AI-Driven Advances In Consumer Product Development

AI is transforming consumer goods at every level, including product development. Through the use of machine learning algorithms and data-driven insights, consumer goods companies are able to rapidly iterate and refine offerings, optimizing processes at every stage of the journey – from formulation right through to packaging. In the process of understanding AI-driven breakthroughs, CPGs can anticipate and even stay ahead of changing shopper preferences, reduce time-to-market, and deliver innovative, personalized products that resonate with consumers. In this edited webinar transcript, Scott Isaacson, manager of data science for Citrine Informatics, and Rob Kieschke, former senior director for AI and software research in the 3M Digital Science Community, touch on these themes with CGT editor-in-chief, Lisa Johnston. Over the course of the discussion, Isaacson, Kieschke, and Johnston paint a picture of what the future of automation within product innovation might look like, and how to best gain stakeholder buy-in and manage change as the pace of these innovations gains momentum. Lisa Johnston: Hello everyone. Welcome to “Accelerating Consumer Product Development with AI.” My name is Lisa Johnston, I’m the editor-in-chief of CGT, and I'm excited for you to join us today. As consumer goods organizations continue to navigate supply chain disruption, growing sustainability mandates, and global economic uncertainty, today's companies must confront these challenges while continuing to grow and engage customers. Innovating in mature product spaces means more than just new technology and processes. Today, we'll discuss how companies can accelerate product innovation with a mix of new technology, data, and artificial intelligence. As part of this, we'll share how to leverage technology for product innovation to better meet today's quickly changing consumer preferences. Joining us are Scott Isaacson, manager of data science for Citrine Informatics, and Rob Kieschke, former senior director for AI and software research in the 3M Digital Science Community. Gentlemen, welcome. May you each tell us a little bit about yourself, your role, and your company? Isaacson: Hi, I'm the manager of data science at Citrine Informatics. Citrine provides a software platform to chemicals and materials companies, including in the CPG industry, for accelerating product development. This involves taking people's data, creating machine learning models, and suggesting the best experiments that are most efficient for getting companies to their goals faster. We work across materials and chemicals companies, and most relevant to this discussion, the CPG industry, as well as food & bev, and agriculture companies. Kieschke: Hi, I’m Rob, I spent 32 years working at 3M — every position I held included developing new materials, everything I did was developing new stuff. I had started out as a material scientist and a very cutting edge aerospace application, and then moved into products that ended up in cellphones — you're very familiar with what those things look like. Then, I ended up looking at the way we did R&D. 3M is one of the most hallowed R&D companies on the planet, how did we actually do that and what were the best practices in an analog sense? At the end of my career, I finally ended up in a digital role where I started looking at not only what we could do with digital assets, but how we could reinvent the way we do things with digital assets. Again, for the last three years I was the chief product owner for materials informatics efforts inside the company. Johnston: Can you talk about how you came to know Scott and how you came to work with him? Kieschke: In that sort of span of time, we started to realize that there was a really different way to develop materials. Like any company, we were stuck in this paradigm where it takes 10 years to do anything. When we looked at why that was, it was the mode of operation of a researcher, usually an individual, looking at two parameter optimizations at a time: I'm going to do X and Y, then I'm going to introduce Z. Well, once you introduce Z, it messes everything up. You have to go back and re-optimize X and Y. Once we started looking at the end state, we could compress the whole time domain. The limitation becomes the human approach of a limited parametric vision. How can we start to get data-driven machines or data-driven algorithms to start thinking in a sense in this 10 dimensional high-dimensional space. We started to see that these technologies were becoming, if not mature, certainly available enough to start investing in. I proposed a significant initiative inside the company to start reinventing the way we develop materials. At that time, a cursory look around who else was working on this, Citrine's name popped up, and we started to look at how we could engage to train with data, data structures, AI, machine learning management, etc.. How do we start to generate early proof points to convince the stakeholders that we're on the right path? Certainly, working with Citrine was a major aid in doing so. Johnston: Scott, one of the technologies that we're going to talk about today, and the approaches, is integrating materials informatics. This is not something that we commonly discuss on CGT. Can you set the stage for us, provide a definition of what it is, and why it's relevant for this industry and our audience. Isaacson: I think at its core, materials informatics always starts with data. We're all familiar with data. There are things like experimental results — did things perform well, did they fail? To what degree? There's other data, including data about my materials. I know something in particular about the surfactants I use in this chemistry, or the naturally derived polymers that I use in that chemistry. Bringing that information to bear, even if we don't have experiments on them yet, is also useful. That's the data piece. Moving beyond data, think about making a model. This might be a machine learning model, maybe it's something simple, maybe it's just fitting a line. At the end of the day, the purpose of the model is making a prediction and part of the value of those predictions is speaking to what Rob said, being able to predict quickly, in many dimensions, at once. Not just X = Y, but X = Y, Z, W, and so on. Finally, with that model, we would like to not just be able to make predictions, but also understand what's the best that we can do. That is where the concept of inverse design comes from. Inverse design is the question, not "I have this recipe, how will it perform?" It's asking the question in reverse, "I want something to perform in this way, what recipe do I need?" Then, in order to answer that question, moving from just a model to a model with constraints — I only have an oven that goes up to 300°C or something — also a sense of optimization. Now I know how it's going to perform, but how do I find the optimal performance in this massive 100 dimensional design space? Those ingredients, data, model, and then optimization over an unknown space, outlines the problem pretty well. Johnston: What are some of the forces that are pushing consumer goods companies to have to make these types of decisions? Isaacson: This goes back to Rob's intro, when he spoke about shifting from optimizing for X, to X, Y and Z. At Citrine, my team has a bit of a joke that every problem is a multi-optimization problem even if you don't know it at first. It always is. You're not just optimizing for viscosity, never is that actually the case, even if you frame your problem that way as a simplification at the beginning. That's definitely one thing that's driving that. At the forefront of those today is sustainability. Consumers are interested in sustainable project products. There are regulations that are coming down causing that to be, so people want to adapt to this fast and don't want to be left behind. A materials informatics approach is one way to meet that paradigm shift in a faster, accelerated way. Johnston: I'm glad you brought up sustainability. We are going to talk about that in a little bit. For now, Rob, I'd like to learn more about some of the pain points that you faced before the partnership, and how this technology changed the ways that you were doing things. Kieschke: I love what Scott said, but I am way more bullish than he is. There's a temptation to think, “I'm just going to do the same things but a little bit better.” There's an opportunity when you look at this system, this way of doing things as a whole, you have a way of almost reinventing the way people come to solutions or solve problems. I'll share this story. In the early days of this program, we started to look at where to apply this, which materials could we apply to first? I went to talk to a pressure sensitive adhesive (PSA) developer — think of the adhesives on Scotch tape. Now, a little history lesson, PSAs were certainly developed by 3M in 1925. When Dick Drew eventually invented what became masking tape, it was the two-tone car painting process. Think about it: 1925 is pretty much 100 years ago. I talked to this PSA developer who had a certain problem. There was a certain set of attributes he was shooting for. When I asked how he does it, he said, "I take these six chemistries, and they vary — each one by 5% — and then I have 65 different formulations. Then, I walk through them and test every single one." I couldn’t help but wonder where the acknowledgement was of our 100 years of experience. Where is the function? Where's the algorithm that is at least guiding you towards knowing, for example, that one should only be at 23%, no more than 27%. I'm making up these numbers, but the researchers' approach is, "This is what we've done in the past, therefore we're going to keep doing it." Once we start developing for predicting algorithms, and particularly when you have the reverse problem, you're able to actually say, "We have this goal, go and focus on this formulation space." I'm not sure we'd ever get to a precise formulation, but this formulation space is massively accelerating and changing the way people solve problems. No one uses paper ledgers for accounting anymore, and there's a good reason for it. These tools that are available, have become ubiquitous and commonplace — we just accept them as tools — they have changed the way we can do calculations, the way we can do modeling, and caused us to be able to solve problems in vastly superior ways than previously. Johnston: Some of this is related to new technologies, but a lot of it is related to adjusting to new processes and ways of doing things. I'm going to touch on that, come back to you in a minute on that point. Scott, what other unique challenges are facing companies that are innovating in mature spaces? Beyond just your work with Rob. Isaacson: One challenge I see, particularly in mature spaces, is there is a lot of data that exists and because that exists, there can be a temptation to try to use all of it, even if it is of questionable quality — put it that way. One example is from a materials' company I was working with. They worked hard to pull together five years of development data on a particular product line. We met to take a look at it and see how we could integrate it, but it became obvious very quickly that they were missing a lot of the inputs and measurements. It’s not actual input data, so it turned out to be useless. This can happen in a number of ways, but the broader challenge is knowing what data to invest in, what's important, and what should be left alone. At the end of the day there's infinite more data existing in the future than the past, and we have to make a correct call on what to pull in and what to leave behind. Johnston: It's certainly a challenge that's not going to get easier as we go through time. Rob, I'm sure some of this rang true, but in dealing with your own process, can you talk about some of the changes that you had to make, and your teams had to make, as a result of going through this — whether that be the team, the company — in terms of the business processes that you had to change as a result. Kieschke: Everyone kind of knows this, but maybe not well enough: The data is super important. As Scott alluded to, having the so-called metadata coupled with the real data — those are fancy words, but what it means is, "I intended to do this. Here's the formulation and here's the outcome." It’s how those originally linked together. Think of it as being good scientific practice, but in actual practice, it's pretty loose. Developing a way of working is important. It can be done internally, which is what we tried to do, and is expensive. It taught us a lot, including where to look. But having a system that people will use easily and will naturally associate the metadata with the data is invaluable. I fully agree with Scott that there is a certain amount of data where it's not worth mining it anymore, particularly if it's written down, that's awful. Written data is not worth the bother. Ingesting from Excel is almost a necessary evil. It's not that I dislike Microsoft, but Excel is customized by the individual researcher too much to be mass ingested. That's an issue. You need to have a front-end that people use which leads to something hugely valuable, like an AI prediction. It has to start to build that connection between, my data causes me to get the answer more rapidly with more precision, and satisfying both the researcher and the executive that this can be done in a way that generates value in a relatively short period of time. Again, start to allude to challenges with stakeholders. There is a generational difference. Young researchers, certainly within the last 5-10 years, come into a company and expect you to have this digital tool set. If you don't, there's something wrong. Workers that have been involved, let's say longer than that, will have an old way of working and that's difficult to get away from because it worked in the past, so it will work in the future. Inviting those people to participate in this new way of working as a tool — not something that's going to replace them, but going to help them, accelerate them — is massively important. It’s necessary to have a proof point, a success story, to invite others in the company to start to participate from an opt-in mechanism, rather than a senior executive saying this is the way you'll do it. Having an executive onboard is helpful, but having a success story that this problem was solved in this way, come and find out how they did it, is a necessary and powerful step in changing the culture. This is at least partly a culture problem. Johnston: Getting key stakeholder buy-in at all levels is crucial, we know. What are some of the other ways that you were able to get buy-in from the younger generation? Having this technology is now becoming table stakes, but all the way up to the board level? Do you have any tips you can offer? Kieschke: First of all, we were wise in that we put together very cross-functional teams. This was not led by IT — we put data folks (data scientists, data engineers), as well as software coders, material scientists, and chemists on the same team, so that they all had to understand each other's perspective. The chemists on those teams became evangelists for this approach. They gave talks to the company about how great this was, "I solved this problem in a way that I couldn't normally have seen." Or we had one example of a researcher who openly admitted he was going down the wrong path, "I was convinced this additive would improve my adhesion, but this tool basically told me it wouldn't, and I saved five months." Generating that internal story was incredibly important to the researcher level, but also to the mid- to junior executive level. What started to sway the CEO and board-level was having the vision about how we could address one of the major pain points: Why are we so slow with getting to this endpoint? We traditionally hadn't thought of speed as being a competitive advantage. We were traditionally focused on product leadership — we make the best stuff available and charge accordingly. We began to realize that if we are second, third, or fourth place, particularly if we're responding to a market need that everyone can see. For example, let's say we have a big sustainability challenge and we end up being fourth or fifth place in terms of time sequence getting there, we have a very low share. Speed becomes another component of sustainable competitive advantage, which we have to embrace. Certainly, having the vision of this 10x acceleration, which is my vision of how we could sequentially do multiples of two to get to close to a 10x speed effort in the company. The rationale to do that got the board interested, it was a board-level presentation. The vision on the senior executive side was articulated with a certain set of actions. My argument was that I don't know whether it's a factor of 2 — it could be 1.7, it could be 1.5, it could be 3 — but the fact is this is a non-incremental change at every level. By committing and doing this before we start to lose ground to competition is critical. Johnston: Scott, can you build on that and add on, in your work with other companies, what else do you advise them in getting this buy-in? Isaacson: I agree with everything Rob said. It's important, though, to emphasize that it is difficult to achieve your goals by only dipping a toe in the water. The reason is, as we all know, research and development is unpredictable. We don't know if something's going to succeed or fail at the beginning, and counterfactuals are hard to come by. It's hard to say what would've happened had I done X instead of Y. There is a bit of a critical mass, as well. Rob alluded to having software people and chemists in the same room. There is a certain minimum investment needed to gather enough momentum to have this evangelism, this momentum build up, and start to precipitate an organizational change. Johnston: Rob, if you can, I'm going to press you a bit to share the benefits that you experienced. You've talked about the challenges and why you made this investment, about getting your stakeholder buy-in. Can you talk about the benefits that your teams experienced and whether this was across functions, or just within your core team? Kieschke: There were several products developed in the structured adhesive space. That's what we chose to start, it was the simplest thing we could think of doing. Again, we wanted that near-term proof point. We had one instance where we wasted maybe a year or so going down the wrong path. Having this approach taught us quickly, in an objective way, it wasn't personal, but it showed us which direction we should go in. This has now become part of that development process. Then, the next step is to take it into the next more complicated thing, which is pressure-sensitive adhesives. It’s a much bigger business. That's the number of parameters you start to introduce, the more complex the approach becomes. It involves tangible things, such as speed and being able to get to market and solve problems for customers in that space. There are some softer things. Since I was in the digital lab for the last six years, my colleagues from the materials and chemistry side were saying, "Oh you guys don't do real science," even though I came from that world. It brought an expansion mindset. Now, chemists and material scientists start thinking about digital as being a collaborator, a tool set, something we can develop together and can work together on. This is not the enemy; this is not some weird stuff. This is not some package I buy and use. This is now part of the way I do my work. That cultural mindset, as I mentioned a couple times, is super important. If a lot of the company is resisting, it's going to be a lot slower. Once people start to get on board and realize that everyone is pulling on the same rope, trying to solve the same problems with a different approach, it begins to flow. Again, the cross-functional team, the ubiquitous communication about success stories, challenges, and being open to communication made a major difference in the way we now start to solve problems. Johnston: It does. Scott, can build on that, maybe share some other perspective from what you saw in the relationship. Isaacson: One quick win that I see frequently is the active uniting of data for a product line, or something, under one umbrella. It eliminates a lot of wasted work. Researchers have told me that once we have a data set on-boarded onto our platform, that I’ll probably do 25% less experiments simply because I'm not repeating myself. That’s a big relief, right? There's no need to do that again. I go back to what Rob said about acknowledging — forget acknowledging 100 years ago — what about 18 months ago? We were working on a project that had both performance and cost components, and with this approach, when a computer is helping you evaluate the options alongside your scientific intuition, you can come up with results that are unintuitive and yet performant — that’s really powerful. For example, we had a project where we were reducing cost but had to maintain performance. We found that a formulation with one of the most expensive components was meeting our goals, but it was because that expensive component was able to be used in a relatively small amount, which met the performance criteria and yet ended up being less expensive. The engineer said, "I never would've thought of this, this is outside of my idea of what I was even considering." That's the power you can have by having this tool as a partner to help evaluate these possibilities alongside your expertise. Johnston: It could almost be a humbling moment for many. Rob, what's your reaction to that? Kieschke: There is something I hadn't mentioned so far and maybe it’s obvious to some. Again, we have a bunch of researchers attempting to solve a problem. Using this toolset, they're able to generate a model that predicts some outcome, and maybe even the reverse. But there is also a networking opportunity that happens then — in solving that problem, you can start to realize how many other people are asking for the same thing, or how many other people are accessing that same algorithm. That is massively powerful in a company of any significant scale, certainly if it's global, because you don't necessarily know who is trying to solve the same problem at the same time. If we take a sustainability effort, let's say we're trying to make a more sustainable film, maybe more bio-derived, there are probably individual teams of researchers across any company who are trying to do the same thing and not talking to each other either because they're geographically separated or are in different parts of the organization. Having that common algorithm feature set, essentially pipeline from concept to notional solution can bring people together trying to solve the same problem in the same company. I would confidently predict that in our company, half the time was probably treading over the same ground that someone else had already done. The ability to have a tool, or a framework, which allows collaboration or visibility to who is trying to solve the same problem across the organization, would give about a factor to acceleration. Part of the stakeholder pitch I was claiming was going from data to predictive model, it's an investment. Let's be honest, we have to try and do this, and it is hard to justify an ROI along the way unless you get to that predictive model, and it's something Scott said earlier: you are able to prove that you couldn't have done it the traditional way. It's really hard to do that, by the way. It's very hard to completely predict a very accurate ROI, but there are intermediate points of value along the way. In other words, by building towards that, you get pretty near-term value in terms of having a different approach, in terms of having a richer, cleaner data set that contains the things that didn't work. If you allow researchers to write down their experiments, they will only write down the ones that worked. They don't want to write down the ones that didn't work. You get a more holistic view of the actual stuff happening to build the model. Along the way, while building the model, you will generate significant value in the organization through things like ease of collaboration, having a much richer data set to actually develop material in the first place. Johnston: Did you run into any roadblocks that you can talk about as you were going along this process? How did you work to overcome them? Any anecdotes to share? Kieschke: I’ve touched on one of them — the more seasoned researchers who would think this was an instrument of the devil and declare, “We're not doing this." Our reaction to that was not necessarily to fight with them, but to invite them to participate. Our reaction to that was actually to attempt to enable the brand new employee to compete with them. Can we take that institutional knowledge and the ability to predict, to enable that researcher to look as good as the person who's been here for 35 years. We thought that was a realistic goal. It wasn't meant to be threatening, but more questioning if we can really do this? We've had the reaction, "You're trying to replace the scientist." That isn't true. Of course, the toolset vs. the robot scientist has an immediate reaction to attempt to go to a fully automated LAM, which is tens of millions of dollars just to get started. You need to build towards that, which is why I talk about the progressive value build. Another is viewing things as competitive approaches that truly aren't. Someone would say, "Well, we can do this with the inner LAM, we can do this with a LIM system." Well not really, those are tools to address one tiny problem in a really good way, but they don't have the extensibility of something we're talking about today, which is a fully fleshed out way of doing R&D differently. Getting back to the question, the appeal of working with someone like Citrine is that it’s a company that has done it many times and is very experienced at doing this. Another issue we ran into was: Do we want to be the company that actually develops our own software and, particularly, maintains it? That's something we didn't want to do. This wasn't something we could hand over to an enterprise-driven IT organization. We thought about having researchers interact with digital folks at a much more intimate level, where we're on the same page. However, the digital people we were hiring in a lab environment were not the people who would want to run and maintain enterprise software. That was something we were always considering, we didn’t want to take the program beyond this point. Johnston: That really demonstrates the value of this partnership. I want to revisit something we spoke about earlier, and that was sustainability. Scott, every consumer goods organization that we work with has sustainability mandates, and product innovation is a core component of that. How can these types of investments help companies advance their sustainability goals? Isaacson: Well, first sustainability is so cross-cutting, it’s a broadly-impacting initiative with so much work to do. We have to reformulate whole product lines; rethink whole supply chains. There is a need to have a tool that lets us go faster because if we just eat this elephant one forkful at a time, it'll take too long. Somebody's going to beat us to market or we're going to fall behind the regulations. There's real consequences for going too slow because the consumer preferences and the regulations are moving so quickly. The other point around sustainability is that it is forcing us to move towards sets of ingredients that are much less well-understood. That's a challenge and an opportunity because now it's harder to rely on intuition to understand how a formulation is going to behave if I swap out this, for this etc. Yet, it's an opportunity to be able to say that we have dozens or hundreds of potential ingredients that could satisfy the demand. Why don't we take a more structured approach? Why don't we use tools that help us evaluate, and narrow this funnel down in a way that is objective and faster than what we could do by then? For example, trying each of the 63 combinations one by one. Kieschke: That is the traditional approach. To your point, Scott, we don't have time to do the traditional approach. Talking about sustainable approaches, chemicals are not fashionable these days and may well be mandated out of all products before we have time to replace them. That is both a scary thing, but also a huge opportunity. How do we start to shorten the time frame? We do not have 10 years available to do this, we've got to do it in about 2. The opportunity then, is how do we start potentially working with suppliers with a much more data-driven approach to develop viable, environmentally-sensitive solutions in 2 years, rather than 10. Johnston: I’d like to hear from both of you: When you look at the future of automation within product innovation, what does that look like to you? Isaacson: What I see is scientists being able to contribute data to, not their own lab notebook or their own Excel file, but an ecosystem of data and models that can be shared and collaborated with across the organization. If I do an experiment in Ohio, then someone in Brussels will be able to benefit from that. Not only because they heard about it, but because now the knowledge that is coming from that experiment is integrated into a model that knows about Ohio, Brussels, and everything that you've done for the last 10 years. The effects of being able to learn from colleagues and mutually experience that acceleration is going to continue accelerating the pace of practicum. Kieschke: The near-term value is the way to connect disparate researchers into a common goal. It's not a near-term goal, it's a very tangible thing and it's going to happen, coupled with the notion of easy forward predictions. In other words, formulation to performance, embryonic reverse prediction capability, i.e. from performance to formulation space, is hugely powerful. Then, finally, full lab automation, which is automated equipment that is going to ingest the data then build the models. That is all very feasible. There's nothing impossible about what I've said. The accelerations that could be gained, I would estimate an order of magnitude compared to what we do now. We should have an expectation down the road of these problems being solved in a year, not 10 years. When we look back on this time, it will truly be a head slapping moment about why weren't we doing this sooner? Literally, the way we do things now will be unimaginable — why did we do it that way? Johnston: In terms of determining the ROI, Rob, how did your teams work to determine the ROI with this? Scott, maybe you can build on that, with what else you're seeing with other companies you're working with. Kieschke: Fortunately, I was never asked for an exact ROI on the program. I had enough and was able to convince our SVP and CTO that this was the way we were going to do R&D in the future. Our metrics were about technology diffusion rate, which is about how many people are adopting this, how quickly, and how many kinds of problems are we solving. In other words, that branch diffusion rate was not a very quantifying metric, but a metric we'd be using to measure success. Plus, functionality of the software, what could it do? What could it not do? Given we were such a diverse company, how many problems could it solve? Isaacson: ROI can be hard to quantify because there's a lot of different outcomes. Sometimes at Citrine we have to think about what happened on a particular set or product line of materials, and then decide what the ROI was from there. It can come from a number of different places. Sometimes you have to make estimates, just because you don't replay history twice. For example, you could say, "All right, we went from performance level X to performance level Y." How long would that have taken you in the real case, or in the traditional case with traditional tools? Maybe they'll say a year, or maybe we did it in three months. That is an actual example. That's pretty easy to quantify the ROI there. But there's more challenging ones sometimes about the opportunity costs, “I stopped early because I realized this, I need to pivot.” That has a real effect. Also, an expert integrates domain knowledge into a model and a more junior researcher is able to use that information to increase productivity. What's the ROI of that? Those are unintuitive results, all of these things contribute and can be hard to quantify, but they're very real. Kieschke: It is hard to quantify. You certainly can have an array of opportunities, which you can have a win-lose approach and have, essentially, a risk-adjusted portfolio of opportunities you would get in one way and not the other. It's tough to not acquiesce and say, "maybe some of these would've worked anyway." It’s really hard to quantify that. On the culture side, if you look at the way we want to modernize the organization or change the way we solve problems, there is certainly something powerful in generating a much more collaborative environment where workers jointly contribute towards the solution. Something most companies suffer from is the individual researcher alone in the lab, where they will claim credit for the entire product. People know that isn't true, but it's the way that people position themselves. It's almost like the single-handed hero model. This approach has the opportunity to share credit appropriately. The people who contributed the most data to the solution would actually get to share the credit. Once you think about that, you have not only a much more data-driven, but much more objective and egalitarian approach to how you credit people for contributions, which I think is powerful as well. It's culture change, which is tough to quantify. But again, if you look at what you want to be in 5 years as an organization, you'd certainly be picking those attributes, not necessarily sticking to the individual researcher mode. The other piece, of course, is collaboration with external entities. It may be easier, maybe less challenging, to actually have a pure data-sharing approach rather than having some joint development, which is always difficult to share foreground IP rights, too. People start selling data to each other. Johnston: Well, those are great words to leave us with. Rob and Scott, I'd like to thank you both so much for coming on to share your insight with our audience. It was incredibly valuable. I'd also like to thank Citrine for sponsoring today's webinar. All of our guests, thank you for giving us your time and I hope you have a great rest of your day.

Related Topics