Gen AI’s productivity promise: Huge potential but most have not yet reached scaled impact

Businesses across sectors are looking to generative AI (gen AI) to address increasingly sluggish productivity in recent decades. Many companies are simply finding that productivity gains are no longer that easy to grab—and this has not been helped by economic challenges in the form of higher inflation, geopolitical tension, and supply chain disruption.

Recent analysis from McKinsey client engagements indicates that around 80 percent of activities across the operations environment could benefit from some level of automation through gen AI, or AI more broadly. The potential gains from the arrival of gen AI could help offset decelerating productivity growth in advanced economies. In the decade 2012 to 2022, for example, productivity growth fell below 1 percent per year, from an average 2.2 percent per year in the five years to 2002, and 1.6 percent per year from 2002 to 2007.

Despite the enormous potential of AI to improve productivity, a McKinsey survey of 150 executives at major European and North American companies with revenues exceeding $1 billion suggests that many organizations have not yet been able to implement gen AI at scale.1 Just 3 percent of respondents to the survey said their companies have already scaled at least one gen AI use case in an operations-related domain. Nearly a quarter (24 percent) are piloting use cases, while 63 percent remain in an exploratory phase or have not started deploying gen AI operationally at all.

Companies are often not yet clear on the value they can capture from AI. More than two-thirds of organizations (68 percent) reported that they have not quantified the potential impact of gen AI for their business. Only 23 percent have quantified the potential for their whole organization, while another 9 percent have analyzed the potential gains for certain functions. This can make it difficult to prioritize AI versus other investments, and slows adoption.

Frontrunners pull ahead in the gen AI race

The survey results indicate the emergence of two groups in the gen AI race: frontrunners with active pilots and quantified benefit assessments, and laggards still focused on research and early exploration.

The main reasons cited for slower adoption were lack of AI maturity and governance (cited by 25 percent)—where technical, organizational, and employee capabilities do not match the needs of the business; and insecurity about the human experience element (19 percent)—where concerns remain about how employees will interact with gen AI solutions, or the possible risks related to gen AI output, like hallucination or biased outputs. The third most cited reason for delayed deployment was an unclear road map to value—cited by 17 percent of respondents—where the value of use cases or the path toward realizing potential value remains unclear.

Companies in the slower-to-adopt group could be missing out on important benefits, apart from productivity. Frontrunner companies are realizing a 5 to 10 percent decrease in costs, on average, while refocusing their employees on more value-adding activities and boosting innovation—whether through shorter cycle times in R&D testing or code development, or by enhancing employee creativity.2

Early lessons: How to capture gen AI value in operations

To speed up adoption and capture benefits across productivity, efficiency, and innovation, organizations can learn from what frontrunner companies are doing differently in five key areas: business case, technology, talent, data architecture, and change management.

Business case. Organizations tend to jump quickly into single use cases and narrow applications for gen AI, when, in fact, the real value tends to come from the smart combination of different use cases to truly transform a process or domain. To capture the true value at stake, organizations could begin to reimagine the end-to-end digitization of processes, and the role gen AI can play at different points.

Technology. Businesses have tough choices to make regarding technology architecture and the type of gen AI tools to adopt. Solutions fall into the three archetypes of “taker,” “shaper,” and “maker”—with each requiring different levels of investment and capability, and each presenting different levels of potential risk and reward (exhibit).

  • “Taker” use cases include software as a service (SaaS) and off-the-shelf solutions such as copilot tools for developers—for example, Microsoft’s GitHub product. They are typically fast and easy to implement, though the value to be captured is also generally lower.
  • “Shaper” approaches include use cases that enrich pretrained models with proprietary data sources. These are more time-consuming to implement than off-the-shelf solutions but are suitable for a range of applications. Examples include tools linked to internal knowledge sources, such as McKinsey’s own Lilli tool for our consultants.
  • “Maker” use cases encompass bespoke gen AI tools and are usually geared toward narrow applications. While these are more expensive and difficult to implement, the rewards can be significant in the shape of true differentiation. For example, consider a company-owned R&D tool developed by a McKinsey client that features gen AI capabilities that can suggest new parts for development. This could be a real source of competitive advantage via accelerated, high-quality development.
Three archetypes of generative AI use cases have different implications for technology and data requirements.

Talent. The biggest current challenge to gen AI success is scarcity of talent. This includes both a lack of proficiency in using gen AI among the existing employee base, and the specialized IT, business, and legal competencies needed to build and refine gen AI applications. To address the first talent problem, companies could shape targeted learning journeys to build critical gen AI skills in data- and AI-related roles. For the second issue around specialist skills, organizations could look to build out their inventory of AI and gen AI skills within the enterprise and define the roles and skills they need to deliver their gen AI plans in the next three to five years. Identifying current skills gaps can also help inform whether companies must build, buy, or partner with an external vendor to get their near-term gen AI plans off the ground.

Data architecture. Data is the fuel for generative AI, and those companies that have worked on their data architecture in recent years may be best placed to capture early value from AI. Having one integrated IT landscape with a consolidated data model and clear sources of truth for each major data category eases the introduction of gen AI solutions in the shaper and maker categories, for example. Access to quality data and good data governance are integral for those organizations wanting to train or adapt a large language model (LLM) to their business.

Change management. As with any digital transformation, change management can make or break success and dictate return on investment. Organizations may need to rewire their operating model and get IT working better in tandem with business owners to ensure gen AI solutions meet real business needs and fulfill their organizational potential. Responsible deployment demands an “AI trust” framework that highlights potential harms and assures that mitigations and guardrails are in place—and tested in an ongoing manner. This is key to effectively introducing AI-powered changes into employee and customer experiences.

Organizations across industries are looking to gen AI to solve their ongoing productivity challenges. While some are pulling ahead and capturing real results, most are still stuck in the exploratory phase—hamstrung by concerns around risk, reward, and organizational readiness.

These organizations may need to find some courage and build up the confidence of their stakeholders to ensure they harness this powerful technology to boost productivity. For inspiration, they can look to those companies further along on their gen AI adoption journeys to understand which factors can make the difference in success or failure.

Be part of the conversation with McKinsey Talks Operations. Learn more and register for events with our Operations experts here.

1 n = 150 executives at European and North American companies with revenues exceeding $1 billion (February 2024).
2The economic potential of generative AI: The next productivity frontier,” McKinsey, June 14, 2023.

Connect with our Operations Practice