Video
This transcript was automatically generated.
Julie Hyman: A new study says that 90% of companies are increasing AI budgets without seeing the desired returns. Research from consulting firm Bain shows 50% of companies are seeing savings of only 0% to 10%, and they were actually expecting a lot more, about double that figure. Joining us is one of the authors behind that report, Bain & Company senior partner Michael Herrick.
Michael, I saw this report last week. I was like, we got to talk. Well, we did talk about it when it came out. And I wanted to talk to you specifically because it goes to what we've been discussing almost without cease for the last few months, which is this sort of return on investment. And part of that has to do with how much can you cut costs, which is really what you dug into with this report. So what surprised you the most about it?
Michael Heric: Well, thanks, Julie. I really appreciate you taking the time here. Well, I would say a few things. First of all, I don't think it's just about cost. I think when many companies are looking at AI, they're looking at benefits beyond productivity.
So for example, it could be like speed. Let's say you're forecasting revenue or something. You're a CFO. You want to get those results faster. Could be working capital improvement. It could be a better customer experience. So it could be a broad range, but clearly, productivity is part of it, and cost savings is one element of that. And you are right that returns typically are lower than what people expected.
Hyman: And is that just a matter of it being early, do you think?
Heric: Yeah.
Hyman: Why is that happening?
Heric: Well, I think some of it is, if you look at like past cycles, let's say, AI has been very, very quick. Adoption has been very, very rapid. But if you go back in history and let's say you look at the internet, for example. The internet was adopted pretty quickly, let's say, about 50% adoption within 10 years. But it's taken 20, 30 years, even now to really transform business models and reinvent retail, financial services. That's all ongoing, and it's sort of similar here.
AI's in the hands of a lot of different people, a lot of different companies, and it's like, what do I do with this new technology? How do I actually make it deliver returns? And that takes time because the issue is, for a lot of companies, what you're doing is you're sort of jumping out there, you're building these great AI pilots or use cases, and you're hoping that there's going to be some value there. But you're just layering it on top of existing work, and you're not really changing the business processes. And when that happens, you're just doing work, and you've got something sitting on the side like AI, but it's not necessarily fundamentally changing your business process. So then they look and they say, ‘The returns are not great. I'm not really seeing the cost savings,’ as an example.
Hyman: Is the idea then to go backwards, like to strip it away, to look at the core of the problem you're trying to solve and then build up from there, rather than that layering on top that you're talking about?
Heric: Exactly. It's like taking a look at your actual business processes themselves. So it's just like that internet example. It's great. I can go online, I can do different things. But until companies fundamentally changed the business model and how they work due to the internet, then basically, they're not getting the full benefits from it.
It's the same thing with AI as well. People are going out there. They're like trying lots of different things. But at the end of the day, the business process hasn't fundamentally changed. That's for some companies, Julie. There are other companies that are actually fundamentally changing their processes. I gave you the example of there are CFOs using traditional machine learning, generative, and even agentic AI that are getting returns and are seeing the benefits, but they're really addressing these business processes head on.
Hyman: And they're still the minority.
Heric: They're still the minority today, which is not surprising. I mean, a lot of the early adopters¬, and I think one thing that we noticed, for example, in our study is that it's not by accident that those companies that have had more experience in these prior technology adoption waves—so let's say, deterministic automation like RPA or before that with Cloud as an example—many of them are actually leading the charge here.
So if you figure that out in your organizations, you figured out how to harness those technologies in your business processes, and now you're going to do the same thing again with AI, and those tend to have a leg up.
Hyman: The other issue with this is the cost, of course.
Heric: Sure.
Hyman: Right. And some of those other technological waves you talked about, you reached a steady state of cost, or there was an initial investment and then like it got to—the line of sight here is very different. I mean, it just seems like it's just straight up in terms of the cost that's being spent. So then, the calculation about what you're getting out of it also becomes very different, right?
Heric: Totally. I mean, I think there's so much focus on tokens, token max, et cetera, and we know all of those examples that have been out there. But the reality is that's only one part of the cost bar, so to speak, here.
There are other portions like the governance around it, like all the work around data. So you really have to look at the total cost of ownership, not just the tokens. The tokens is only one piece of the puzzle. It's a big piece, but it's only one piece. And the reality is that there's a lot of AI use cases where you really don't get a return on investment. And then there's others where it's incredibly attractive and the returns are huge. And the problem is a lot of the examples of where you're not really getting ROI drowns out all of the good, if you will.
Hyman: And what's interesting is the companies are betting on savings from this to keep investing, but they're not necessarily getting that savings.
Heric: 100%. I mean, I think there is a lot of just optimism around—and this was in the study, for example—what percent of agents are truly autonomous. It's very, very low because that technology is still developing. It's not as if it's fully mature yet. And a lot of the business cases are about not taking into account the full total cost of ownership, one, and number two, assuming a world that doesn't necessarily exist today.
And so to your point, there's a lot of pressure to just do things. And so a lot of those business cases are artificially attractive, if you will. And then, with AI, in particular, if you haven't done well in, say, automation, you've got the pressure to do what you need to do around AI, and you don't necessarily consider what happened in the past.
Hyman: So, Michael, obviously, at Bain, your job is to sell—you want to tell people how to do this the right way. Setting that aside for a moment, is it worth it? Even if you're making the argument you can do this in the right way, is the cost going to have to come down? Is the demand for AI going to come down because it's being applied in this sort of scattershot approach? What does this tell us about the trajectory of the whole thing?
Heric: Well, one is, I would say, Julie, the stakes have never been higher. And like you said, the amount of spend on AI is a much, much bigger factor than, say, other cycles. So if you took, say, RPA and deterministic automation, people would have a shared service. They might spend a few hundred thousand dollars on some pilots and things around RPA. If it didn't work out, it's not going to crush the company.
Now, you're talking $10, $20, $30, $100 million spend around generative, as an example, that doesn't even get into what they're spending going forward around agentic and what they spent already on traditional machine learning. And so as a result, the stakes are much higher. So I guess I'm an optimist, which is I think companies are going to make this work. I think this technology fundamentally transforms business processes. But I think the key is in where we spend our time with clients is really, let's look at business processes. Let's look at them from scratch. Let's say, if we had to do this process from scratch today with AI embedded in it, what would that process look like?
People are using AI to, for example, automate or do work with AI on work that shouldn't even be done in the first place. This is low value work. It's not going to add any value. They should really focus on things that are really going to drive business performance, customer experience, those sorts of things. So really taking a look from scratch now with AI and how that kind of works. And, again, I think many previous cycles are a good example. It's like the PC didn't get adopted because people had training courses saying use a PC. This is wonderful. The reason it got adopted is, for example, spreadsheets, way more powerful than paper ledgers or email, much better than doing a memo in the mail.
Hyman: It was figuring out how to do this in the best way.
Heric: It's like these use cases pulled in the technology. And our view is like when you start transforming those business processes, the pull around AI is going to be even greater than it is today.
Hyman: Do you think the cost is going to have to come down?
Heric: I think the cost will come down with all of the innovation today. I mean, if you just look at all the progress from when ChatGPT came out in late 2022 to now, it's just phenomenal in terms of the level of progress. I don't have a crystal ball, but in six months, I do think what you can do today, for example, with AI agents will be dramatically improved six months from now. It'll be a completely different world. And I think for most of our clients or just most companies out there, it's hard to sit on the sidelines and not capture the value of that, especially when you see other cycles.
I mean, look at cloud computing, the companies that sat on the sidelines there. Look at deterministic automation and the ones that sat on the side there. Look at the ones who didn't fully embrace the internet. So I think you see cycle after cycle, if you didn't actually innovate with those new technologies, then you tried to be a follower, and you were a late follower, it's really tough to catch up.
Hyman: Yeah, but to your point, there's the right way to do it early and the wrong way to do it.
Heric: There's a right way to do it early, and there's the wrong way.
Hyman: It seems like it. All right, Michael, thank you so much for coming in.
Heric: Well, thank you, Julie. Really appreciate your time.
Your AI Budget Is Growing. Your Returns Aren't. Here's Why.
AI keeps underdelivering on its promise. The fix is organizational, not technological.