for example, if you're a retailer putting AI capabilities against your vendor negotiations and your procurement, you can see some cost out coming out of the COGS line item, for example, as well from some of these technologies. Yeah, what about because this is something that's helping just consumers, the ability to find things that you're paying for that you don't use. Yeah, you definitely see that, you know, both in terms of, I mean, this is always, if you're a CTO, the bugaboo, I've got all these licenses, but I'm, you know, nobody's actually using. How do I find that more quickly and kind of shut that down in a more automated way?

15:39There's also really interesting, you know, go through all of the contracts that I have. And am I actually enforcing all of the terms of those contracts? Or, hey, actually, I was supposed to get a rebate from this vendor and I just never, like nobody ever followed up on it because the person that was leading that relationship left the company and we just forgot about it. I think we're seeing some like pretty big outcomes from even just getting in and looking at that contract structure. Yeah, in the report, it says about 17 % of the value that the leaders are getting is coming from agentic systems.

16:17But that's expected to grow to 30%, I think. Yeah, over the next three years. Yeah. Yeah. I'm really interested in the value that people are getting from agent systems. And I actually would like to go and sit with a team that's running agents to see what they're doing, how many agents they have running. because my experience, again, it's just from talking to people and playing around with agents myself, is that they're still pretty unreliable. Yeah. Yeah. So what are your thoughts on that? Yeah, it's a great question. And I will say this is one part of the report that I always struggle with a little bit too transparently, which is what even is an agent?

17:17I think everybody defines that a little bit differently and kind of has their, and so, you know, I would take those numbers with a grain of salt of, you know, what is truly new and agentic versus what is something that a company has been, could have done two years ago and is just defining as agentic because that's the nice term du jour. All of that said, as we see agentic outcomes really working, I think a few commonalities that we see and how that works. One, very rarely is this a set it and forget it, agent just runs. More often, it is still kind of a human in the loop in the process, making the final decision, but having an agent or a set of agents speed up some of the processes and take some of the toil out of the day-to-day, one.

18:14Two, I think we are seeing the best agentic workflow design actually starts from a zero-based process design and says, what should this look like? Let's let's work with the human to say what it looks like today. And then how do we actually redesign this to take the time out of it? And on this, I think the important piece, you know, there's this, the story that I always tell of using agentic marketing transformation where the company said, okay, we redesigned it. We took a 10 day process down to one day. And then we found that 10, nine, you know, 10 minus nine still equals 10. And basically what they found was they took the process down to one day, But the upstream process and the downstream process still took the exact same amount.

18:59I think this was, you know, the example here was they had a content review cycle that happened every 10 days. So even though you shortened the content generation to one day, review cycle still not for nine more days. And so it's and so like really thinking through that end to end process design, I think matters a lot, too. And then three, and I think this is something that companies and organizations are starting to get more of the hang of. But, you know, early days, people just said, like, put an agent on it, it's going to solve everything. And it's actually, you know, agents need context, just as much as humans do.

19:32They need, what am I optimizing for? You know, what's my objective function? They need access to organizational data, and more than data that's easy to find. There's actually a lot of like tacit and implicit knowledge that people within the organization have that you've got to feed agents with. And then they need the guardrails. They need the constraints. They need the, you know, what can you do and can't you do? And if you don't put that in place in a really thoughtful way, I think this is exactly to your point. You can see a lot of failure modes that are coming and frankly, a lot of disappointment with agents that, you know, companies are seeing if they haven't thought about all of these pieces.

20:08Yeah. Where do you see agents being applied? What part of the business? You mentioned marketing. Yeah, we've seen marketing being a big one. As you think about just as companies have gotten more and more complex on their marketing technology stack, like the orchestration around that is ripe for agents to come in and support. And so that you can have marketers be marketers, do what they're good at, you know, be the creatives and be really thoughtful about what they're trying to drive. So that's a big one. I think the customer contact centers has been one that overall, generative AI over the last couple of years, that's been a ripe function for disruption.

20:52But again, another one where you've got kind of multiple systems that maybe need to be coordinated. You're trying to pull data from multiple different places. There's different communications channels. And so having kind of an agent orchestration layer around that can be really impactful. And then I think some interesting things on corporate functions, you know, one that a lot of my clients are getting excited about is, you know, if you're a retailer, everything that you're doing is revolving around the results that happened over the weekend, right? You wake up Monday morning and it's like, okay, what do we actually go change this week?

21:24What do we need to do differently? What is marketing going to do? What are our stores going to do? What is merchandising going to do? and like everybody on a Monday in a retailer is spending all this time getting all the data together so that they can get to if you can actually replace some of that with an agent that over the weekend is prepping and getting the right reports in front of people actually then have a back and you know conversational layer on top of that where you can kind of go back and forth and ask questions I think there's some uh real fundamental value that be that can be created there and you know I think it's something that our organizations get really excited about as well.

21:58Yeah. And when you say contact, when you were talking about contact centers, is that customer support or is that sales? We've actually seen it in both. I think customer support, you know, certainly happening across industries. And then I think we're more and more seeing the future of sales with, you know, whether it's inside sales, you know, some of the targeting that you're doing, being able to do that more at scale with agents. Yeah. Who's building these agents? Are most of these companies building themselves? Is BCG building them for them? Or are they buying off the shelf agents? I mean, increasingly, they're these, just as there are quote unquote, model gardens, there are agent marketplaces, you know, people have built these standard agents to do tasks that everyone has to do.

22:54So it's really a hybrid approach. And we actually have some data in the study. I'm making sure I've got the right number here. Yes. So we found that only about 11 % of organizations are primarily building themselves. So there are some agents that organizations are building on their own. Typically when there's not a great external solution, it's a bespoke use case that's important for that organization. And it's incredibly important for their core competitive advantage that they're trying to build against. But what we're more seeing is actually this hybrid approach to your point, which is some combination of going and getting point solutions, the agents that are out there, some combination of working with your big tech partners, hyperscalers, and and partnering with them on on agent build and then i think the most advanced companies are creating these agentic platforms that are allowing kind of low code no code agentic build deeper within the organization and i think that's something that is going to continue to grow as we as we continue to monitor the space yeah and the investment that we were talking about where these leaders are investing twice as much is there a breakdown of what they're investing in Is it people?

24:18Is it systems? Yeah, it's really both. And, you know, I think if you look at the overall breakdown of the numbers, it's probably half and half. True tech and data cleansing platforms, you know, your licensing costs, all of that, and then half on the people side. And within the people side, you know, I think seeing a few different components to that. Some of that's hiring new talent, AI-specific, AI architect, AI engineering, AI centers of excellence, I think are cropping up in a lot of companies. But it's also investing in upskilling, reskilling, training. One of the stats that's in the data in this report is the AI leaders are investing six times as much in training and upskilling as the AI laggards.

25:18And that's probably not a big absolute number, by the way. I don't think it's millions and millions of dollars that are investing. But it is a meaningful show of putting your money where your mouth is and actually spending the time to upskill and reskill people. There's that. And then there's also a big component of this, which is the process redesign and the change management around that. And I think we're seeing that companies that invest in the people and the process design are much more likely to get to the impact than those that are just invested on the tech or the platforms. Yeah. So where do you see you?

25:55This is, as you said, longitudinal. I mean, you've been tracking companies for a while in this regard. Are the benefits among the early adopters short-lived and kind of one-off? And then do you have some sense of the long-term benefits is or would they continue to invest and evolve um and and then what what do laggards i mean it is it's a gap right it's a widening gap yeah does that mean that that there's uh sort of in the internet age the beginning of the internet age companies that that did not go online eventually disappeared because they just missed out on the new market. I mean, I do think that's what we're seeing in the data, which is the widening gap.

27:04And if you have, you know, whether it is a reduced cost base or increased revenues or a combination of the two, which is typically what we're seeing for the leaders, that gives you an advantage relative to the companies that don't have. And you can, you know, there's many ways strategically that you can think about what you do with that. You know, sometimes it means I'm going to reinvest some of this into additional innovation and AI and continue to get better and increase that gap. It could be that I invest some of that in pricing. And so I bring my price down for consumers I can take share within an industry.

27:38But I do think what you're seeing is companies that aren't investing here are going to have higher cost bases, lower revenue, and they slowly are, I think, going to wither away. I think the other piece to say, though, is it's not too late. We're only a few years into this ballgame, and there were a lot of and still are a lot of investments in AI that aren't paying off. And so there is a version of this, which is a lot of companies that I work with like to think of themselves as a fast follower, which is basically monitor what's working. So I don't actually have to experiment myself and prove myself what's working or not.

28:19I can see what others are doing, whether that's my competitors or whether it's in corollary industries. And then I take what's working and I adopt it. And so I'm much more focused in what I'm investing in and getting value from. And I think that can be a successful strategy. I think what you can't do is stay still, ignore this, put your head in the sand, pretend it's not happening. Because as we've said, the value gap is widening, you've got more to reinvest, all those good things. The other thing I would say is there is an organizational muscle around being able to change and being able to adopt new technologies and adopt new processes.

28:59Building that muscle, I think is incredibly important if you want to continue to survive and thrive in a world that changes faster and faster every day yeah and and as part of that whether it's early adopters or fast followers if you don't get involved it remains this kind of scary unknown scary nebulous yeah you have to get involved oh that's that's what they're talking about this is these are the problems this is you know that you begin to understand the ecosystem and how uh ai can be applied i mean i tell people that my personal life start paying attention because play with it try it like there's not there's nothing better than getting your hands dirty to have a view actually the other i'm trying to remember who i heard this from so much ai stuff happening.

29:56But I'm starting to hear also that on a recruiting front, it really matters. Like candidates will ask, what AI tooling are you using? Or how has AI made this process better? And if you're not, if you don't have a good answer to that, you're losing that candidate, which means they're going somewhere else. And, you know, that somewhere else is going to get even further ahead, because they've got an AI forward person who took the job because they're excited about using AI tools in these spaces. And so I think just everywhere you look, there's arguments for you got to get started. You can't just sit and wait on the sidelines.

30:27Yeah. Do you, well, before I get on to what you're advising companies, the laggards to do, what's the, where do you see the penetration going? So it's 5 % of companies, even their AI is, is there except for as you say the ai first companies it's it's a small part of their overall operations is is the and i'm surprised myself at how slow the the adoption has been and i'll just take an example that we we're all familiar with this customer service i mean i'm still uh you know pressing one for you know this or two for that on my phone or you know or or listening to some infuriating chatbot or typing uh with you know that's not answering my question and that's been solved i mean i i can think of half a dozen companies that provide these solutions that take care of that, but they're not being adopted across the economy.