What AI Can and Cannot Do for Business

Shapiro: What separates more successful applications of AI from less successful ones?

Buluswar: So much of the narrative in AI today is focused on infrastructure. And it’s amazing. The tools are phenomenal. But the challenge for institutions is being able to see a process that is manual, repetitive, error-prone, expensive, and consequential and rethink that with an AI-first principle to say, “What should that look like tomorrow?” That’s where this power of imagination—the power of human intelligence—is absolutely critical.

Yet when we think about the AI landscape today, the capability of the tools is far outpacing firms’ abilities to adapt and adopt. We don’t have enough leaders who bring that perfect blend of imagination, technical aptitude, and operating know-how to be able to lean into what Suzanne and her team have to offer. They can’t picture what operating problems they would tackle, how they would do it, and how they would measure the ROI, which should be 50- to 100-fold.

Early on, it’s understandable that you focus your efforts where you feel the most overt pain. We all need a starting point. When you think about expenses, that pain is much more visible than lost revenue or opportunity cost of revenue, which are more below the surface. And given that the cost of compute is not trivial, there’s much more pressure to get to the more concrete problems and use a more direct, a priori, one-to-one measurement of the financial impact of those investments.

That’ll morph over time.

Executives need to be able to reimagine how they would want to tackle an opportunity or an overt problem in ways that they have not done historically. However much we like talking about growth mindset, the reality is that the more successful people are, on average, the more likely they are to hew to a linear extrapolation of what got them to that point. But we don’t live in a world of linear extrapolation anymore.

El-Moursi: It’s partly the people, and it’s the courage. People think all AI is equal. Whatever I tell them, they’re going to believe it’s all more or less the same. Technically savvy founders can tell us, “Here’s how I wrote every code.” But we’re at a point where people don’t always know—or want to know—the intricacies. They want to see that it works.

The challenge for us as a startup is to educate the market on why our AI approach and our product are more successful—because during the hour that we’ve been sitting here, another three companies were born that will compete with Brighthive.

Market participants are trying to understand the change that will come and identify the most important things to focus on, anticipate problems they’ll face if they don’t get ahead of the curve, and then move with conviction to set a strategy and execute a plan. As they’re thinking about all of that, AI adoption is happening in a waterfall manner but also at the speed of light.

Shapiro: How do you think about measuring outcomes and ROI when the benchmark keeps changing?

Buluswar: I go back to the three-layer cake of how we think about the power of AI. At the first layer, you’re not really measuring. It’s more of a soft feel of how comfortable you and your colleagues are with all things AI. At the second level, you don’t measure it on the margin but you might measure it in the collective. You can actually see the activities. For instance, you might track what employees do when they’re logging in, what activities they’re engaged in, and how those activities are getting more efficient in the collective. Even if it means they’re just working fewer hours, that’s a win all around.

Then there’s that third one, the big rocks. At its core, AI is about rearchitecting the what and the how of decision-making with the goal of achieving cheaper, faster, better decisions. This core thesis will reinvent people’s roles and their sense of professional identity—and that change is not trivial. The results will be externally visible.

When you’re making cheaper, faster, better decisions, you measure the outcomes in terms of lower costs, fewer errors that can be connected to financials, or some sort of upside customer experience or revenue components.

In the case of an AI-enabled process for error detection, banks experience significant productivity improvements. At the same time, they have the ability to look back and see all the mistakes that were made historically, what would be different in the brave new world, and what financial and regulatory consequences would be mitigated.

I think of defining success through the lens of the CEO, CFO, and head of audit: being clear around how you measure it, being confident across functions that it is material, and knowing that it is attributable.

In the absence of attribution, we’re all spinning stories and creating a narrative for ourselves that’s a feel-good, high-five moment with no real discipline around recognizing that on the other side for all of these tech investments, there’s a material cost of GPUs. And we’re not tackling a fundamental question: How do we take the 10 most critical processes in our respective firms and with confidence build a disciplined blueprint for how we’re going to redesign them, front to back, in ways that we couldn’t have imagined three years ago?

Shapiro: In academic research, it’s really easy to measure how much faster I can write a given paper using AI. It’s harder to think about the return from being able to write a much bigger paper. I’m solving a problem that I was never able to think about solving before.

Murli, you talked about detecting errors at a much faster rate. What about jumping up two levels to redesign the product or reimagine the product so there are no errors in the first place?

Buluswar: I love that question. Now when these errors happen, we can go to the root cause, and we can predict just when they are about to happen. Then we can ultimately tie that to our product design and our terms and conditions and the confidence that we have with our customer.

Citi didn’t go there because it was still solving the more fundamental building blocks, but there’s a very natural progression to those next two levels of sophistication.

El-Moursi: Right, because you can save the prompts. In prompt engineering, you can see the prompts I used, and I can see the prompts you used. That’s their reusability. The folk knowledge in the business is in the prompts. They’re the new English. When the product improves because of an idea that was tested, that becomes a valuable prompt. When you save it, everyone in the business gets to use it. The compounding effect is huge, simply because you have a platform that allows you to do that. Of course, you don’t only jump to the second and third levels; you could be in infinity if you change those behaviors.

Buluswar: For Booth faculty, the imperative is in equal measure educating students on how the world is evolving and teaching them how to hone the skills to be deeper thinkers, to have next-level curiosity, to challenge legacy assumptions, and to recognize that you have to think about processes and problems fundamentally differently than you did historically.

El-Moursi: Last summer, I approached the Harry L. Davis Center for Leadership and said, “Let’s have some interns come to Brighthive for the summer.” This was an experiment to see how much these current-day students, who are much younger than I, know about using tools. We ran this as a study, collected data, and hope to publish it.

We ended up with four students from Booth and one from Northwestern. Sure enough, in the interview process, I discovered that most of them have pretty outdated knowledge before they graduate.

Are they smart? 100 percent. They’re coming from many different countries. They’re the McKinsey, Boston Consulting Group crowd. Yet they came in not knowing simple things like a go-to-market strategy, let alone the new world that is AI GTM. They came out 12 weeks later with a 12-tool tech stack. They did Unify. They did Conversion. They did Navattic. Each one of them owned a mini tech stack for GTM.

To me, it’s personal. I earned this degree in 2012, when the market was different from what students are going to be entering now. It comes down to betting on yourself and then jumping in—being able to say, “Because I did this with AI, I can do that.”

Shapiro: Here’s a question that came in from an audience member. Can you think of an example of a company’s application of automation today that seems risky or concerning? Any examples for which you’d recommend against using automation?

El-Moursi: Yes and no. I wouldn’t recommend against automation. I would say look at the governance. In the United States—and in the world—we are low on our governance posture. Technology leaders who come into our funnel tell us that one of the things they’re hoping to solve with Brighthive AI is for governance policies to run autonomously, so that we can prevent the damage that happens when governance is poor.

That’s why data sharing and data collaboration become difficult. Your data’s great, my data’s great, and together we’re powerful—but we don’t know if we can trust each other. Autonomy requires great governance. If you understand that as a business leader, you’re going to find yourself embracing the right AI because you’re going to be looking for the governance and the policies and platforms that help you drive that.

Buluswar: I’ll share a slightly irreverent view, if I may. We’re breaking new ground, and when we break new ground, we are going to make mistakes. We live in such a dynamic world that it is virtually impossible to predict the hallucinations that a large language model will have.

For me, the question isn’t whether we’ve got a perfect system. The question is whether we’ve got a healthy, dynamic system that is, a priori, trying to at least put some parameters around what could go wrong and manage for that, and then learn and adapt. In the absence of that, you will never get to a level of comfort where you’d say, “Oh yes, this is going to be as close to perfection as possible.” It won’t, and that’s OK. You just have to be able, to the best of your abilities and at any point in time, to look around the bend and imagine what the consequences could be.

Murli Buluswar, MBA ’01, is head of decision sciences at Optum and former head of US consumer analytics at Citi. Suzanne El-Moursi, MBA ’12, is cofounder and CEO of Brighthive, an agentic AI data management platform. Bradley Shapiro, is Chicago Board of Trade Professor of Marketing and True North Faculty Scholar.