Podcast with McKinsey's Ivan Ostojic

My guest today is Ivan Ostojic, Partner at McKinsey & Company. Ivan and I talk about what success looks like - and how to avoid failure - in quantum computing proof of concept projects. We also discussed how we know whether 1000 qubits will be a quantum computing panacea, and much more.

Listen to additional episodes by selecting 'podcasts' on our Insights page

THE FULL TRANSCRIPT IS BELOW

Yuval Boger (Classiq): Hello, Ivan, and thanks for joining me today.

Ivan Ostojic (McKinsey): Hello, Yuval, it's pleasure to be with you.

Yuval: So who are you and what do you do?

Ivan: Well, who am I? There are many facets to that question, but let's focus on the professional part. I'm partner in McKinsey and I actually hold a great passion for technology-led innovation. So, we have something within McKinsey that we call the global technology council. It's like a think tank where we gather people that are external to McKinsey and some that are internal. And we think about future of technologies and there is 70 people there. I lead that part operationally and we have some deep dive groups. There is an overall group that is looking at tech trends and how are they changing faces of the industries. And then there are some deep dives that they're looking at particular technologies and one is dealing with machine learning ops and the other one is looking at quantum computing.

I lead all the quantum activities within McKinsey within that emerging technology group. And then also, I lead our innovation practice in Europe, Middle East, and Africa, and as well, like large parts of business building practice. So that's a bit more than what you want to know, but actually it's interesting because these technologies will either find its way like quantum will either find its place, transforming the core business or by companies building new ventures. So that's why I mentioned it. And by training I have kind of a dual background. I work a lot with tech companies and they often talk about horizontal, vertical. So it's a little bit like me. I have a PhD in life science, so that's my industry vertical that I know the most, although I serve clients in other industries. But also horizontally, I actually have a master's in technology management and innovation from ETH in Zurich, so I combined those passions also in my work.

And quantum, I've been spending out more than three years looking at it. I think it's so disruptive and it holds great promise that we can solve some big problems. So I'm very passionate about it. And we have a large group within McKinsey, and when we started digging, there were kind of tens and tens of people who have a PhD in quantum. So we are, I got the group of friends that is working with me there that has very deep technical competencies.

Yuval: Excellent. And do you work primarily with European clients or global clients?

Ivan: I work globally. I've worked globally, but for quantum it’s concentrated in some parts of the far East, Europe, and then United States. And Israel as well, but that's kind of accounted as Europe, also, in our division.

Yuval: You mentioned technology-driven innovation. Do you see quantum as a standalone activity where companies are just engaged in quantum projects or do you see it as part of a bigger innovation strategy or growth strategy for a company?

Ivan: I think it is a great question. It depends sort of who you are, right? So if you're a tech company, it can be like a standalone product in, in sort of a business to business terms. So, like either manufacturers or hardware or cloud service providers or cloud service plus consulting and so forth, I think if you're an enterprise, it's rather going to fit somewhere within your business. So it could be like a business model transformation.

Let me just take you through kind of a semi made up example, but relatively realistic. If you're a lubricant company today that produce all kinds of lubricants today. I think with quantum, it holds a promise that it could kind of simulate the optimal mix. So you could create a platform company that just literally simulates that part of the chemical process.

And there is some lab automation that produce that and it goes into the value chain. So, it's still within the core business, it's a bit of a different business model, but it could be like a new venture. And then there are companies or industries where this will be deeply entrenched within their current business model and it will dramatically or less dramatically, but significantly improve. So take banking... we all read, where in this quantum field, around recent research with optimizing portfolios when that's core, what banks do. And if you can improve something few percentage point, it's a huge revenue geneation. So they will need to master that activity. Similar it's within pharmaceuticals or others, where this is kind of part of the core of what they do.

And then there will be some potential for improvement of operations. We also read about the use cases like scheduling or logistical optimization. So it's going to depend on which industry you are and where in the value generation chain will hit you.

And to summarize, for many enterprises, they'll go through their core business. For some, this would be either variants of their current businesses, but more kind of platform driven or actually genuine new value creation. Think about sustainability and carbon capture. The fact that we might be able to simulate some of the molecules. When we have the right quantum hardware, not today, that's genuine new businesses. So that's kind of a bit how to think about it. Now we would need to cross different industries versus where the value is being created and then decide what's the optimal path.

Yuval: A lot of companies today are experimenting with quantum. Sometimes it comes from the top. Sometimes it comes from engineers or scientists who are just interested in starting to do it in their off time. But what does success look like today with a quantum project? Is it just proving that you could do something with quantum that you did with classical? Is it truly a unique advantage? What does success look like in a quantum project in your eyes today?

Ivan: Yeah, so I mean, to be quite frank, I spent a lot of time on this. We want to kind of understand this technology as much ahead of the curve. And we try to map the use cases that companies were experimenting with and so forth. And so far, it was relatively difficult to find the use case where, genuinely quantum hardware - I'm not talking about quantum inspired - what we created can beat or outperform, any kind of classical algorithm. And Yuval, I follow your publication. You know a lot of your audience, error correction, et cetera. So simply hardware is not ready today to produce something that would have superiority in terms of business impact. Maybe there are a few use cases, but jury's still out, there, whether we benchmark... But I think just not to over-hype where we are.

So, given where we are and being realistic on the roadmap of the quantum development, I think the biggest success would be few. Number one, understanding whether quantum has a potential to be completely disruptive to your industry. Something like implicit industries where lots is driven by science. And now we can simulate things like in chemistry. Or, actually, it will be enhancement of what you do today. And then understanding, okay, what are the most promising use cases and getting your hands dirty. I mean, there was a good example of Airbus challenge and Volkswagen has a challenge. I'm pretty convinced, although I didn't speak directly with these companies that, they're not going into it expecting big dollars, but they do want to see where is the technology frontier; what can we learn; and, what can we actually, at least, on the research level, produce sufficiently realistic algorithms to prove that, this case will be feasible if the hardware works?

So, I think with those three, I would say those would be my kind of areas to answer your question directly. First understanding where in my value chain, it will hit or create opportunity, and how, number two, learning with these experiments, where is the frontier? And number three, maybe in the research area, kind of proving that some of these algorithms could at least, sort of forward looking theoretically help. I think that's what's realistic to expect given where the hardware is. And then some offshoot, I spoke to many companies, was that then they produced like quantum inspired algorithms that were doing useful thing and they were quite happy. And I think if you come in with the appropriate expectations, I think you can actually manage your business stakeholders and not over-hype the technology and actually drive the steady growth that we are witnessing now.

Yuval: So given that that's your definition of success, it sounds like failure would be just setting expectations that are too high, going to the CTO and saying, give us $200,000, and we're going to revolutionize the way we do TSP problems or anything like that. And then of course, because the hardware is not quite there yet, failing relative to these very high expectations. Is that correct?

Ivan: Yeah, I would say yes, over hyping and promising kind of unrealistic results within a year or, or two in a way. I think that would be really something that would be not successful. I think having a strategy though, would be very good because, some of the hardware players are essentially more or less confident, but there is a probability that they will have hardware earlier than later. And you know, if there's going to be a limited capacity, it's also a question of how do we almost get engaged now in the project to have like a first seat in the row? And for that you need a strategy. Do you need the first seat in the row or not? Because if you're going to be disruptive you rather are there early, if this is enhancement of your business, you have a little bit of a privilege to wait.

That kind of is important. I think the case you described is probably absolute disaster because you know how these corporates work, the metabolism, you go there, something promise they do something, they tick the box, it didn't work. And next time somebody is coming with a good idea, maybe technology is more ready, people will say, oh, we tried this, it didn't work. And that memory somehow, memory of failures in many corporations, I witnessed it, it pertains during time. So I think, I think that's something we should avoid. We should just responsibly position where the technology is, but also what's the promise and then kind of build a strategy. How do we experience it without setting expectations too high, in a sense of, we will get something here and now.

Yuval: On the hardware side, a good approximation to the strength of the hardware is the number of qubits, because people say, oh, you know, I'll have a hundred qubits next year and I'll have a thousand cubits thereafter, and at certain number of qubits, I can do error correction and I can do this, that or the other, but obviously no one can simulate what a thousand qubit computer is going to look like, or is going to be able to do because that's beyond the capabilities of classical computers. How do we know, how do you know, how do others know that a thousand qubit computer is really going to be breakthrough in terms of the algorithmic abilities and the business value that it delivers?

Ivan: So, I think that's a great question. So far, what I've kind of seen is actually quite a good research from either smaller companies or actually large tech giants, where they focus on a particular problem, let's say some chemistry simulation problem. And they really go in depth, mathematically and so forth. Explain why quantum computers could solve this and then give some parameters. I mean, there were even some very fundamental papers that are a hundred pages long and go into very much detail. I think there is no way to know until we put the whole thing together. But I think this theory and what we know from the theory and how the different hardware work and the equations that went into there could give us confidence. So it's all about de-risking. It's not about being a hundred percent certain could give us reasonable confidence that we can... The risk... That there is a probability that if hardware work as intended, meaning no faults and so forth, that the algorithm will work.

But I think that's the best we can promise. It's the levels of de-risking. It reminds me a little bit of the pharmaceutical industry. When you're putting a new drug at the end of phase one, there is like a 90% probability that it will not work. And then as you cross certain stages, the probabilities get updated and enhance our expectations and valuations and everything else. And this kind of, on a very abstract level, also reminds me, so at the moment we have a theory and it should kind of work. Then we do a little thing in a simulator. We get some, as you said, it's not easy to simulate thousands or more qubits, but okay, this result give us next level of confidence. And then there's the next level above this. So where are we on this de-risking path is sort of what gives us confidence that this could work, but it's not like 100% guaranteed, as you say. I mean, we won't know until we put everything together.

Yuval: When I speak with companies about what they worry about, what they would like to see, usually three things come up. One is stronger hardware. Okay, we need a thousand qubits or a hundred qubits or whatever number they need. The second is people. We need more trained, qualified people that understand quantum information science that can create algorithms and so on. And the third thing is the software development platforms. How do I create software for a thousand-qubit computer? I'm no longer going to be able to manually connect the gates to the qubits and so on. So I have a two-part question there. First, do you agree, maybe there is a fourth or fifth one that you think is missing? And the second part, do companies really worry about the software part or is it just our wishful thinking at Classiq that that's the case.

Ivan: I think when you say companies, I mean, there are really few companies that move at scale that they think really, let's say, quantum transformation is they would think, let's say artificial intelligence, but there are groups within the companies that are thinking about this, and we spoke about it. And so I think there is one more point that I would add, which is causing, let's say a lot of confusion or so, in many companies and that's more or less, I call it strategy, but it's much more than a high level of strategy. It's actually, which use cases are where I should start experimenting and so forth. Some get lost really there in a way. Where in the value chain this has opportunity, and so forth. And what we sometimes advise them is that you need something like, we call it in artificial intelligence, we call it business translators.

So you need people who understand quantum theory of quantum computing sufficiently enough to understand where is this applicable, but then at the same time, have sufficient understanding of business to know what business problems they are. And this part is a little bit difficult because, you would need to kind of order cases across your value chain. Let's take pharma with chemistries, imputing missing data based on expectations from the hardware and software. So I would add that part for the use case roadmap and so forth. Now, I completely agree with you. People... We don't have enough and the gaps will become probably even bigger. I would agree hardware with some nuance, right? Because I mean, in our tech council, we have representations from very different companies that have different approaches and so forth. There is constant debate. Can we do something within NISQ or not?

And some people say, forget about it. Nothing useful will come. The others say yeah, we will be able to error-correct with software and we will be able to produce something beneficial. And I don't want to be a judge of that. It's a debate as you're aware, but I think, whether companies want a thousand qubits or more, that's a question. I mean, I think they want the useful quantum computer where they can kind of deliver the business problem, just a bit of nuance there. And then I agree with you with software, because I think even in our predictions, and this happened few years ago, now... You see this kind of in practice, many of the business problems will be solved in some sort of a hybrid sequence.

So, I have, let's say, sort of optimization problem. I can solve 70% of equations on using high-performance computer. And then I need some part of quantum computer in order to get coefficients or something like... Having somebody who can understand what needs to be done, translate it into actually lower level of arbitration and so forth, and then translate it into software workflow that will produce that at scale. I think that's quite, actually, critical. Also we can learn from artificial intelligence. When you had a bunch of data scientists doing sort of, some AI experiments, everybody was impressed kind of, wow, cool, you can do this, but there was no business impact. And the reason why there was no business impact is because this was not productionized for people at the front lines, to use it at scale. So now you have all this wave of machine learning operations.

And I think we can learn from that. I think, I think we can kind of preload that period in quantum. Yes, we need experiments, but I think we will need software learning from this to be able to scale that at industrial grade. So for these two reasons, because I, my prediction that workflows will be complex and it'll kind of call different libraries and so forth. Do you need binding tissue of that which is software and... All I'm trying to say, it's not like, oh, let's do a little bit of data and do something in R. No, it's going to be more complex. And then the second is, ability to scale gets much more powerful with software. So that's why I think these three or four components. Use case roadmap. When do we do what, when, with appropriately managed expectations, people, hardware, useful hardware, whether it's stronger, but definitely useful. And then the third one is the software. I think those are probably the most critical elements.

Yuval: So you mentioned quantum ops or the equivalent of machine learning ops or AI ops. Do you worry that a quantum computer is on the cloud? You know, you need to have an SLA, you need to have good response time. You need to have reliability, or you're saying no, the Amazon and Googles, and IBMs of the world will figure it out because they've done it so many times in the classical world.

Ivan: Yeah. I think curious, curious, kind of... If I may say something a little bit provocative... So I think they will figure it out. But for me, this question on the tech side, what really is the business model? It's also not a hundred percent resolved, and it's the question is, who's the first to get to market. And here's what I mean: in certain areas, quantum computing will generate so much more disproportionate value, take something like sustainability or so, you're literally creating new markets by simulating, let's say, molecules that don't exist today.

The question is, do company want to sell capacity for that? Or they would want to participate in the value that's being created? And we going to have abundant capacity? Because if we have abundant capacity, it's pretty obvious it's going to go into a cloud. And many companies will, hook on the cloud, have their workflows. And I believe these companies will figure out how to do it because many of the applications, unless, kind of, real time optimization of logistics and supply chain. There was an accident that I need a response in a second... Actually, quantum would perfectly be more powerful than, than standard, but you know, for that, I need the response time that's very fast. Others, you can still live a bit with, if you need a couple of days to do a simulation and so forth. So that's one, but I'm just going back to my train of thought.

So if you have an abundant capacity, I think they will figure out if things will go with the cloud. If you don't have abundant capacity, if we are generating extremely high value in certain industry verticals, we might not even see many of these quantum computers, at least in the first era, let's say first five years or so, being exposed very broadly into a cloud, if you see what I'm trying to say. Or maybe there'll be a third option where you have a vertical and a horizontal kind of a hybrid strategy. So there'll be certain vertical solution for certain industry and then a cloud for everyone else. That could also be, but I'm not so worried that they won't solve...

I think the more thing is by the way, just, sorry, I'm a bit kind of speaking as I'm thinking, but there is one thing, though, that is important. So we analyze one of the quantum use cases and we realize, actually, there might've been an uplift in some optimization thing, but actually there was a lot of loss on the information transfer between... Because it was done in a hybrid sequence between a quantum and then a cloud and so forth. So what I'm more worried is this quantum information transfer. Because to exploit the full power, we'll need certain information to be exchanged in a certain way. And that's another technical bottleneck, let's call it like that. It's fast enough so that you reap the full benefits of quantum.

Yuval: As we get closer to the end of our discussion today, some companies look at quantum as, oh, we're going to be generating this completely new product or type of service or platform as you described that was not there before. And now they're say, oh, I'm spending X amount of dollars on high-performance computing. And maybe with quantum, I'll be able to spend 30% less and get similar results. If you were a betting person, which one do you think will be more prevalent, the new product or the cost savings side?

Ivan: So, given the cost saving as you described, I would bet for the new product. You see what I mean? So, the way you're describing is like less spend on infrastructure costs in a way. Because I'm spending so much on AI, now I can do it faster. I'm not sure if that's the consideration. I think the cost in a sense of, oh, I have this supply chain and now I can cut 20% more costs because I have a better scheduling algorithm. That might be for a lot of industries. And we'll publish now one report where we analyze very detailed use cases. That might be for some industries. I still kind of think there's a lot of opportunity, if you ask me, in this upside, because lots of applications are what are they called, small data, big compute. And I'm referring back to chemistry, molecular simulations, and so forth. And there, inevitably, you come more to really opening new areas of application. So if I'm a betting person I'm betting in number one.

Yuval: Excellent. So, Ivan, how can people get in touch with you to learn more about your work?

Ivan: I think the best is if people contact me via email. And you know, there is a lot of now reports coming from our side where we are trying to analyze... It's difficult to analyze something that's coming few years down the road, but for a number of years down the road... But we are trying to analyze impact on different industries and publications. And on many of these I'm coauthor, so that's the other way to find it. You know, McKinsey, a game plan for quantum computing, was our intro publication. It was high level, but it was just supposed to put the basics for executives. And to read that, that would be the other way. And my email address is ivan_ostojic@mckinsey.com.

Yuval: Very good. Ivan, thank you so much for joining me today.

Ivan: I hope it was interesting. Thank you, Yuval, for having me. I enjoyed the conversation with you.

‍