In our last video, we learn about the effects of AI on innovation. Specifically, we looked at how your firms should structure your innovation to receive the most benefit from analytics in AI. We find that AI analytics can complement decentralized innovation to improve the firm's ability to create recombination innovation. In this video, we're going to talk about what kind of skills should the firm pay attention to, but where should this data analytics skill come from? And how do you organize the skills to best facilitate innovation for your firm? On one hand, AI skills and didn't let the skills are in high demand. As you see in the prior slide, you see that AI skills and data skills has increased dramatically over time, especially the last 10 years. So one hand is the corresponding strategy should be well if AI skills matter for innovation, we should just hire more people with AI skills, especially AI skills of researchers who are at the hire innovation and activities, okay? That's probably really hard to do and very expensive, imagine hiring a drug innovation expert, and we're also really well versed in the AI, okay? That's probably pretty hard to find, and if you find them is probably very expensive to find too, to hire them, or you can hire anyone with the AI skills. If so, where would you put up? So what does it matter to have inventors who have AI skills, or it doesn't matter, you can just hire any AI skills, and they can help the researchers to use AI for discovery. So we actually look at this precisely in our research. Here, we looked at inventors who also have the AI skills. We also look at the general non-inventors who also have AI skills in the firm. Look at which type of employees contribute more to AI output in the AI productivity. Interesting, you see that having AI skills and inventors is large but is also very high, so we can't really tell the effect of that. But the good news is that if you look at the AI skills for non-inventors has a pretty strong effect almost as large as the inventor with the AI skills, but the effect is much stronger in the sense that variance is low. On the first hand, we can't really tell if AI skills and inventor really matters, but we can tell precisely that AI skills for non-inventors really matter. So the good news is that you don't need to hire people with AI skills who are also inventors. You can hire people who knows the AI skill and help these inventors utilizes AI skills to help them discover the things they need to do and help them to achieve the innovation goals. Okay, good news that you can hire AI skills broadly doesn't have to be people who are in research who are who have strong sales in both, okay? Should AI skills be concentrated in a single department or should they be embedded within different functional departments? This is an interesting question because IT traditionally, information technology departments are often concentrated in a single department managed by a CIO or managed by a head of technologies, okay? And that's generally how firms manage their IT resources and IT department. Should the same configuration apply for applying for managing AI skills, okay? Should firm decentralized their AI skills and get them into specific functional teams, specific product teams, specific functional department. Well, this is not an obvious answer to have because firms tend to do this both ways. For example, Twitter at first centralized their AI skills in the single department just like you know how you organize IT skills, and they created many interesting tools and products. And they believe that these tools can help teams at Twitter. Unfortunately, most is product team actually did not find the tool that these centralized AI teams have create to be very useful. So after a few years, Twitter decided to decentralize their team by embedding AI skills within the functional teams to health each product, reaching their goals using AI. And this is not just for tech companies, General Electric also experienced something similar. On a conversation with the head of research at General Electric, the person told me that it's actually most optimal for General Electric to pair an industry expert with someone with high data and AI skills to jumpstart their product innovation. The industry experts can provide intuition and expertise and providing enough context and problems that relevant to the company. People with AI skills can help the person solve the problem, figure out what data to use, how to implement it, okay? So because it's very important to have a contextual knowledge, AI is a general purpose technology. It can be used to solve the problem, but you need to define the problem well. So at first again, GE has centralized AI skill and found it not to work very well. The solution, the product they developed was beautiful, but again, they're not very useful for their product teams. So later, GE also decentralized their configuration by pairing industry expert with a person with the AI skills. So we actually look at this phenomenon, more details. So if this is just specific GE was specific to Twitter, we actually looked at the dispersion of data skills across seven functional groups. These are functional groups are typical in most all firms. We looked at the manufacturing department, engineering, sales and marketing, human resources being number four, five, accounting and finance, six, R&D, and seven, administrative. So these seven function that's often occurring in all firms. And we look at how data skills are dispersed across the seven groups, and we look at both with industry and also across industry comparisons. And we looked at it generally how this is distributed. Again, we service about 600 firms and looked at dispersion questions and their corresponding survey responses as the following. The question is to what extent do you use the analytics in each of the following groups? We asked them to say frequently, or not frequently, for each of the seven functions. And we also looked at globally, all firm globally or just US firm, just European firms, and just Asian firms, okay? You can see there's a huge heterogeneity across different countries, across different regions of the world. So it's not very obvious that data skills should be dispersed, and data should be concentrated, and this is dispersion scores by industry. So basically, if you have very high dispersion score, that means your data skills are distributed quite evenly across seven functions, okay? If your dispersion score is very low, that means you highly concentrate your data skill in a single or one or two departments, okay? So this is a median data dispersion score by industry. We look at various industries out there were classified into each, the highest being the management of companies and enterprises. 86.4% of these firms are having a highly dispersed score, meaning that data skills are highly dispersed across different functions within the firm. We see health care and social systems industry having a very low score, meaning that their data skills or AI skills are still relatively concentrated into a few departments. There's definitely industry differences. And this is the cluster of scored by region, okay? You can see that for data dispersion in the middle column, you can see that Europe is actually more dispersed in the US, and APAC about the middle. And you can also see the other type of data related capital, the left-hand side data related human capital. They're pretty much the same across regions, and you can see data decisions. Europe is being a little high and US is being low, okay? There's definitely cluster of scores by regions, okay? And across different industries and if you look at the scores and average maybe the same, the variance of them is very, very large, okay? So it shows you there's definitely heterogeneity, people haven't figured out exactly how to organize the data skills yet. So we actually look at this phenomenon, and we examined to what extent of the dispersion actually beneficial firms, okay? So we actually find that economical example with Twitter and General Electric is correct. In a sense that when firm dispersed their AI and the skills across different functions, meaning by embedding them into different functions or product teams, okay? These firms are far more effective, are far more productive than firms that did not do this. On average, a one standard deviation above the mean in dispersion score have about 3 to 5% higher enterprise value, okay? And given the size of most firms in a sample and 3 to 5% change enterprise value for a medium firm in our sample. So 3 to 5% change of a $10.7 billion is between $322 to $534 million, whose is a substantial change. So again, percentage rise may be low, but translating into a large number 1%, 3%, 5% is huge.
