In our last video, we looked at the relationship between AI analytics and a type of product information that the AI is best suited for. But there might be also many other organizational factors out there that may affect that precise relationship. In this video, I'm going to talk about how you organize your innovation may mediate that relationship. Some organization structures are just well suited to support AI driven innovation and some organization structures are not. In this video, we'll explore whether organizational factors such as how you organize your innovation teams, team composition, and how the skill set of employees could change the type of innovation that AI can help firms create. In this video, we'll focus primarily on the structure side. Organizations have very different ways of organizing their activities. They may organize their general activity in one way, and they may organize their innovations in another way. It's very important to look at specifically how firms are organizing their innovation activities, and also how they compare to the general organization structures. This structure can both vary within industries and also across industries. That means it's not just about industry differences, for example, drug R&D may just organize very differently from automobile R&D. All could be within the same industry that within drug discovery, one firm decide to organize their activity for innovation in one way and another firm would choose a different way. To give you an idea of what I'm talking about, let's see some examples. These are two firms, Google and Apple, and they organize their innovation activities in a very different way. These two are very innovative firms in technology sector, and if you look at Google, their individual structure is much more dispersed whereas on Apple it's very much concentrated into a few clusters. What are these clusters? What are the nodes? Where are the links in this graph? Nodes in these graphs are inventors. Two nodes are linked when they wrote a patent together. You can see Google, they have several cluster, but they're pretty much decentralized whereas apple, you see the innovation structure really concentrated on a few cluster at the big one in the center, also in another one in the bottom, where in Google is much more dispersed. This is a way how firm naturally organize themselves. It could be many different reasons, could be historical reasons, could be different way of thinking about organizations, thinking about innovations, so many factors. We don't really know why they organize it the way it is. But that's not the most important points but I just want to show you that firms organize innovation differently. These structure once formed are actually difficult to change because it's hard to change the innovation relationship, a formal structured by fiat, or changing the CEO or changing research R&D head. It's not necessarily going to change the structures quickly. But analytics in AI are changing very quickly. Can we see the adoption of AI analytics change these firms differently? Can this technology help Google type of the world better, or the Apple type of a world? In a sense, can they help a more disperse innovation structure more so than a concentrated version or vice versa? This is not just Apple versus Google because they are two famous firms and they are both very innovative and maybe just specific to tech sectors were very innovative firms. You see the same thing in various other industries as well. This is two firms, Sanofi and Roche. These are big drug makers. Again, you see they're very productive drug makers. But these pharmaceutical companies have very different way of organizing. Sanofi is slightly more dispersed, where Roche is much more concentrated into a few clusters. Similar question we can ask is that can analytics and AI help the Google, Sanofi of the world or Apple, Roche of the world to innovate. There are pros and cons of centralization, decentralization. There are many advantages to decentralization in the sense that people are focusing on their specific type of work. They're very much local to their market and they tend to straight, sticky knowledge. By sticky knowledge are just things you know, and it's hard to move from one group to another because there are lots of tacit knowledge thoughts of things you just know amongst each other within a team, but it's hard to translate that information to others. When people worked together on a specific problem for a long time, they are more likely to identify problems and create alternative solutions. But because they are decentralized it's sometimes difficult to facilitate that cross-group coordination. They're very good at working with each other within a group, but might be a little difficult to translate that knowledge or worked with another group who may have a very different way of communicate with each other or a different way of creating their own lingos or own sticky knowledge. On the other hand, centralization can also provide many advantages. Centralization tend to crave very radical innovation applicable beyond that individual group. In decentralized world, you may be very good at identifying a problem, a solution of your own group. But your solution may just work for you, but it may not work broadly across all groups, whereas centralization can help you do that better. Centralization also has a broad search for external information beyond individual group that they may decide that certain resources, certain things are just better for the entire organization as opposed to thinking about that individual group. Individual group tend to just think about relatively myopically, what is good for my group without thinking broadly. Centralization can overcome some of the cross-departmental co-ordinations that decentralization doesn't have, but decentralization really good at solving problem locally and identifying novel solutions specific to that business context. There are pros and cons in both cases. How do analytics and AI come in? Do you think knowing what you know about analytics is going to help centralized organization better or decentralized innovation structure better? If you think about what analytics and AI can do, one of the key advantage is that they can conduct broad search for diverse knowledge. By doing that broad search and finding that hidden relationship that you haven't seen before, they're also linking existing silos of information in a way that helps decentralized structure. By having a tool that automatically mine across different fields and finding hidden patterns, it can in some ways mitigate the central weaknesses of decentralization in terms of coordination. It can not solve all of the coronation, that's just impossible. But even chipping a little bit of that difficulty, that may facilitate decentralization structures in performing innovation activities. Remember in our last video, we showed that artificial intelligence, machine learning, data analytics are especially great at finding that new combination or new diverse combination or a new way of using existing technology, existing combination in a new way. That is all going to be great for helping decentralization structure in solving their problem. It says, it seems like another technology invented somewhere else may be used to help me finding a solution to my problem. Again, if I mentioned earlier, you think about cancer oncology, p53 protein. If you think of a new chassis or music, these are all great new combinations by reusing the existing combination in a new way or reusing existing technology in a new way, or finding a diverse combinations together. That essence can help decentralization by putting the silos of information and silos technology, and by combining them a new way. Let's see if our intuition is correct. Again, we can measure the innovation structure of the firms. Here we're using machine learning to understand machine learning. Here we're using community detection algorithm on co-authorship network. Basically, what it does is capture how many groups are there in this graph and how connected they are to each other. In a very dispersed crowd in the Google's role, you'll see a very high measurement on decentralization, on the concentrate and in Apple's role you see a high-value on centralization. I want to make a note that this is different from formal hierarchies. Number one, they're very difficult to capture, and then number two, they may not reflect on what people are really doing. Just because you and another person who has worked together many times before are reworked into different organizations, it doesn't mean you won't work with each other anymore, you won't think about new patent to file. They may affect to some extent, but it won't affect that dramatically. Here we're actually capturing the real way people are working together, the real way people are innovating together, by looking at their co-authorship on a patent. This is the distribution of dispersion innovation. Again, the high value means you're very dispersed, and low value it means you're very concentrated. Again, you see a huge spectrum of firms. Some firms are very dispersed, some firms are very concentrated. You see a spike of zero here just for a firm that have never filed any patent, so that's a zero by default. But if you ignore any zero column, you see that dispersion is quite spread out. You can use this measure to gauge your own organization structure. You can see how your formal hierarchy of your structure, how that different from the informal collaboration structure, what's actually being done by the employee themselves. Because it's you own firm, you don't need to just use public data search patent, you can look at trademarks, product design teams, white papers, and publication they've created together. All of that can help you measure the actual internal collaboration structure of a firm. You can also see how this structure is different from your competitors. You can look at your own patent co-authorship network, that's a collaboration pattern, and compare that with your competitors to see how you organize your activities differently from your competitors. We'll look at data analytics and we look at AI investment. These technologies complement decentralized innovation structures in firms. That means decentralized innovation structures benefit tremendously by having investment in AI and investment in data analytics. Not only are they more likely to adopt, and when they do adopt both together, that means they invest in both AI and they have decentralized innovation structure, they're more productive than the firm with only AI or only decentralization. They are about 3 percent more productive and that's pretty big difference considering how competitive some of these industries are. We'll look at the far between the most recent years and the earlier years. The effect is even stronger in a recent years, that's a firm learn to know about how to deploy analytics in AI innovation, but structures matters dramatically. Data analytics and AI can help decentralize innovation structure, and they can do that better than centralized structure. I showed you that AI can support decentralization better than centralization. But again, there are nuances to that. It's not going to support all types of innovation, it's going to support a specific type of innovation, as I mentioned earlier, that is the combinations of existing things in a new way. Recombination innovation, diverse combination innovation. Again, I mentioned earlier that a new combination is something you combine existing technologies in a new way. For example, in our earlier example, we have four technologies, A, B, C, and AB. You can think of a new combination will be BC, combining B and C together, or a diverse combination, putting A, B, C together, forming ABC. We can also classify them in a little different way in the sense that we can call AB, ABC, new combinations that hasn't existed before. Or we can see A prime, B prime, C prime are slight improvement where refinement of existing technology A, B, and C. A prime will be an improvement on A, B prime will be an improvement B, and C prime will be an improvement on C. Again, we can also mention D, a brand new technology hasn't existed before or new subclass in [inaudible] , will be D, will be novel technology. Again, we can classify innovation into three types. Brand new technology will be like D, or a new combination, like of AB, BC, ABC, or reuse A prime, B prime, C prime. We can benchmark the new names against globally or locally to your firm. In a sense, we can say, are you the first average who generate a technology called ABC and that has never existed in an entire world? Or you can say, although I'm not the first to create ABC, but I'm first one to use it in my own industry. You can say local to that particular firm. We can measure this three by two combinations, novel new combination, reuse, at least three can be applied globally and also locally to the specific firm. Whether it is new, is it global to the entire world, and whether is this local to that firm, to that industry. In this graph, I'm going to show you how likely is it for AI investment to help decentralized innovation structure to produce all different 6 type of innovation. When a firm had decentralization and they invested in AI, what is the likelihood of creating a new global technology class? That's the first graph. Or a new combination globally? Or a new combination locally to it's firm, or a reusing of existing technologies. I'll explain each one in turn. The first one is on new technology class. This is how we read, this is that one of firm that has decentralized innovation structure and they invest in data analytics and they invest in AI, do that combination of decentralization AI investment help that firm create new global technologies? You'll see that the effect is almost negative. Now let's look at the three graphs, the scores very high, but notice what they are. The second column is called new technology local. What's does it mean is, that is a brand new technology class that this firm or industry hadn't seen before, but that technology class existed elsewhere. Again, you're incorporating a brand new technology class that exists elsewhere in the world and bring that into your industry, your own industry, and in your own firm. The highest bar is actually called new combination globally. That means you combine technologies in a new way and that's the first time ever in the entire world that combination has existed. Slightly lower is new combination locally that this combination has happened in other industries before. But you're first one to do it in your own firm, in your own industry. These three type of innovation are the key to be benefited from AI and decentralization. On average, you look in the middle bar on the highest bar on new combinations, global combinations. That's an average four patents per year. In a sense, if you invest in AI and you have decentralized innovation structure. If you have a decentralized innovation structure is seemed to really pay off if you using AI to innovate patents that are combination of existing technologies together in a new way or refinement of existing technology that has happened in other industries already. Again, bringing AI in the innovation into your own firm or industry or in own content in a new way to solve the problem specific to you. Next, we see reuse. Reuse I just think about the A prime. Basically, the A prime is a slight improvement of existing technology A. It's still innovation, but it's incremental. You see that fact is there, but it's not that large. If you think about it, if you already have a technology that you know pretty well if you wanted to create incrementally. You probably don't need a heavy duty AI technology to help you do that. You probably know already what you need to do because in here, the search area around that technology is relatively small. It doesn't really benefits much from using AI. Whereas for global combination you had to go through two to the N in combinatorics. N is the number of technologies. You have, say, 1,000 technologies. You had to go through two to the 1,000 combination to figure out which one works. That's a much bigger search space and that's where AI is really helping you figuring out which combination is valid, which combinations not. That combination is especially powerful when you have decentralized innovation structure because that decentralized innovation structure really needed the actual power of expanding the search space that AI can provide. Centralization doesn't need as much because they already have ability to coordinate a search broadly to this, to some extent. But for decentralization, they precisely need that it's capability of searching broadly. Here you see that in our empirical analysis is really about having analytics and to create new combinations for decentralized innovation. This is just not by combination, we can also examine it outside of the technology costs. In the prior measurement, we only rely on the technology cost that patent authorities allocate. But here we can look at other citations. If you believe your patent is very noble, you should see that in some of the patent statistics. Here we look at patents statistics in a sense that in the first one bar will be 90-100 that means a 100 percent or 90 percent of the citation in your patent are coming for your own firms. Here is a entirely incremental of your own existing innovation. Also 10 percent it could be I only said it cited a 10 percent of my own patent that firms own patent and 90 percent are from outside. Again, we did this analysis based looking at AI investment decentralization. That's what happens when you have both AI investment and you apply it on a decentralized innovation structure, what would the outcome be? Again, you see, that's really the middle range, between 40-60 range that see the greatest being of buck and this says that you are not borrowing entirely from knew from outside but borrowing just enough. You're sighted just enough. You utilize the existing technologies and you cited just enough from outside, and not combination in the middle range. This intermediate nobility seem to be the sweet spot for AI to help decentralized or under the structure to innovate. You'll notice on the very new ones on the variant zero,10 percent, in the very right-hand side of graph. Again, you see a negative act. Small but negative fact. Probably because that one is something so new you may not be able to know how to use it well. When it's half and half you know exactly how to apply some way to fit your own needs and contexts. Ones is brand new complete new, you just bring him outside. That doesn't seem to be as effective and then doing them in middle-range. In productivity again, we see the same thing. We see that when you invest in AI and having decentralized innovation you are much more productive and the first team is stronger in recent years. In the first bar is for all years, and the second bar is from 1988-2007, and the third bar is 2008- 2013. You see the move, the biggest effect cause really fun the reason years between 2008 and 2013, where do you see the biggest improvement? In about 3 percent more productive that's firm that adopt AI and have a decentralized innovation structure are 3 percent more productive than other firms.