How advanced industrial companies should approach artificial-intelligence strategy | McKinsey & Company

Artificial intelligence (AI) has reached a commercialization tipping point. As a result of several technology advancements that are now converging, major investments by technology companies and start-ups, and demand from businesses, AI is starting to have a major impact across markets. Advanced industries such as automotive, semiconductors, and industrial manufacturing could harness AI over the next decade to discover entirely new ways to make things better, cheaper, and faster. While popular reporting on the topic often tends to focus only on generating new business ideas, companies can directly apply AI to their current core business processes and operations. However, many companies have yet to think through how they could embed AI in their strategies and businesses. Instead, they take ad hoc approaches that might not scale, cannot prove out new technologies, and fail to build systematic capabilities. Ultimately, such efforts make little impact. Hype galore, but many businesses are not yet clear on using AI technology While projections vary, forecasters seem to align on an annual AI revenue-growth rate of roughly 50 percent. Analysts are bullish about AI, claiming it has the potential to drive significant productivity improvements via the automation of multitudes of manual tasks across nearly every industry. While the hype surrounding AI has reached critical mass, many companies remain unsure what to do with the technology. For example, one big concern: in a McKinsey survey of respondents from Chinese companies, results showed that AI is not a strategic priority for 43 percent of C-level executives. At the same time, the top three barriers to executing AI were the lack of required talent and capabilities, AI technology not being mature, and top management not being clear on the value of AI (Exhibit 1).

How to define an AI strategy Given the buzz and confusion that surrounds AI in general, business leaders need to determine what the technology can and cannot do for their company, and build an AI strategy based on those findings. Typical steps include identifying potential applications, playing out scenarios of AI-generated industry disruptions, defining a strategic stance and selecting underlying AI initiatives, and making the AI transformation happen. The first two steps focus on understanding how the external environment could evolve, while the second two cover what the company should do about it. First, move quickly from big-picture concepts to concrete, specific applications where AI creates business value (either sooner or later). Generate a list of these real-world applications by discussing the topic with everyone, including venture capitalists, research labs, the company’s own IT department, and consultants, and by conducting internal brainstorming. It often proves helpful to brainstorm on the applications along three archetypes. For example, operational enhancements could cover research, product development, manufacturing, supply-chain management, or marketing and sales. This might include introducing AI-enhanced predictive maintenance in manufacturing or developing R&D applications that could recommend modification areas to drive productivity. Another archetype focuses on product innovations, such as automated driver-assistance system modules, or even new business models. The latter likely builds on the earlier two, for example, using an autonomous-driving system to unlock shared-mobility and in-car services. Once identified, assess the level of potential disruption each AI application could deliver using seven product and value-chain questions. Does it provide an efficiency boost or cost reduction? Can it enable product substitution, thus producing an impact on the quantities of current products sold (and increasing others)? Does it integrate the value chain by combining segments? Will it speed up product commoditization, and thus affect product price and margin? Will it enable the company to better allocate resources? Does it provide alternative monetization models that can generate additional revenue from existing customers? Can it change the value chain from a linear structure to an ecosystem or a platform? The more that a company can answer these questions with a “yes,” the larger the likely disruption should become. When doing this exercise, do not allow technology changes to limit your answers. Experience shows that many times, AI combined with other business concepts creates the real disruption. For instance, think about self-driving robo-taxis enhanced with AI. Given the inherent uncertainty that surrounds AI, a war-gaming approach can help companies generate ideas. For example, have teams role play as start-ups, Internet companies, and direct competitors, and then play out how to invest in and use AI in the industry. This can enhance brainstorming and quickly separate reality from theory. The next step is to determine the industry impact of AI and how to respond to these disruptions, taking into account the top-ranked applications just identified. Understand which AI-induced disruptions could soon pass their tipping points for adoption, how they could change industry profit pools, and what strategic stance the company should take in response to them. To understand potential industry profit shifts, use the structure-conduct-performance framework. Examine any changes in performance from a structural point of view among existing value-chain steps. Will access to customers increase or decline, for example, and are upstream players more—or less—competitive? Next, analyze the conduct of current companies: Is their ability to differentiate products and services growing or declining, and have switching costs become greater or smaller? Shifts in structure and conduct typically result in associated profit-pool shifts. Conduct a tipping-point assessment to determine whether the AI initiative will go mainstream or remain a niche offering. Several early indicators can signal an approaching tipping point, such as growing customer awareness of AI, the solution’s ability to deliver a better total cost of ownership (TCO), and the introduction of relaxed regulations regarding the technology. Likewise, by testing the application against seven factors—universal utility, customer awareness, switching barriers, TCO, ecosystem compatibility, scalability, and influencers—companies can determine whether it’s ready for mass adoption. Offerings that include more of these factors have a higher probability of mass adoption in the near future. The first two steps of this process provide an understanding of how the outside world could change. Now define what to do about it. Understanding the magnitude and timing of profit-pool shifts will provide insights when companies formulate their stance on AI. For instance, should the organization drive adoption or simply strive to keep pace with others? A simple decision tree can help leaders work through these issues (Exhibit 2). Sharing this stance across the entire company and not just among leaders matters. Given AI’s complexity, creating clarity regarding which direction to follow is important, because it is likely the only way people will make the required effort and investments.

Translate the strategic stance by identifying the initiatives that could enable the company to outperform competitors. Define the feasibility of these applications for the business’s specific situation, determine their potential return on investment (ROI), and then prioritize the AI initiatives accordingly. Conduct a feasibility assessment. Almost every company will face gaps in its capabilities to implement the identified initiatives. The typical four gap categories are computing power, algorithm use, data availability, and adoption levels. An AI application focused on automated product ideation might experience feasibility gaps with regard to data (for instance, a lack of quality information or diverse data sets), or algorithms could lack the sophistication to deliver needed decision-making insights. Determine ROI. Work through the expectations for each AI initiative, either financially or strategically. For instance, a product-ideation application might feature deep-learning models capable of making trade-offs across cost considerations, production limitations, and customers’ preferences. If successful, the app could reduce design cycles and lower labor costs. Another application focused on manufacturing quality might feature an AI-enabled visual wafer-defect-inspection process that boosts quality levels (sometimes by more than 65 percent). Prioritize AI initiatives. Based on both feasibility and ROI, prioritize the AI applications, create a road map, and choose where to start. Companies can also use a project’s development timing and expected impact as additional parameters to fine-tune the sequencing. Although AI’s potential to make a company’s core business obsolete gets most of the attention, many experts instead expect it to inject that core with a powerful transformative juice that turns it into a learning organization. Exploiting this ability will be the key to AI success for most companies. Becoming AI ready has five major dimensions, each with explicit building blocks: Concrete action plans. Establish a concrete action plan to implement AI, including determining the appropriate partnership models. These plans should connect with the company’s overall business strategy to make use of the new technology’s capabilities. At a minimum, the plans should stipulate how the management team will adopt AI-influenced ways of thinking in daily business. It’s also important to connect with a partnership ecosystem. Most companies need to work with AI specialists to build the required solutions. Although many show a strong preference to own the algorithms and data, gaps in talent and technology know-how can stand as barriers to this option. Governance. Determine the target organizational model for AI. Doing so often involves establishing a centralized “AI engine” to provide support across the businesses. Decide how to build needed AI capabilities and competencies, including an AI team of data scientists, analytics engineers and translators, and the necessary infrastructure and algorithms required. Focus on redesigning work flows appropriately, to reflect the new operating model. This includes integrating data-management solutions with the enterprise architecture. Likewise, establish data-quality principles, policies, and practices that reflect AI considerations, and build a clear “playbook” of methodologies to consolidate cross-functional learnings. Infrastructure. Focus on the nuts and bolts of AI, including the establishment of data architecture with data lakes and data-search layers. This could mean consolidating redundant data-warehouse assets, where remaining relevant will require close oversight and the effective allocation of resources. Talent. Companies need to develop a special, multiyear plan to attract and retain the right AI talent. Given the different background and the ability to be more selective in their job search, AI talent requires more attention than talent in other areas. Typically, the CEO needs to be personally involved to show how serious the company is about AI and the individual. If attracting talent is hard, retaining it is often more challenging. We have seen examples where companies were able to convince talent to join with a grand strategic vision, but the enthusiasm vanished when the person’s career turned out to be less exciting than expected. Culture and understanding the best working environment play a role here. Culture. In some ways, fashioning an AI-focused culture can be the toughest part of a transformation. The key is to bridge the gap between the technical sphere (how do we make AI work?) and the business world (which real-life problems can it solve?). Leaders determine how to implement a change-management process and set up agile ways of working, potentially establishing an AI council to make strategic decisions. It also helps to nurture a start-up-like environment to support capability building and boost motivation. In many ways, successful tactics used for advanced analytics are also applicable here. For instance, some organizations establish an “AI war room” to bring people together to share knowledge rapidly and promote iterative working styles. Successful initiatives involve both the business units and top leadership in these discussions, since the entire organization needs to understand the new roles and responsibilities the AI transformation will necessitate. Fundamentally different from the current digital revolution, the coming AI surge will likely redefine 21st-century business practices. While potentially overestimated in the near term, its long-term impact will likely be profound. Therefore, leaders need to get some skin in the game, which means investing despite currently high levels of uncertainty. Company leaders (especially those not already thinking about the AI implications for their businesses) need to invest to turn AI into value that can be captured. While some may choose to wait and see at this point, such a posture will make later attempts to harness AI even more difficult in the face of disrupting competitors. Wouter Baan is an associate partner in McKinsey’s Beijing office, where Christopher Thomas is a partner; Joshua Chang is a consultant in the Taipei office.