00:00:02Mayowa Ayodele: There are only three main things you need to know. What are the decisions you're trying to make? What are the rules we must follow? And what is the quantity we are trying to optimize? Once those three things are defined in a mathematical way, that's it. And one project we worked on recently was one where we reduced the solve time from about 10 hours to five seconds. That's a very massive gain for the client.

00:00:32Murray Thom: Hello, and welcome back to Quantum Matters, where quantum computing gets real from D-Wave. I'm your host, Murray Thom. As we move past the hype and theoretical to explore practical real world applications of quantum computing today and where the biggest opportunities lie in the future. Let's open the box and see what's possible.

I'm sure you've had this experience. When we first learn about quantum computing, we don't know anything about it, what it is, what it looks like. Is it the size of a breadbox? Is it the size of a room? How is it used? What is it used for? And this is a challenge for effective decision making, not just for ourselves, but even for executive teams in organizations, businesses, and the public sector who want to understand the value that it can create. In this episode, we're going to be exploring what kinds of problems quantum computing is being used for today, as well as the typical customer journey towards quantum hybrid applications.

Now, there are a variety of resources available online for folks who want to do it themselves, but many of our customers ask us to accelerate that journey for them, and we do that through our professional services team who work hand in hand with organizations across their entire quantum journey, from problem discovery to proof of concept, and ultimately to in production quantum applications. That means helping teams identify the right problems, implement solutions, demonstrate results, and confidently deploy them in real operational environments. To help us explore that journey, joining me is Mayowa Ayodele, Team Lead Solutions Architect at D-Wave's professional services team. Mayowa, thanks for joining me today.

00:02:06Mayowa Ayodele: Thanks, Murray.

00:02:07Murray Thom: Now, Mayowa, I wanted to begin with your story because as I'm speaking with people about quantum computing, they're sort of surprised and delighted to find out that they don't need to know quantum mechanics. And that's actually an important part because you didn't come to D-Wave and quantum computing with a quantum physics background. Can you tell us a little bit about what you've done and the path that's brought you to this role?

00:02:25Mayowa Ayodele: I come from a computer science background. I have a master's in computing information engineering and a PhD in operations research. I don't have any experience in ... At least within the context of my academic degrees, I haven't learned anything related to quantum computing, quantum physics. In fact, my first experience of formulating problems within the context of quantum computing was maybe six years ago when I learned about qubits. So it's a steady upgrade from computer science degree, exposure to optimization problem to working as a data scientist and then a researcher formulating problems as qubits and gradually moving into real quantum applications here at D-Wave.

00:03:11Murray Thom: And in that role in professional services, what types of organizations are you typically working with on the customer side and what does your day-to-day look like?

00:03:21Mayowa Ayodele: Most of the problems I work on are optimization problems. I have one use case that is completely different from optimization of the simulation, but 90% of the use cases I work on would be large, complex optimization problems for different clients across different industries. 90% of the customer projects would be on our hybrid technology.

00:03:46Murray Thom: And why is there a drive to look for new solutions here? Why quantum hybrid and why now?

00:03:53Mayowa Ayodele: There's been a lot of research in classical computing and classical optimization. And I think initially the idea was quantum is just going to replace everything because it's so much faster. But more recently, people have found that we can mix together the best of both worlds and be able to solve problems more efficiently. So most times when we apply our quantum hybrid technology to customer problems, we are often able to outperform the existing solution once we've done our homework right.

00:04:28Murray Thom: It's interesting because in 2025, Wakefield did a study across executives and what they found was that 81% of executives reported that they felt like they were reaching the limits of what classical computing could do for their optimization problems. And I think the reason why that's happening is because when we, in today's modern world, run into challenges that are very difficult, very technically complex, we often try to approach them by basically getting more compute resources and that involves parallelizing problems and optimization is this space where just raw brute force parallelization doesn't work out super effectively. So one of the things that's unique about D-Wave's approach is that it involves four schools of thought and bringing them together to help bring high quality, reliable solutions to these problems. Can you speak a little bit about those schools of thought and the structure and D-Wave's sort of modern quantum hybrid solver technology?

00:05:20Mayowa Ayodele: Tensor programming, which is very well known in machine learning, is one of the school of thoughts. What's unique is how we've been able to combine all these four things. We have tensor programming, we have mixed integer linear programming, which is very common in many classical solvers. We also have the concept of being able to represent decision variables as computatorial variables like permutation. This is often used in many metroistics like genetic algorithms. And then combined that with our own quantum annealing puts us in a unique position where we're able to solve a lot of things much better than things that have just maybe one of those or two of those combination of things.

00:06:01Murray Thom: So to help folks understand, I mean, we're talking about tensor programming, mixed integer linear programming, quantum optimization, and then you were mentioning combination lists and sets. So can you give us an example of how to make use of that last category? What would that look like in a problem context?

00:06:19Mayowa Ayodele: Let's talk about the very common example of a traveling salesperson problem, which has relevance in routine problems or in delivery where you want to deliver things to many different locations. Rather than have a very complex representation of that decision variable, we could directly represent it in a more natural way where we're saying you're visiting A, B, C, D in that order. So we have a variable called a list variable, which is what people can easily relate with. And the solver simply tries to find different permutations of that list in order to create a solution to that problem, which is a lot more efficient than other ways of solving that problem.

00:07:03Murray Thom: Okay. That makes a lot of sense. I think that a lot of folks can take inspiration from the fact that your background is in optimization, but some of the folks I'm talking with are Python programmers with high school level mathematics. If someone is new to quantum hybrid applications, what are the types of things that you can learn to be able to do and what is that level of effort required?

00:07:22Mayowa Ayodele: The platform itself is very easy. It's not more difficult than making a call to any cloud system, which is define some mathematical variables, upload that to the cloud, get the solution, and that's it. In terms of people with experience, say they have mathematics, they know a bit of Python, it's not difficult to get there, really. A lot of people can be discouraged by the way you implement things in quantum gates and having to understand all of these logical gates. It's completely different here because there are only three main things you need to know. What are the decisions you're trying to make? What are the rules we must follow? And what is the quantity we are trying to optimize? Once those three things are defined in a mathematical way, that's it. So just being able to learn the concept of optimization, understand a bit about what happens or how quantum optimization works can be helpful in terms of thinking about the backend, but it's really easy to go from just understanding Python maths to being able to solve a simple problem.

00:08:33Murray Thom: In that answer, you touched on gate and gate quantum computing, so let me just contextualize that. In quantum computing, there are multiple models of quantum computing. There's the annealing model, which is part of D-Wave's technology in our hybrid algorithms for solving these optimization problems, and there's also the gate model. The difference between them is the way that they're using quantum effects to solve problems and also the types of applications that they're focused on. With gate programming, that's a technology that really we're looking to impact commercial technology in areas like molecular simulation, quantum chemistry, and three-dimensional fluid flow over vehicles. But as Mayowa was mentioning, the programming model for that really requires quantum physics expertise, whereas in the annealing model, thankfully, we don't have to necessarily have that expertise. So Mayowa, building on what you were saying in terms of the accessibility and in your confidence that people can learn to program this, as a team of developers gets some experience, if they're working with you and they're seeing you build a couple of problems, how much of the work can they actually do on their own if they see a second and a third application they want to build?

00:09:35Mayowa Ayodele: I can answer that with some examples from our previous client project. So when we work with most of our clients, we enroll them on a training course, which is just about two weeks, which gives them some introduction into our solvers. While we're working with them on the project, we're intentional about explaining how the solver works, how we're modeling the problem within the context of our solver. And we have indeed had many clients been in position after the projects where they're able to add additional constraints to their model, and some can even now apply our solver to new use cases. And most of our projects are just less than six months. So they've gone from knowing very little about our solvers to being experts themselves where they feel comfortable to be able to apply our solver to other use cases within their business.

00:10:31Murray Thom: I would say that it's probably not commonplace for folks who are watching this podcast, at least not as commonplace as it is for me, to meet with someone who has a PhD in operations research and optimization. So let's say someone gets a PhD in optimization. What is the special skills that you're able to bring as the team lead in the professional services team when you're working with customers? What is the capability that you can bring that's really helping those teams with their productivity and boosting their results?

00:10:57Mayowa Ayodele: One of the things that I learned during my PhD is the effect on formulations on the search space. I'll use the word search space. By search space, I mean the number of possible solutions to the problem. There are many different ways to formulate exactly the same problem. Understanding the effect of changing formulations in terms of how we then get results or how we make the landscape a lot more complicated for the solver, that is one skill I learned during my PhD that I'm actually using in my work today. Sometimes we have client problems that we've never seen before. I try a particular formulation. If I don't get the kind of results I'm expecting, I know why and I know what to change and I know how to make the formulation more suitable to take advantage of the strengths of our solvers.

00:11:49Murray Thom: Of the customer teams that you're working with at the outset of a project, how many of them, what proportion maybe, have experience working with quantum hybrid applications? And maybe you could even expand from that proportion to what are the kinds of teams that they're made up of?

00:12:05Mayowa Ayodele: Maybe 10 to 15% actually have experience in quantum hybrid applications. We have people who have learned how to use D-Wave solvers before. That's definitely a very small portion. Most people are learning about our solvers for the first time. Typically engage with a team that's made up of subject matter experts, someone who can explain the problem to us in great detail. Sometimes we have data scientists on the team, other times we have optimization experts, and sometimes we have software developers or even non-technical team, people who are comfortable with just using Excel to generate their solutions. So we have a wide range of clients that we work with. And at the end of most projects, they get to a point where they feel like they're comfortable enough to be able to solve something using our solver.

00:13:02Murray Thom: That point really resonates. The importance of having subject matter experts who understand the business problem and the way that it's getting solved so that we're not solving a theoretical model of the problem, but the way that the business is currently running it. Let's say if there are teams who are looking to D-Wave to do a lot of the heavy lifting, you've got the subject matter experts, the ability to describe the problem, what's the importance of the availability of data?

00:13:25Mayowa Ayodele: That's one of the most important things because without the data, we can't build the model in any useful way. Having clean, good quality data is very, very, very, very important because in the end, our solver sees a bunch of equations with numbers. The numbers assigned to it come from the data. If the data isn't accurate, we don't get good results. So it's one of the most important things that we ask our clients to provide good clean quality data as quickly as possible early on in the project.

00:14:04Murray Thom: Let's get to, I think, a very exciting question, a question that's definitely on a lot of folks' minds. How do you recognize a problem as being suitable for quantum computing technology and a quantum classical hybrid solution?

00:14:17Mayowa Ayodele: The markings of problems that are suitable for a quantum hybrid solver would be one, an optimization problem. If it's an optimization problem, that's already one checkbox ticked. Also, where the problem is complex and large enough. The reason I say complex and large is because if the problem is small, most solvers can solve that problem very efficiently. You probably don't need a quantum computer for that. But where you start to reach the limit in things you're struggling to then solve a larger problem, that would be one where we think would be a good fit. And also where we have very complex interactions between variables. So for instance, we are talking about a production scheduling problem, and the manufacturing of each product consists of many different steps and different steps require many different machines. So when we have large, complex problems and problems where we have discrete decisions. And what I mean is, for instance, we might have decisions like how much electricity we take from a generator, and that could be something like 24.5 real variables.

Whereas in other decisions where we want to choose a staff that is suitable for a particular shift, or we are looking at the best way to create an ordering of products to manufacture, all of those more discrete decisions are better suited for a hybrid solver than continuous variables. So three things I've said so far, complex, large, and problems that have discrete decisions. There are few other use cases that may not be so common as optimization problems. So when people ask us questions about AI and ML, we've also had pure quantum solutions focused on sampling some form of Restricted Boltzmann Machine, which has been quite useful in material simulation, drug discovery, etc.

00:16:30Murray Thom: When I'm trying to help maybe give an example of this, and let me sort of provide one here, I'll sometimes explain it in the context of energy grid optimization. And in a traditional energy grid, there's a central power plant, and then it's got transmission lines heading to the east and heading to the west. And what makes that problem easy for classical computing is that as the load increases, let's say going down the west grid, they just kind of have to turn a dial. They're ramping it up. That's that continuous variable that you were talking about. You're kind of continuously varying the power generation to meet the load. And also what makes that much easier is the fact that the decision of the power going down the west grid doesn't affect the power that's going down the east grid. And if we consider, let's say, where energy grids are going towards modern decentralized energy grids, you're in a different situation.

You now have power plants all throughout the network. People have solar panels that are generating power on sunny days. They've got maybe electric vehicles in their garages that can store energy. And now you're in this situation, kind of like you were talking about with discrete variables, you're throwing switches, right? Sometimes you're pushing power to a particular site, sometimes you're pulling power, and each of those decisions you make at one area of the network affect the sort of decisions you make elsewhere.

00:17:42Mayowa Ayodele: Yeah. That's a perfect example of how we can have very simple problems for classical, and you can just add a couple of more restrictions around it and it becomes more suited for our hybrid technology.

00:17:57Murray Thom: Now, you also mentioned something which I think is probably worth contextualizing a bit, which was the size of the problem. One of the points I'll raise with people is that all of these problems ... Let's say you start up a company and you're doing at-home grocery delivery drivers. You start with maybe two drivers and 20 deliveries that you need to make. At that point, the problem is very easy. But as the system grows, as you get up to maybe 250, 300 deliveries to be made and you've got more like nine or 13 or 17 delivery drivers, I think what surprises people is that even at that scale, that problem can be complex or it can go through a transition where it's relatively manageable and then you reach the threshold where now you're loading the trucks to the point where you have to consider their capacity in terms of the weight that they can carry and then you have to take that constraint into account. How do you help people understand, let's say in workforce scheduling and production ready, that that is large for a problem?

00:18:53Mayowa Ayodele: It's easy when we think about all the possible combinations of things that could be a potential solution. Even things that are unrealistic, where a driver is assigned to five different trucks at the same time, or when you've missed out a delivery, all of these very large possible set of solutions exist in the search space. So we need a clever way to be able to find a set of good solutions out of these very many solutions, and that's what we are good at. We extract solution. Not just one, but we could give you a few different options to choose from, and they would be ones that satisfy all the business rules, the ones that make sense in terms of minimizing the objective that are important to the stakeholders.

00:19:46Murray Thom: Yeah. And I mean, you and I have talked before. I mean, you made this point to me, which is like, if you don't have business constraints, then you can really take your time and then you've got the luxury of really exploring a lot of different options. But it's these constraints. It's the fact that you're getting information from your employees about when they're available to work, and then you have to create a schedule for next week. You want that information to be sort of live and current. The fact that if they're working a night shift, they can't work a morning shift the next day. So I think that those kinds of details people may overlook when they're first thinking about workforce scheduling, that really introduces these really complex relationships that can make these problems challenging.

00:20:25Mayowa Ayodele: Indeed. And some of our clients, particularly, they have solutions that work, but the problem is it takes too long. So a lot of these scheduling type problems could take hours on many commercial classical solvers. And one project we've worked on recently was one where we reduced the solve time from about 10 hours to five seconds. That's a very massive gain for the client, and the client was thoroughly happy with that. So in those instances where things are taking way too long, it's one of the situations where we should be thinking about quantum or quantum hybrid.

00:21:04Murray Thom: Yeah, that's remarkable because the thing is that if our ability to replan and reschedule is slower than the time it takes for the disruptions to happen, then that's really limiting the effectiveness. Being able to reduce the times that way, I mean, that's amazing.

One of the things you touched on was the combination of quantum and artificial intelligence, and you mentioned our Restricted Boltzmann Machine. And just to kind of allay any concerns for the audience listening in, I mean, the community of folks who work in machine learning, who are building machine learning models, that's actually a very familiar term, but would you mind maybe just providing a little bit more detail about what you mean by a Restricted Boltzmann Machine and what it means to draw solution samples from it?

00:21:43Mayowa Ayodele: To oversimplify it a bit, sometimes we have situations where we want to generate a lot of samples, say within the context of drug discovery. Not just random samples, but samples that have certain properties. Sampling from the QPU is very, very fast. And because we are taking advantage of things like quantum superposition, we are able to explore that such space quite efficiently. When we have those kind of use cases, we are able to do things more efficiently and we get higher quality solutions than using a classical solver for those kind of problems.

00:22:27Murray Thom: Yeah, that's phenomenal. I mean, I think that for a lot of folks, it's going to be really surprising and interesting the fact that some of the computer science models of the brain involve neurons that are either firing or not firing. There's those binary decisions where switches are getting thrown and those neurons have synaptic connections to other neurons so that if they're fired, they either might excite the other neuron or inhibit it and that is introducing those relationships and that's the sort of solution samples, the sort of state of the brain and real conditions that you're talking about, so it's so cool. Okay. So let me ask you another question that I think a lot of folks, they have on their mind, which is that you've worked on several projects, several very successful projects where a proof of concept has led to a really strong return on investment. Demonstration of business value, companies who have either taken applications into production or have plans to take those applications into production.

What are the lessons learned that you've seen for projects that are going really well? What's the sort of recipe for success?

00:23:25Mayowa Ayodele: We've touched on some of those points, but I'll bring them all together. One is it's all about collaboration where we're working together with our client and the goal is success. The client is ready to provide the resources that we need. And by resources, I mean a subject matter expert that is available and can answer questions so that we understand the full context of the problem. We have maybe data scientists, optimization expert or software engineer, someone that understands the data, understands the infrastructure within the business. Another factor which has been quite useful for our clients is we are intentional about our knowledge transfer to the client. They feel like they have control over their own solution. What if a constraint were to change? I want to be able to easily change that constraint on my own. And another factor we mentioned was the data. Good quality data, good collaboration with the client, good use case selection based on the factors we've mentioned before, and we are almost guaranteed success at that point.

00:24:42Murray Thom: If someone is thinking about data, this is a solution that really can come in. It's not about redefining your entire IT landscape, but you can really sort of build a point solution, bring a solution into a place and really work within that IT landscape. What are some of the database management solutions or enterprise resource planning solutions or ERP solutions that customers may have that these models will work positively and interact with on the customer side?

00:25:08Mayowa Ayodele: We have such a huge flexibility in this area where as long as there's some API that can produce the data that we need to be able to generate the solution, it can come from anywhere. We have clients that have very well advanced database solution. We have people that even have some spreadsheet somewhere. As long as the data is accessible in some form, it's very easy for us to then plug in our solution to the data source and then plug back the solution into whatever form they want to visualize the solution. And that's really easy. There's a lot of flexibility in that area.

00:25:50Murray Thom: So if someone has, let's say like SAP S/4HANA as an ERP and they've got, let's say like a production planning submodule in it, those types of systems are compatible. You can make API calls to those.

00:26:01Mayowa Ayodele: Yes. We can connect our solution such that we receive the data and return the solution back into their own infrastructure.

00:26:15Murray Thom: I want to turn our conversation a little bit towards talent and training. So when we're thinking about students, let's say, who are looking to approach this field, what lessons can be learned about the journey they might take or what advice you might give students thinking about entering into the quantum computing field and building quantum applications?

00:26:33Mayowa Ayodele: I can start answering by giving many examples. Even for people working in my team, we have people who come from very different backgrounds like electrical engineering, chemistry. The core choose the important or the learnings or some understanding of programming because we need to be able to implement the solution in some way. And at the moment, most of our solution platform relies on Python. So an understanding of Python programming language definitely helps and understanding of how to convert a problem into mathematical equations. Say in the very traditional knapsack problem where we're saying, are we selecting this item or not selecting this item? We want to be able to sum up the weight of all the items. We want to be able to say the sum must be less than some capacity. Just a basic understanding of how to translate a problem into mathematics. Once we get that where we are like, "I know what the decisions and the constraints and the objectives are," the rest is very easy, really.

00:27:47Murray Thom: And what kind of resources are available to folks? I mean, are there examples they can look at in open source?

00:27:52Mayowa Ayodele: Yeah. D-Wave has many materials online. On GitHub, we have D-Wave examples that people can look at. We have formal training available for people who prefer the more structured way of learning about our solvers. There are many, many different resources. Research papers. There's so many resources that can get people started.

00:28:14Murray Thom: Thinking about your path, what's led you to this current role, what advice would you have given yourself as you were starting out?

00:28:22Mayowa Ayodele: I think I would have said to myself that I did not need to panic so much about things because a lot of things turned out to be much easier than I expected. So coming from a background, operations research where I solved a lot of problems with biologically inspired methods to then going to physics-inspired method where I'm formulating problems as qubits, it wasn't intuitive, just solving problems and formulating it as qubits was a bit different for me. And I didn't see the sense in it because I thought it complicated things a lot. But if I were to advise myself