Why the C-Suite Needs to Know About Neuro-Symbolic AI

Ivan Pollard: Welcome to C-Suite Perspectives, a signature series by The Conference Board. I'm Ivan Pollard, leader of the Marketing & Communications Center at The Conference Board and the guest host of this podcast episode.

By admission, I'm also an AI nerd, and today, we're going to discuss ?neuro-symbolic AI: what it is, how it works, and what we can do with it. And I'm delighted to have Jeff Schumacher, growth platforms leader at EY-Parthenon, join me. Welcome, Jeff.

Jeff Schumacher: Thank you, my friend.

Ivan Pollard: Great to have you here. And we'll go straight into it. So for listeners who are new to this facet of AI, how would you define what neuro-symbolic AI is in layman's terms, and how does that advance what we can already do with generative?

Jeff Schumacher: The simple statement we go around with is, neuro-symbolic AI is a business growth engine that turns data into value, right? If you think about it in AI terms, if you take corporate value and you branch it into two simple MECE branches—mutually exclusive, collectively exhaustive—one branch is productivity. That's where generative AI lives and agentic. And then the other branch is growth, and that's where ?neuro-symbolic lives.

If I can drill that for you a little bit, it's the company's commercial model, how they make money. If I drill that a little bit, it's how they price something, how they forecast, how they promote, how they launch a new product, new service, how they buy something, divest something. That's where neuro-symbolic plays.

Ivan Pollard: OK. And so it's more of the generative AI produces things and predicts things. The neuro-symbolic helps us make decisions around those strategies.

Jeff Schumacher: Correct. Yeah. You might say generative AI learns, neuro-symbolic reasons.

Ivan Pollard: Ah, very good. So the power of learning with generative AI and neural networks. The power of reasoning or the ability to reason with symbolic AI. And those two things coming together sounds interesting.

So if I get it right, then, what you are saying is neuro-symbolic AI understands. It's not just guessing at what comes next. It actually reasons. So it's kind of got causation, not just correlation.

Can you explain that difference and how, if that's true, if it can see causation, how's that changing the way executives will view its use in developing strategies?

Jeff Schumacher: You're getting it exactly right. So most of growth, when you're thinking about growth you're thinking about you're going to correlate the growth. You're going to, if you're a consultant, you'll do a two by two. If you're an investor, you'll look at a comparable. If you're a corporate, you might do a survey, right? You're getting to correlations of growth. But it's not necessarily causal.

So, I live in Los Angeles, right? And so in the summer, consumption of ice cream goes up, but so do shark bites. They're correlated, but they're not causal. Now my 7-year-old daughter who's getting ready for second grade here, she's heard that story so many times. She's convinced that she goes down to the ocean with her mint chip ice cream. She's going to get eaten by a shark. So there is some 7-year-old logic to that.

But neuro-symbolic gives you those causal relationships. The neuro side is the unification of data to understand behavior. The symbolic side is the rules. So Ivan, it's almost like you're getting a historian and a mathematician coming together, right? That's what you're getting with neuro-symbolic.

And that causality for executives, and this is a business play, not an IT play. This is more of a lens than how you look at the market, but it gives CEOs certainty. And certainty is, gives you confidence. And then you can act on it. You can get behind it, you can invest behind it. And that's what we're finding with the executives that we're working with neuro-symbolic.

Ivan Pollard: That's interesting. I love that story, Jeff, about your 7-year-old daughter. Thank her for it cause our members will learn from her.

But essentially, what you're saying is generative AI might spot that correlation in the data and tell me that I should make carcharodon-flavored ice cream, and that would be a big hit. But what you're saying, the symbolic side then comes in and starts looking at the reasoning and whether or not that would actually work.

You just used a word that's very unusual in the world of AI, or has been for the last two years: certainty. So why does the symbolic side of neuro-symbolic bring certainty?

Jeff Schumacher: One of the main reasons in what we find—you know, I invented the platform that we use, that's built on neuro-symbolic, is Growth Protocol. That's the company I founded four years ago. It took us a couple of years to figure this out. If you look at the analysts, they'll tell you neuro-symbolic is two to five years away, so we're talking about the only enterprise application in there.

Generative AI. It's a learning platform, and it's natural language, right? So it gives you the answer when you ask the question. Now whether or not that answer is right is another question. And the reason why that is difficult, if you ask a gen AI platform a question, if you ask that same question five times, you'll get slightly five different answers. That causes uncertainty.

The other problem is you can't audit it. You don't know where it comes from. Neuro-symbolic is just the opposite. It is fully auditable. You can see where it's coming from. You can go all the way to audit, and if you ask it the question five times, the same question, you will get exactly the same answer five times.

Now, the other part of that is if it does not have the data, it will not compute it, it'll come up null. That aspect, it's very binary and where we're deploying it, we have it in, call it 40-plus situations now on the market and 50 more onboarding, if the data isn't there, and there's nothing to reason with, there's nothing to do. And that's an important point.

Ivan Pollard: So that's interesting. We talked about growth and growth protocols. So business goals, commercial strategies, work processes, those sorts of things. How is neuro-symbolic AI best suited to drive growth through the expression of strategy and process?

Jeff Schumacher: It's another interesting question. So if I put my consulting hat on, and we've been at EY now for a few months. In that world. I'll do a two by two for you. So where does ?neuro-symbolic best play? Where something is high complexity and high impact. So go on your X-axis complexity, go on your Y-axis impact, and that top, that's where you want neuro-symbolic. And it deals with complexity.

And I can give you an example. So we're working with a manufacturer, manufacturers construction equipment and such, right? And the power of neuro-symbolic is the ability to unify a lot of data. So our software unifies currently 110 million data sources. So where we applied it initially was to unify all the messy parts of the internet to give us a foundational aspect there.

So we unify all the financial information that's public. We unify all of the news that's global and public. We unify all the scientific, the research, global and public. We unify all the e-commerce information, global and public. We unify even the social. So that goes into what we call a workflow.

And in this equipment manufacturer, what they were looking at is they finance a hundred percent of all the products they sell. And they think about construction and mining and that type of thing. And in there, because they finance it all, they have all the financing information—they have the credit rating of the customer, they have how much the product is worth, how long the financial terms are.

And if you think about it, Ivan, if you finance something, what do you have to do? You have to insure it. So the question became, could we build a neuro-symbolic insurance platform? And the answer is yes, because what neuro-symbolic allows you to do is now say, OK, I can take the telemetrics data off the machine. I can take the financing data. I can take OSHA data because I can take—in neuro-symbolic, you can take both structured and unstructured—so I can take OSHA data in. And therefore, now as an insurer, I can understand what price, how long, under what conditions, what's the repair versus replace versus finance cost? What creates loss? What prevents loss?

All of those are reasoning questions that goes into this model and then allows them to underwrite better than what an insurance company could do. And so therefore, you create an entirely new revenue stream for this company. You create an entirely new margin profile for this company because it has a far significant better margin profile, and therefore you change the EBITDA profile of the company. Therefore, you change the multiple of the company, and therefore you change the market capitalization of the company.

And these are the power of neuro-symbolic. And the other advantages, Ivan, if you think about it, my background is computer science, MIS, right? So, most people think in years in technology. We're talking in weeks that this gets put together and then months in deployment. So the ability for executives and CEOs to truly affect their top line and bottom line in a near term basis is significant for them. And we're finding that to be quite exciting as we show this to CEOs that we bring it to.

Ivan Pollard: Now, if we were talking in the 1970s, we'd have had the same discussion about how to generate reliable information about insurance, for instance, for this machines operator. And we would have a huge group of people doing it. They would never be able to crunch 110 million data sets. They probably would still make good reasoning assumptions using the logic that they've got from all of their mathematical and probabilistic methods, but also the data that they've got.

What you're saying is this is doing better than any of they could ever do it, faster than they can do it. We could argue whether or not it's cheaper than they can do it. But essentially, we're going to get the brilliance of everybody put into this neuro-symbolic machine, and then when you ask it the right questions and build it the right way, it can almost answer anything that drives commercial growth. Is that correct or am I getting off piece?

Jeff Schumacher: I think it would drive anything that requires a reasoning or risk. But if you think about it, so this is another question, Ivan. So now let me go put my mathematician hat on. So in most executives that we see, most people think of growth, let's say it that way.

Most people think of growth as an arithmetic. It is the sum of the parts. Therefore, different types of growth are completely different. Like whether you're pricing something or you're entering a market or you're creating a new product. Those are very different arithmetics, very different sums, different parts that go into making that equation.

?Neuro-symbolic says no, they're all the same. They're all risk. And so pricing is a risk. Underwriting is a risk. Entering a new market is a risk. So it studies the motion, and it learns from the motion. And that motion, if you go back to your math classes, that's calculus. Calculus is the study of motion.

Therefore, at Growth Protocol and what EY and Growth Platforms that absorbs the software, we build one reasoning model, and we just configure to different questions that you're asking, and that becomes quite powerful in what you're doing. And then if you give me new product, new service, new market, new capability, new process, or just re-engineered process, if you give me all those words and you allow me to call them workflows, that's how we see it, as workflows. But then those workflows are always on. That insurance example that I talked about is always on. That workflow is always on.

So what does that mean, Ivan? It means it's always learning. So I don't have to, if one of the attributes change, it's going to learn from that. I don't have to go back and rerun a regression or redo the whole thing as you did back in the '70s, or we were talking about in the '70s, that you would have to get all the machine learning guys together and get it redone and redo the regression.

This is why pricing has to be redone when you do regression and you replan it on planes, on the factor charts. The point, the value of neuro-symbolic is that it learns. The value of neuro-symbolic is, it's always on. The value of neuro-symbolic is it gives you certainty. And the value of neuro-symbolic is it treats risk and reasoning questions all the same.

So you're building one model. What does that mean? You can build an entire growth operating system for a company. And for an executive, they don't need 19 different solutions. They can have one reasoning solution that becomes their growth OS.

Ivan Pollard: So always on, Jeff, always learning and always connected. Like you said, the idea that we had silos of different units managing those vectors of growth that will now disappear. So when you're talking about this with neuro-symbolic, and as you say, it's starting to come pretty fast. DeepMind have already got it working with AlphaGeometry and a few other things. You've got it working with Growth Protocol and EY-Parthenon. But who is responsible inside a commercial organization? Who's the person that's advocating for the implications and the implementation of neuro-symbolic AI?

Jeff Schumacher: It's the business that is responsible for it. And what we're finding, Ivan, is when we started with this back in April, when EY did the acquihire of the services team from Growth Protocol and I came over and we founded Growth Platforms and EY, we said, "Hey, we built a business plan. And I can tell you currently, every executive we've shown it to, we planned it to be like, hey, we'd show it to 10 executives, and three would take it. But currently every single executive we've ever shown it to is, take it. It's 100%."

Because what they're getting to Ivan is, it's almost like they understand that they're playing "Moneyball" here. And they're going to see value in the market others don't see. And they're going to say, others are doing this arithmetic of an equation when they should be looking at it as calculus. And it's all the same with just different configurations. And therefore, if you think about "Moneyball," in the movie, everybody's out there buying players. You're buying wins, and you're buying wins by getting runs, and you're buying runs by getting on base. So the arithmetic in baseball was on base percentage, or the calculus in baseball was on base percentage. Same goes here.

Ivan Pollard: Brilliant way of putting it. So you've been pitching to Brad Pitt and others. I like that. It's also evidence that you are right, that it doesn't matter what executive to you pitch it to in any commercial organization, whether they're the head of HR or the head of marketing, or the chief revenue officer or the chief strategy officer. It binds them all together. It goes there.

OK, so we're going to take a short break, and we'll be right back with more of my conversation with Jeff Schumacher. And we're going to be fascinated to explore where this is going next and what the implications are for business.

Welcome back to C-Suite Perspectives. I'm your host, Ivan Pollard, leader of The Conference Board's Marketing & Communications Center, and I'm joined by Jeff Schumacher, Growth Platforms leader at EY-Parthenon. So Jeff, can you share a few examples when ?neuro-symbolic AI has already shown tangible business results, whether it's new revenue streams or operational efficiency, or faster decision cycles? Give us a real example.

Jeff Schumacher: Yeah, Ivan, and I think there's multiple, so we could spend the entire call on different ones, but I'll give you another one. Remember, so we're talking about it's a reasoning model, and when neuro-symbolic looks at something, they don't care if you're forecasting, entering a new market, buying something, divesting something, pricing something, or creating a new product or service. So it's basically the commercial model.

But let's take one in pricing. Pricing is always an immediate impact, right? So you can get to it. And it's been around for eons. So most people, when they price, and again here we're talking about, we will go into the retail channel here. When you price, you run a regression, and you try to figure out where the elasticities are and what the right price is.

But pricing can get extremely complex in the reasoning of all the different attributes. And what happens is most of the time, you take a more simple version of that so that you can get to a price, and you know that there's a lot more out there. But you just don't have the resources, the time, or the energy to go after it.

So in this case, we looked at price in a channel. Again, this is in the HVAC space. So think about an air conditioning unit, right? So you got to stay cool. And so what's interesting about that is, so when you do price, you first do, you got to figure out, am I going for share? Am I going for margin? Am I going for growth or profitability? And what happens there is, by market, you might have a different answer. So now that regression tree grows, right? What you're doing grows.

And then you have to define market. Is market Pittsburgh? Is market Pennsylvania? Is market Eastern United States? Is the market the United States? Is market another country? So now you have that complexity. Then, OK, now you overlay competitor. And when you overlay competitor, what happens is it's rarely Coke versus Pepsi, right? The products don't line up. It's more like Coke versus Nestea or something like that. So you have just differences.

And so now you have to account for those differences between products. And then you say, "OK, now if I'm in Pittsburgh, I want ease of use in Minneapolis. I want feature functionality." So now I got to account for all of that. And then you add time of year. In the summer in New York, you're probably, if you're in a five story walkup, you're like, "Hey, speed of implementation is pretty important." Whereas in, say, Seattle, you might be more price sensitive. So now you have to do all of those reasoning -related questions.

And that's where regression, when you typically see it, people just leave all of that on the table. So when we were able to plug all of that into a ?neuro-symbolic system and basically build a pricing workflow in a $50 million EBIT category, we found $8 million in a matter of weeks. So that's the speed to which you can do.

And now, if one of those attributes change, like feature functionality versus ease of use, the model will learn and just adjust so you don't have to redo everything, and therefore it's always on. And so once you have that ontology for the company—what is their language and how they think about it, and then the knowledge graph of all the attributes and things that you're doing—once that's set up, you can just keep clipping through.

So for me, I built one of the largest incubators in the world previously,. So I've founded hundreds of companies. What happens there is, you hack a company or hack an idea and then you have a long build once you think you got the idea right. Here, once you have the ontology and knowledge graphs, the follow on is fast. You could keep adding and adding and adding. So when you say speed to market and certainty and expandability, these are the things that executives get pretty excited about.

Ivan Pollard: That's cool. Can it go the other way as well? It can go from Philadelphia to Pennsylvania and then beyond, but can it go from Pennsylvania to Philadelphia to Pollard?

Jeff Schumacher: Yes. It can go from Philadelphia, I don't know Philadelphia very well, but it can go from Philadelphia to Allentown to downtown store.

Ivan Pollard: Can it get to me, though? Can it get to the individual price that it knows I'm willing to pay?

Jeff Schumacher: Exactly. Yeah. That—

Ivan Pollard: That's cool.

Jeff Schumacher: Willingness to pay is a key factor, and a whole bunch of factors go into what is your willingness to pay.

Ivan Pollard: Yeah, that's super cool. In the middle of that, Jeff, as you were talking you talked about the model doesn't care. It's just doing the reasoning based on the data and all the inputs that are coming in at the moment. But we often hear about the guardrails, so sure, it's doing the reasoning with the symbolic bit, but it's pulling the data. Are there ethics still applied to neuro-symbolic in the same way that we've argued about them for generative AI.

Jeff Schumacher: What I can tell you, Ivan, is what we're finding from like in the regulatory environment, the auditability regulators really like. But the auditability also, you can show—we do a significant amount of work in insurance, because who needs auditability more than insurance, and who needs certainty and causalities more than insurance companies—is that we can show there's no bias in the data. And that is important, that you don't have the human biases that you otherwise might have. And you can prove that you don't have that.

So from an auditability side and a regulatory side, this I think is quite significant, versus what a gen AI and what, which, what I think you're hearing in the market is, the ability to prove how you got to that answer and prove there's no bias in there.

Ivan Pollard: Yeah, cause I like that. So even if there was a little bias in the data you're using, there's absolutely no bias in the logic. So the logic removes any bias in the data.

Jeff Schumacher: And even more powerful, if the regulatory body changes or they want more restriction, that is just a constraint that goes into the model, and then it's always on again.

Ivan Pollard: Start with the logic. As we're thinking about this, you touched on this about the different executives, that you've got 100% hit rates at the moment. It doesn't matter what executive they are, they're interested. But from a leadership perspective, what sort of organizational cultural shifts are needed to make ?neuro-symbolic more of their operation effectiveness at play? What shifts need to happen for the organization or the culture?

Jeff Schumacher: I think it's, neuro-symbolic is targeted at growth or a reasoning related question. But I can give you a simple example. So we were at a retailer, and we were showing them two categories that were performing. They were contributing equally the same in a revenue and margin profile. But one is, you were looking at the data and how it we were representing cause we were unifying. Remember the neuro side of the data getting unified.

One you could see was in a headwind and predicted to stay that way, so that merchant was doing all of these tactics to maintain his revenue and margin profile. The other was in a tailwind and so basically, high tide raising all boats. So the CFO was looking at the data, he goes, or looking at the output of the model, and he was saying, they're contributing the same revenue and margin profile, therefore we're paying them the same, but should we be paying them the same? And we're like, absolutely not. Because one is doing a lot more to get that. That revenue shouldn't be treated equal.

So Ivan, to your question, the implications of understanding that this becomes your growth OS is you're going to have operational implications. You're going to have frontline implications. You're going to have strategic implications. So where this is owned, almost always, it is championed by the CEO, and then different quarterbacks have different components of this. But it relies on the business adopting it and understanding that it's going to change the metrics that they have, and what gets measured, gets managed.

And then IT is the enablement of that function, but not the leader of that function. So that's I think a shift. A lot of AI, generative AI solutions tend to reside over in IT. ?Neuro-symbolic almost always resides over at the business.

Ivan Pollard: Yeah. Smart. Looking ahead, just to wrap this up. When you think about forward looking companies, what do you think they'll be doing with neuro-symbolic AI by, let's say, 2030? What might be the lasting strategic effect on their market cap and their commercialization business models?

Jeff Schumacher: I think it favors first movers. So those that adopt quickly will gain a significant share over the others. I think those that understand that it can go through their value chain and change the way that they operate so the incentive structures will change and such. And then even then, I see, IT is an incredibly important aspect for organizations, but it has been largely centralized. And that was done, I think, because of Y2K and hey , I have to get behind all of that and get a centralized look at my value chain.

Neuro-symbolic causalities, you want to get as close to the revenue as you can, and therefore, I think you're going to see IT begin to federate as neuro-symbolic becomes more ubiquitous in the market.

And if you ask me where I see it today versus 2030, neuro-symbolic is where gen AI was in 2019. That's where neuro-symbolic is today. So if you take that forward, that is just about the timing of your 2030 aspect, I think you're going to see IT federate. I think you're going to see it'll be in that word cloud, one of the biggest words in the market.

I think you'll see that it's going to bolt on to gen AI to help it improve the outcomes that gen AI has, which you're seeing some organizations do. But it will become fairly ubiquitous. And right now, I think it'll favor significantly—as we did talk about "Moneyball"—in 2002, Oakland A's won more games than anybody else in the American League. That's case in point of favoring first movers.

Ivan Pollard: Brilliant. Thank you so much, Jeff, for sharing all of this. If we had to summarize it, we've got the learning power of the neural networks matching to the reasoning ability of the symbolic knowledge. And that's going to change business, the way we do business, and even what business is.

We should probably just comment that it's also going to apply to things like autonomous robotics and scientific discovery, and this ability to bring reasoning together with learning. Always on, always effective, a little bit more auditable, and a lot less bias. But let's just leave everybody thinking about that.

This is coming, this is going to come fast. And those who get there first will be the ones who win in the long term. Jeff Schumacher, thank you very much. Keep going with what you are doing with evangelizing for the power of this next step in AI to help businesses do better. Thank you very much.

This has been your C-Suite Perspectives. I'm Ivan Pollard, and this series is brought to you by The Conference Board. Tune into the next one. Thank you.