AI has been gradually integrating into our daily routines, from smartphone technology to self-driving car features and retail tools enhancing customer experiences. This slow but steady progress has made the AI revolution almost invisible. While milestones like AlphaGo’s victory in 2016 were celebrated, they quickly faded from public attention.
However, generative AI applications like ChatGPT, GitHub Copilot, and Stable Diffusion have captured global attention in a way that previous AI advancements did not. Their broad usability—accessible to almost anyone for communication and creation—and their remarkable ability to engage in conversations have sparked widespread interest. These new generative AI tools can handle routine tasks such as data reorganization and classification. Yet, it’s their capacity to generate text, music, and digital art that has made headlines and encouraged individuals and businesses to experiment and explore their potential. Consequently, a diverse group of stakeholders is now grappling with generative AI’s effects on business and society, often lacking the necessary context to fully understand its implications.
About the Authors
This article represents the collective insights of Michael Chui, Eric Hazan, Roger Roberts, Alex Singla, Kate Smaje, Alex Sukharevsky, Lareina Yee, and Rodney Zemmel, drawing from perspectives across QuantumBlack, AI by McKinsey; McKinsey Digital; the McKinsey Technology Council; the McKinsey Global Institute; and McKinsey’s Growth, Marketing & Sales Practice.
The rapid pace of generative AI development adds complexity to this understanding. ChatGPT emerged in November 2022, and within four months, OpenAI released GPT-4, a significantly more advanced large language model (LLM).1“Introducing ChatGPT,” OpenAI, November 30, 2022; “GPT-4 is OpenAI’s most advanced system, producing safer and more useful responses,” OpenAI, accessed June 1, 2023. Similarly, by May 2023, Anthropic’s Claude could process 100,000 tokens of text—about 75,000 words, or a novel’s length—per minute, a vast increase from 9,000 tokens in March 2023.2“Introducing Claude,” Anthropic PBC, March 14, 2023; “Introducing 100K Context Windows,” Anthropic PBC, May 11, 2023. Google also announced generative AI-powered features in May 2023, including Search Generative Experience and PaLM 2, the LLM behind its Bard chatbot and other Google products.3Emma Roth, “The nine biggest announcements from Google I/O 2023,” The Verge, May 10, 2023.
To understand the future, it’s crucial to recognize the breakthroughs that have fueled generative AI’s rise, decades in the making. For this report, generative AI refers to applications built on foundation models. These models utilize extensive artificial neural networks, mirroring the neuron connections in the human brain. Foundation models are a part of deep learning, characterized by multiple layers within neural networks. Deep learning has been central to recent AI progress, but foundation models powering generative AI are a significant leap forward. Unlike previous deep learning models, they handle vast and diverse unstructured data sets and manage multiple tasks.
Foundation models have enabled new capabilities and greatly enhanced existing ones across various formats, including images, video, audio, and code. AI trained on these models can perform diverse functions: classifying, editing, summarizing, answering questions, and creating new content.
We are just beginning to grasp the full potential of generative AI. This research is our latest effort to evaluate the impact of this new AI era. Our findings suggest that generative AI is set to revolutionize roles and boost efficiency across sectors like sales, marketing, customer service, and software development. This transformation could unlock trillions of dollars in value across industries from banking to life sciences. The following sections detail our initial findings.
For the complete report, download the PDF.
Key Insights
Generative AI’s impact on productivity could inject trillions of dollars into the global economy. Our latest research indicates that generative AI could add between $2.6 trillion and $4.4 trillion annually across 63 analyzed use cases. For context, the UK’s 2021 GDP was $3.1 trillion. This contribution would elevate the total impact of all AI by 15 to 40 percent. This estimate could double if generative AI is integrated into existing software applications beyond these specific use cases.
Approximately 75 percent of the value from generative AI use cases is concentrated in four areas: Customer operations, marketing and sales, software engineering, and R&D. We examined 63 use cases across 16 business functions where generative AI can solve specific business challenges and deliver measurable outcomes. Examples include enhancing customer interactions, generating creative marketing and sales content, and drafting code from natural language prompts.
Generative AI will significantly impact all industry sectors. Industries like banking, high tech, and life sciences are poised to see the most substantial impact relative to their revenues. In banking, for example, full implementation of generative AI use cases could yield an additional $200 billion to $340 billion annually. Retail and consumer packaged goods also show significant potential, at $400 billion to $660 billion per year.
Generative AI is poised to reshape the nature of work, enhancing individual worker capabilities by automating specific activities. Current generative AI and related technologies can automate 60 to 70 percent of the work activities that currently consume employee time. This is a significant increase from our previous estimate that technology could automate half of work time.4“[Harnessing automation for a future that works](/featured-insights/digital-disruption/harnessing-automation-for-a-future-that-works),” McKinsey Global Institute, January 12, 2017. This acceleration is largely due to generative AI’s improved natural language understanding, crucial for activities representing 25 percent of total work time. Consequently, generative AI has a greater impact on knowledge work in higher-paying, more education-intensive occupations.
Workforce transformation is likely to accelerate due to increased automation potential. Our updated adoption scenarios, considering technology development, economic feasibility, and adoption timelines, suggest that half of today’s work activities could be automated between 2030 and 2060, with a midpoint around 2045—roughly a decade sooner than previously estimated.
Generative AI can substantially boost labor productivity economy-wide, but requires investments in worker support for activity shifts and job changes. Generative AI could drive annual labor productivity growth by 0.1 to 0.6 percent through 2040, depending on adoption rates and worker redeployment. Combined with other technologies, work automation could add 0.5 to 3.4 percentage points annually to productivity growth. However, workers will need support in acquiring new skills, and some will transition to new roles. If managed effectively, generative AI can significantly contribute to economic growth and a more sustainable, inclusive world.
The generative AI era is just beginning. While excitement is high and initial trials are promising, realizing the full benefits will take time. Business and societal leaders face significant challenges, including managing inherent risks, identifying new workforce skills, and rethinking core business processes for retraining and skill development.
Where Business Value Lies
Generative AI is a transformative advancement in artificial intelligence. As businesses rapidly adopt and adapt, understanding its potential to deliver economic and societal value is critical for informed decision-making. We used two complementary approaches to pinpoint where generative AI, with its current capabilities, can deliver the most value and its magnitude (Exhibit 1).
The first approach examined potential use cases for generative AI adoption by organizations. We define a “use case” as applying generative AI to a specific business issue, resulting in measurable outcomes. For instance, in marketing, a use case is using generative AI to create personalized emails, with outcomes including reduced content generation costs and increased revenue from more effective, scalable content. We identified 63 generative AI use cases across 16 business functions that could deliver a total economic value between $2.6 trillion and $4.4 trillion annually when broadly applied.
This would increase the economic value of all AI and analytics by 15 to 40 percent, adding to the $11 trillion to $17.7 trillion we estimate for non-generative AI. (Our 2017 estimate for AI value was $9.5 trillion to $15.4 trillion.)
Our second lens complements the first by analyzing generative AI’s impact on work activities within approximately 850 occupations. We modeled scenarios to estimate when generative AI could perform over 2,100 “detailed work activities”—like “communicating operational plans”—that comprise these occupations globally. This helps estimate how generative AI’s current capabilities could affect labor productivity across the global workforce.
Some of this impact overlaps with cost reductions in the use case analysis, assumed to be from improved labor productivity. After netting out this overlap, the total economic benefits of generative AI—including major use cases and widespread productivity gains across knowledge work—total $6.1 trillion to $7.9 trillion annually (Exhibit 2).
Estimating Generative AI Use Case Value
To assess the potential value of generative AI, we updated McKinsey’s proprietary database of AI use cases, drawing on insights from over 100 industry and functional experts.1”[Notes from the AI frontier: Applications and value of deep learning](/featured-insights/artificial-intelligence/notes-from-the-ai-frontier-applications-and-value-of-deep-learning),” McKinsey Global Institute, April 17, 2018.
Our updates focused on generative AI use cases—specifically, how generative AI techniques (mainly transformer-based neural networks) can address problems not effectively solved by previous technologies.
We only analyzed use cases where generative AI could significantly improve key value-driving outputs. Our primary value estimates exclude use cases where the sole benefit is natural language capability. For example, natural language is key in customer service use cases, but not in logistics network optimization, where value is primarily from quantitative analysis.
We then estimated the potential annual value of these generative AI use cases if adopted economy-wide. For revenue-increasing use cases, like those in sales and marketing, we estimated the economy-wide value by assessing generative AI’s impact on sales and marketing expenditure productivity.
Our estimates are based on the 2022 global economic structure and do not account for value creation from entirely new product or service categories enabled by generative AI.
While generative AI is exciting and rapidly advancing, traditional AI applications still represent the majority of AI’s overall potential value. Traditional advanced analytics and machine learning are highly effective for numerical and optimization tasks like predictive modeling, with ongoing new applications across industries. However, generative AI’s continued development promises to unlock new frontiers in creativity and innovation, expanding AI’s overall potential (see sidebar “How we estimated the value potential of generative AI use cases”).
This section highlights generative AI’s value potential across business functions.
Generative AI can impact most business functions, but a few stand out in terms of impact as a share of functional cost (Exhibit 3). Our analysis of 16 functions identified customer operations, marketing and sales, software engineering, and research and development as potentially accounting for about 75 percent of the total annual value from generative AI use cases.
Notably, functions like manufacturing and supply chain, previously prominent in AI use case sizing, now show lower potential value from generative AI.5Pitchbook. This is largely due to the nature of generative AI use cases, which exclude many numerical and optimization applications that were key value drivers for earlier AI applications.
Beyond function-specific use cases, generative AI can drive organizational value by revolutionizing internal knowledge management. Its natural language processing capabilities allow employees to retrieve internal knowledge by querying in natural language and engaging in ongoing dialogues. This empowers teams to quickly access relevant information, make better-informed decisions, and develop effective strategies.
In 2012, the McKinsey Global Institute (MGI) estimated that knowledge workers spend about 20 percent of their time, or one day per week, searching for information. Generative AI can take over these tasks, significantly increasing worker efficiency and effectiveness. This virtual expertise can rapidly process vast amounts of corporate information stored in natural language and quickly analyze source material through dialogue, offering a more scalable solution than relying solely on human experts.
In other scenarios, generative AI augments worker productivity through collaboration. Its ability to rapidly analyze large datasets and draw conclusions provides insights and options that can dramatically enhance knowledge work, speeding up product development and freeing employees for higher-impact tasks.
The following examples illustrate how generative AI can deliver operational benefits in key use cases across the business functions with the highest value potential from our analysis of 63 use cases. The first two examples show generative AI as a virtual expert, while the latter two illustrate its role as a virtual collaborator.
Customer Operations: Enhancing Customer and Agent Experiences
Generative AI has the potential to transform customer operations, improving both customer experience and agent productivity through digital self-service and enhanced agent skills. Its ability to automate customer interactions using natural language has already gained traction in customer service. Research at a company with 5,000 customer service agents showed that generative AI increased issue resolution by 14 percent per hour and reduced issue handling time by 9 percent.1Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, Generative AI at work, National Bureau of Economic Research working paper number 31161, April 2023. It also reduced agent attrition and requests to speak to a manager by 25 percent. Notably, productivity and service quality improvements were most significant among less-experienced agents, as AI assistance helped them communicate more like highly skilled agents.
Specific operational improvements from generative AI in customer operations include:
- Customer self-service. Generative AI-powered chatbots can provide instant, personalized responses to complex customer inquiries, regardless of language or location. By improving automated channel interactions, generative AI can handle a higher percentage of inquiries, allowing human agents to focus on complex issues. Our research indicates that about half of customer contacts in North American banking, telecom, and utilities are already handled by machines. Generative AI could further reduce human-serviced contacts by up to 50 percent, depending on current automation levels.
- Resolution during initial contact. Generative AI can instantly access customer data, helping human agents resolve issues and answer questions more effectively during the first interaction.
- Reduced response time. Generative AI can assist sales representatives in real-time and recommend next steps, cutting down customer response times.
- Increased sales. By rapidly processing customer data and browsing history, generative AI can identify tailored product suggestions and deals. It also enhances quality assurance and coaching by analyzing customer conversations to identify improvements and coach agents.
We estimate that generative AI in customer care functions could increase productivity by 30 to 45 percent of current functional costs.
Our analysis only captures direct productivity impacts, not the potential knock-on effects on customer satisfaction and retention from improved experiences, including better agent understanding of customer context for personalized assistance.
Marketing and Sales: Enhancing Personalization, Content Creation, and Sales Productivity
Generative AI has quickly become integral to marketing and sales, where text-based communication and scalable personalization are key. It can create personalized messages tailored to individual customer interests, preferences, and behaviors. It also assists in drafting brand advertising, headlines, slogans, social media posts, and product descriptions.
Marketing
Implementing generative AI in marketing requires careful consideration. Models trained on public data without safeguards against plagiarism, copyright infringement, and brand recognition risks can violate intellectual property rights. Virtual try-on applications may produce biased representations due to limited or biased training data. Therefore, human oversight is essential for conceptual and strategic thinking specific to each company’s needs.
Operational benefits of generative AI in marketing include:
- Efficient and effective content creation. Generative AI can significantly reduce time spent on ideation and content drafting, saving effort. It also ensures consistency in brand voice, writing style, and format across content. Team collaboration via generative AI integrates ideas into cohesive pieces, enhancing marketing message personalization across customer segments, geographies, and demographics. Mass email campaigns can be instantly translated and tailored with different imagery and messaging. Generative AI’s content versatility can improve customer value, attraction, conversion, and lifetime retention at scales previously unattainable.
- Enhanced data use. Generative AI can help marketing overcome challenges of unstructured, inconsistent, and disconnected data by interpreting text, images, and varying structures. It helps marketers better utilize data like territory performance, customer feedback, and behavior to develop data-informed strategies such as targeted profiles and channel recommendations. These tools identify trends, drivers, and opportunities from unstructured data like social media, news, research, and feedback.
- SEO optimization. Generative AI can improve SEO for marketing and sales by optimizing page titles, image tags, and URLs. It synthesizes SEO tokens, aids content creation, and distributes targeted content.
- Product discovery and search personalization. Generative AI personalizes product discovery and search with multimodal inputs (text, images, speech) and deep customer profile understanding. It leverages user preferences, behavior, and purchase history to suggest relevant products and generate personalized descriptions, improving e-commerce conversion rates for CPG, travel, and retail companies.
We estimate generative AI could increase marketing productivity by 5 to 15 percent of total marketing spending.
Our analysis doesn’t account for knock-on effects beyond direct productivity impacts. Generative AI-enabled synthesis could yield higher-quality data insights, leading to new campaign ideas and better-targeted segments. Marketing functions could reallocate resources to higher-quality owned channel content, potentially reducing external channel and agency spending.
Sales
Generative AI can also transform B2B and B2C sales approaches. Sales use cases include:
- Increased sales probability. Generative AI can identify and prioritize leads by creating comprehensive customer profiles from structured and unstructured data, suggesting actions to enhance client engagement at every touchpoint. For example, it can improve close rates by providing better client preference insights.
- Improved lead development. Generative AI can help nurture leads by synthesizing product sales information and customer profiles, creating discussion scripts with up- and cross-selling points. It can also automate follow-ups and passively nurture leads until clients are ready for human interaction.
Our analysis suggests generative AI implementation could increase sales productivity by approximately 3 to 5 percent of current global sales expenditures.
This may underestimate additional revenue potential. Generative AI’s lead identification and follow-up capabilities could uncover new leads and enable more effective outreach, increasing revenue. Time saved by sales representatives through generative AI can be reinvested in higher-quality customer interactions, further boosting sales success.
Software Engineering: Accelerating Developer Work as a Coding Assistant
Treating computer languages as natural languages opens new avenues for software engineering. Generative AI can be used in pair programming, augmented coding, and training LLMs to generate code from natural language descriptions.
Software engineering is critical across industries, growing as companies embed software in products and services. Digital features in vehicles, like adaptive cruise control and IoT connectivity, significantly contribute to their value.
Our analysis indicates that AI’s direct impact on software engineering productivity could range from 20 to 45 percent of current annual spending. This value comes from reduced time on tasks like initial code drafts, code correction, refactoring, root-cause analysis, and new system designs. By speeding up coding, generative AI shifts needed skills towards code and architecture design. A study found developers using GitHub Copilot completed tasks 56 percent faster.1Peter Cihon et al., The impact of AI on developer productivity: Evidence from GitHub Copilot, Cornell University arXiv software engineering working paper, arXiv:2302.06590, February 13, 2023. McKinsey internal studies also showed teams trained on generative AI tools rapidly reduced code generation and refactoring times, with engineers reporting improved work experience, happiness, flow, and fulfillment.
Our analysis did not account for quality improvements and productivity boosts from enhanced code or IT architecture, which can improve the entire IT value chain. However, IT architecture quality still largely depends on human architects, not just the initial drafts generative AI currently provides.
Major tech companies are already offering generative AI for software engineering, including GitHub Copilot (integrated with OpenAI’s GPT-4) and Replit, used by over 20 million coders.2Michael Nuñez, “Google and Replit join forces to challenge Microsoft in coding tools,” VentureBeat, March 28, 2023.
Product R&D: Reducing Research and Design Time, Improving Simulation and Testing
Generative AI’s potential in R&D may be less recognized, but our research indicates it could boost productivity by 10 to 15 percent of overall R&D costs.
Life sciences and chemical industries are using generative AI foundation models in R&D for generative design. These models can generate candidate molecules, accelerating new drug and material development. Entos, a biotech company, uses generative AI with automated synthetic development tools for small-molecule therapeutics. These principles apply to designing various products, including physical products and electrical circuits.
While other generative design techniques using “traditional” machine learning have unlocked R&D potential, their costs and data demands can limit application. Generative AI’s pretrained foundation models, or fine-tuned models, have broader applications than task-specific models, accelerating time to market and expanding product types for generative design. However, current foundation models cannot yet design products across all industries.
Besides productivity gains from rapid design candidate generation, generative design can improve designs themselves. Operational improvements include:
- Enhanced design. Generative AI helps product designers reduce costs by optimizing material selection and use. It also optimizes designs for manufacturing, reducing logistics and production costs.
- Improved product testing and quality. Generative AI in generative design can improve product quality, attractiveness, and market appeal. It accelerates complex system testing and customer trial phases by drafting scenarios and profiling testing candidates.
We also identified a new non-generative AI use case in R&D: deep learning surrogates, which have grown since our earlier research and can pair with generative AI for greater benefits. Integration requires specific solutions, but the value could be significant as deep learning surrogates can accelerate testing of generative AI-proposed designs.
While we estimated direct generative AI impacts on R&D, we did not estimate its potential to create entirely new product categories. These innovations can drive step changes in individual company performance and overall economic growth.
Industry Impacts
Across 63 use cases, generative AI could generate $2.6 trillion to $4.4 trillion in value across industries. The precise impact varies based on function mix, industry importance, and revenue scale (Exhibit 4).
For example, generative AI could add roughly $310 billion in value to retail (including auto dealerships) by improving marketing and customer interactions. In high tech, most value comes from faster software development (Exhibit 5).
In banking, generative AI can enhance existing AI efficiencies by handling lower-value risk management tasks like reporting, regulatory monitoring, and data collection. In life sciences, it is set to significantly contribute to drug discovery and development.
Detailed industry analyses follow.
Generative AI Supports Retail and Consumer Packaged Goods Value Drivers
Generative AI could increase retail and consumer packaged goods (CPG) productivity by 1.2 to 2.0 percent of annual revenues, or $400 billion to $660 billion.1Vehicular retail is included as part of our overall retail analysis. It can streamline customer service, marketing and sales, and inventory and supply chain management. Technology has been vital in retail and CPG for decades. Traditional AI and analytics have helped manage data across SKUs, supply chains, and product categories. These customer-facing industries offer generative AI opportunities to complement existing AI. For example, personalization can optimize marketing and sales activities already using AI. Generative AI’s data management excels for AI-driven pricing tools. Applying generative AI here integrates applications enterprise-wide.
Generative AI at Work in Retail and CPG
Reinventing Customer Interaction
Consumers increasingly seek customization, from clothing to shopping experiences. Generative AI enhances this. Stitch Fix uses algorithms to suggest styles and has experimented with DALL·E to visualize products based on preferences for color, fabric, and style. Stylists can visualize clothing based on consumer preferences and find similar items in Stitch Fix’s inventory using text-to-image generation.
Retailers can create next-gen shopping experiences, a competitive advantage in an era of natural-language interfaces for product selection. For example, generative AI can improve ingredient selection and meal preparation—imagine a chatbot pulling popular recipe tips. Customer value management enhances with personalized marketing campaigns via chatbots. These human-like conversations about products can boost customer satisfaction, traffic, and brand loyalty. Generative AI offers cross-selling, upselling, insight collection, and increased customer base, revenue, and marketing ROI.
Accelerating Value Creation in Key Areas
Generative AI tools facilitate marketing and sales copywriting, creative marketing brainstorming, consumer research, and content analysis and creation. Improved writing and visuals can increase awareness and sales conversion.
Rapid Resolution and Enhanced Insights in Customer Care
E-commerce growth elevates consumer interaction importance. Retailers can combine existing AI with generative AI to enhance chatbots, enabling them to better mimic human agents—responding to queries, tracking orders, offering discounts, and upselling. Automating repetitive tasks allows human agents to handle complex issues and gather context.
Disruptive and Creative Innovation
Generative AI tools enhance product development by rapidly creating new designs digitally. Designers can generate packaging designs from scratch or create variations. Text-to-video generation is a rapidly developing potential addition.
Factors for Retail and CPG Organizations
Retail and CPG executives integrating generative AI should consider factors affecting value capture:
- External inference. Increased need to understand if generated content is fact-based or inferred, requiring new quality control.
- Adversarial attacks. Foundation models are targets for hackers, increasing security and privacy vulnerability.
To address these, retail and CPG companies need strategic human oversight and prioritize security and privacy. New quality checks are needed for processes previously human-handled, like customer rep emails, and more detailed checks for AI-assisted processes like product design.
Banks Could Realize Significant Value
Generative AI could significantly impact banking, increasing productivity by 2.8 to 4.7 percent of annual revenues, or $200 billion to $340 billion. It can also enhance customer satisfaction, decision-making, employee experience, and reduce risks through better fraud and risk monitoring.
Banking, a knowledge and tech-driven industry, has already benefited from AI in marketing and customer operations.1“[Building the AI bank of the future](/industries/financial-services/our-insights/building-the-ai-bank-of-the-future),” McKinsey, May 2021. Generative AI applications can deliver additional benefits, especially with text modalities prevalent in regulations and programming, and the industry’s customer-facing nature.2[McKinsey’s Global Banking Annual Review](/industries/financial-services/our-insights/global-banking-annual-review-archive), December 1, 2022.
Several characteristics position banking for generative AI integration:
- Sustained digitization with legacy IT systems. Banks have invested in tech for decades, accumulating technical debt and complex IT architecture.3Akhil Babbar, Raghavan Janardhanan, Remy Paternoster, and Henning Soller, “[Why most digital banking transformations fail—and how to flip the odds](/capabilities/mckinsey-digital/our-insights/tech-forward/why-most-digital-banking-transformations-fail-and-how-to-flip-the-odds),” McKinsey, April 11, 2023.
- Large customer-facing workforces. Banking relies on service representatives like call center agents and wealth management advisors.
- Stringent regulatory environment. Banking is heavily regulated, with substantial risk, compliance, and legal needs.
- White-collar industry. Generative AI’s impact can span the organization, assisting employees with emails, presentations, and tasks.
Generative AI at Work in Banking
Banks are realizing generative AI’s potential in front lines and software. Early adopters use solutions like ChatGPT and industry-specific tools, mainly for software and knowledge applications. Three use cases highlight its value:
Virtual Expert to Augment Employee Performance
A generative AI bot trained on proprietary knowledge (policies, research, customer interaction) can provide always-on, deep technical support. Frontline spending is mostly for validating offers and client interaction, but data access can improve customer experience. The technology can monitor industries and clients and send alerts on semantic queries from public sources. Morgan Stanley is building a GPT-4 AI assistant to help wealth managers quickly find and synthesize information from a massive knowledge base.4Hugh Son, “Morgan Stanley is testing an OpenAI-powered chatbot for its 16,000 financial advisors,” CNBC, March 14, 2023. This model combines search and content creation for tailored client information.
A European bank used generative AI to develop an ESG virtual expert, synthesizing unstructured information from long documents. The model answers complex questions based on prompts, identifies sources, and extracts information from images and tables.
Generative AI can reduce back-office costs. Customer-facing chatbots can assess user requests and select the best service expert based on topic, difficulty, and customer type. Generative AI assistants can provide instant access to product guides and policies for quick customer request resolution.
Code Acceleration to Reduce Tech Debt and Deliver Software Faster
Generative AI tools aid software development in four areas: code drafting from context (code or natural language), automatic code testing, legacy framework integration and migration optimization, and code review for defects and inefficiencies. This results in robust, effective code.
Production of Tailored Content at Scale
Generative AI tools can streamline content generation using existing documents and datasets. They can create personalized marketing and sales content for client profiles and histories, and multiple A/B testing alternatives. Generative AI can also automatically produce model documentation, identify missing documentation, and scan regulatory updates for relevant alerts.
Factors for Banks to Consider
Banks integrating generative AI should consider:
- Process regulation levels. Varying from unregulated customer service to heavily regulated credit risk scoring.
- End-user type. Varying expectations and familiarity with generative AI—employees vs. high-net-worth clients.
- Intended automation level. AI agents could act autonomously or as copilots, providing real-time suggestions.
- Data constraints. Public data like annual reports can be widely available, but customer and internal data need identifiable detail limits.
Pharmaceuticals and Medical Products Could See Value Chain Benefits
Generative AI could significantly impact pharmaceuticals and medical products—2.6 to 4.5 percent of annual revenues, or $60 billion to $110 billion annually. This reflects the resource-intensive drug discovery process. Pharma companies spend about 20 percent of revenues on R&D,1Research and development in the pharmaceutical industry, Congressional Budget Office, April 2021. and new drug development averages 10 to 15 years. Improving R&D speed and quality generates substantial value. Lead identification, a drug discovery step, can take months even with “traditional” deep learning. Foundation models and generative AI can complete this in weeks.
Generative AI at Work in Pharmaceuticals and Medical Products
Drug discovery narrows potential compounds to those effective for specific conditions. Generative AI’s data processing and modeling accelerates output across use cases:
Improve Automation of Preliminary Screening
In lead identification, scientists can use foundation models to automate preliminary chemical screening for effects on drug targets. Thousands of cell cultures are tested and paired with experiment images. Using off-the-shelf foundation models, researchers can cluster similar images more precisely than with traditional models, selecting promising chemicals for lead optimization.
Enhance Indication Finding
Indication finding identifies and prioritizes new indications—diseases, symptoms, or circumstances justifying medication or treatment use. Indications are based on patient clinical history and records, prioritized by similarity to established indications.
Researchers map patient cohort clinical events and medical histories (diagnoses, medications, procedures) from real-world data. Using foundation models, they quantify clinical events, establish relationships, and measure similarity to evidence-backed indications. This yields a short list of indications with better clinical trial success probability due to accurate matching to patient groups.
Pharma companies using this approach report high clinical trial success rates for top indications recommended by foundation models for tested drugs. This smooth progression to Phase 3 trials significantly accelerates drug development.
Factors for Pharmaceuticals and Medical Products Organizations
Pharma executives integrating generative AI should be aware of factors limiting benefit capture:
- Need for human in the loop. New quality checks may be needed for processes shifting from human to AI, like emails, or more detailed checks for AI-assisted processes like drug discovery. Verifying if generated content is fact or inference elevates quality control needs.
- Explainability. Lack of transparency in content origins and data traceability can hinder model updates and risk scans. A generative AI solution synthesizing scientific literature may not point to specific articles leading to treatment popularity inference. Technology can “hallucinate” incorrect responses. Systems need to point to sources and involve human checking.
- Privacy considerations. Generative AI use of clinical images and medical records could increase protected health information leak risks, violating patient privacy regulations.
Work and Productivity Implications
Technology has been reshaping work for decades. Machines have given workers “superpowers,” like industrial machines for physical tasks and computers for complex calculations.
Technology augments work by automating activities workers would otherwise do. Generative AI application in the workplace may follow this pattern, though the affected activities and occupations will likely differ from older technologies.
The McKinsey Global Institute began analyzing technological automation impact on work activities and adoption scenarios in 2017. We estimated then that workers spent half their time on activities technically automatable by existing technology. We also modeled adoption scenarios and their impact on global work activities.
About the Research
This analysis builds on our 2017 methodology. We examined the US Bureau of Labor Statistics O*Net breakdown of 850 occupations into 2,100 work activities. We scored each activity’s capability needs against 18 automation-potential capabilities.
We surveyed automation experts for technology performance against each capability and its future advancement. This year, we updated cognitive, language, and social and emotional capability assessments based on generative AI expert surveys.
Based on technical automation potential assessments for each activity over time, we modeled global work automation adoption scenarios. First, we estimated solution implementation time for each activity once technology met capability needs. Second, we estimated initial and declining technology costs based on history. We modeled automation adoption for an activity in an occupation in a country (47 countries, >80% global workforce) when automation cost parity with human labor is reached.
Based on historical technology analysis, we modeled 8 to 27 year adoption timelines from adoption start to plateau using S-curves. This range accounts for factors like regulation, investment, and management decisions affecting adoption pace.
Modeled scenarios create a time range for potential automation pace. “Earliest” scenarios maximize automation development and adoption, “latest” scenarios minimize it. Reality likely falls in between.
The analyses in this paper incorporate generative AI’s potential impact on current work activities. Generative AI’s new capabilities, combined with previous technologies, could accelerate technical automation potential and technology adoption, augmenting workforce capabilities. They could also impact knowledge workers’ activities not expected to shift until later (see sidebar “About the research”).
Automation Potential Accelerated, Adoption to Lag
Generative AI advancements mean technology performance is now expected to match median human and reach top-quartile human performance sooner across capabilities (Exhibit 6). For example, MGI previously estimated 2027 earliest for median human natural language understanding in technology, but this analysis revises it to 2023.
These reassessments due to generative AI have increased the total percentage of theoretically automatable hours from about 50 percent to 60–70 percent. The technical potential curve is steep due to generative AI’s natural language capability acceleration.
The time range between early and late scenarios has compressed from 2017 expert assessments, reflecting greater confidence in technological capability advancements by certain periods (Exhibit 7).
Our adoption scenario analysis accounts for technology integration time into automation solutions, technology costs vs. human labor costs in occupations and countries, and technology diffusion time. With accelerated technical automation potential from generative AI, adoption scenarios are also accelerated. These scenarios encompass a wide range of outcomes due to varying investment, deployment, and regulation decisions. However, they indicate how workers’ daily activities may shift (Exhibit 8).
Consider postsecondary English language and literature teachers as an example. Their activities include test preparation and student work evaluation. Generative AI’s enhanced natural language capabilities can handle more of these tasks, initially creating drafts edited by teachers, and eventually requiring less human editing. This could free teacher time for class discussions or tutoring.
Our previous adoption scenarios suggested 50 percent of 2016 work activities would be automated between 2035 and 2070, with a midpoint around 2053. Updated scenarios, accounting for generative AI, model 50 percent automation of 2023 work activities between 2030 and 2060, with a midpoint of 2045—roughly a decade earlier.6The comparison is not exact because the composition of work activities between 2016 and 2023 has changed; for example, some automation has occurred during that time period.
Adoption is also likely faster in developed countries with higher wages, making automation economically feasible sooner. Even with high automation potential, automation costs must be compared to human wages. In countries like China, India, and Mexico with lower wage rates, automation adoption is modeled to be slower (Exhibit 9).
Generative AI’s Potential Impact on Knowledge Work
Previous automation generations were effective at automating data management tasks. Generative AI’s natural language capabilities slightly increase automation potential here, but its impact on physical work is less changed, as its capabilities are designed for cognitive tasks.
Therefore, generative AI is likely to most impact knowledge work, particularly decision-making and collaboration, previously having the lowest automation potential (Exhibit 10). Our estimate of technical potential to automate expertise application jumped 34 percentage points, and management and talent development increased from 16 percent in 2017 to 49 percent in 2023.
Generative AI’s natural language understanding and use for various activities explain the steep automation potential rise. About 40 percent of worker activities require at least median human natural language understanding.
Consequently, communication, supervision, documentation, and interpersonal interaction activities are potentially automatable by generative AI, accelerating work transformation in education and technology occupations, previously expected to automate later (Exhibit 11).
Labor economists often note automation technologies impact lower-skilled workers most, measured by education. Generative AI has the opposite pattern—it will likely incrementally impact more-educated workers by automating some of their activities (Exhibit 12).
Another interpretation is that generative AI will challenge multiyear degree credentials as skill indicators. Others advocate for a skills-based approach to workforce development for equitable, efficient training and matching systems.7A more skills-based approach to workforce development predates the emergence of generative AI. Generative AI could still be skill-biased technological change, but with a different, more granular skill description more likely to be replaced than complemented by machine activities.
Previous automation generations often impacted middle-income occupations most. Automation justification for lower-wage occupations is harder due to lower human labor costs. Some lower-wage occupation tasks are technically hard to automate (fabric manipulation, delicate fruit picking). Some labor economists noted a “hollowing out of the middle,” and our previous models suggested automation would likely impact lower-middle-income quintiles most mid-term.
However, generative AI’s impact is likely to most transform higher-wage knowledge worker work due to technical automation potential advances in their activities, previously considered relatively automation-immune (Exhibit 13).
Generative AI Could Propel Higher Productivity Growth
Global economic growth was slower from 2012 to 2022 than the prior two decades.8Global economic prospects, World Bank, January 2023. While COVID-19 was a factor, long-term challenges—declining birth rates, aging populations—are ongoing growth obstacles.
Declining employment is among these obstacles. Global worker growth slowed from 2.5 percent in 1972–82 to 0.8 percent in 2012–22, largely due to aging. Many large countries already have workforce decline.9Yaron Shamir, “Three factors contributing to fewer people in the workforce,” Forbes, April 7, 2022. Productivity, output relative to input, was the main economic growth engine from 1992 to 2022 (Exhibit 14). However, productivity growth has slowed with employment growth, confounding economists and policymakers.10“The U.S. productivity slowdown: an economy-wide and industry-level analysis,” Monthly Labor Review, US Bureau of Labor Statistics, April 2021; Kweilin Ellingrud, “[Turning around the productivity slowdown](/mgi/overview/in-the-news/turning-around-the-productivity-slowdown),” McKinsey Global Institute, September 13, 2022.
Generative AI and other technology deployment could accelerate productivity growth, partially offsetting declining employment growth and enabling overall economic growth. Our estimates suggest technology-enabled work activity automation could provide a 0.5 to 3.4 percent annual global productivity boost from 2023 to 2040, with generative AI contributing 0.1 to 0.6 percentage points—but only if affected individuals shift to other work activities at least matching 2022 productivity (Exhibit 15). Some workers will stay in occupations with activity shifts; others will need to change occupations.
Considerations for Business and Society
History shows new technologies reshape societies. AI has already changed how we live and work, like phone understanding and email drafting. Mostly, AI has been behind the scenes, optimizing processes or recommending products. Generative AI’s rapid development significantly augments AI’s overall impact, generating trillions in value and transforming work nature.
However, the technology also brings new challenges. Stakeholders must act quickly to address opportunities and risks given generative AI’s rapid adoption. Risks already surfaced include content intellectual property infringement (“plagiarism” in training data), answer accuracy and explainability, and content fairness or bias (harmful stereotypes).
Using Generative AI Responsibly
Generative AI poses various risks. Stakeholders should address these from the start.
Fairness: Models may generate algorithmic bias from imperfect training data or developer decisions.
Intellectual property (IP): Training data and model outputs can create IP risks, including copyright, trademark, patent, or legal material infringement. Even using provider tools, organizations need to understand training data and tool output use.
Privacy: Privacy concerns arise if user input becomes identifiable in model outputs. Generative AI can create and spread malicious content like disinformation, deepfakes, and hate speech.
Security: Bad actors can use generative AI to accelerate and enhance cyberattacks. It can be manipulated to give malicious outputs. Prompt injection can trick models into unintended outputs.
Explainability: Generative AI relies on neural networks with billions of parameters, challenging answer explainability.
Reliability: Models can give different answers to same prompts, hindering output accuracy and reliability assessment.
Organizational impact: Generative AI may significantly affect the workforce, with disproportionately negative impacts on specific groups and communities.
Social and environmental impact: Foundation model development and training can have negative social and environmental consequences, including increased carbon emissions (large language model training can emit ~315 tons of CO2).1Ananya Ganesh, Andrew McCallum, and Emma Strubell, “Energy and policy considerations for deep learning in NLP,” Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, June 5, 2019.
Economic challenges also exist: workforce transitions are considerable. In midpoint adoption scenarios, about a quarter to a third of work activities could change in the next decade. The task is to manage positives and negatives simultaneously (see sidebar “Using generative AI responsibly”). Critical questions to address while balancing enthusiasm and challenges:
Companies and Business Leaders
How can companies quickly capture potential value while managing generative AI risks?
How will occupation and skill mixes be transformed by generative AI and AI in coming years? How will companies enable transitions in hiring, retraining, and HR?
Do companies have a role in preventing “negative use cases” harming society?
How can businesses transparently share scaling generative AI experiences across industries, governments, and society?
Policy Makers
What will the future of work look like in terms of occupations and skills? What does this mean for workforce planning?
How can workers be supported during activity shifts? What retraining programs? What incentives for private company human capital investment? Are earn-while-you-learn programs like apprenticeships possible for retraining while maintaining income?
What steps can policymakers take to prevent harmful generative AI use?
Can new policies be developed or existing policies amended to ensure human-centric AI development and deployment with human oversight, diverse perspectives, and societal values?
Individuals as Workers, Consumers, and Citizens
How concerned should individuals be about generative AI? While companies assess bottom-line impacts, where can citizens find unbiased information about life and livelihood impacts?
How can workers and consumers balance generative AI conveniences with workplace impacts?
Can citizens have a voice in shaping generative AI deployment and integration into their lives?
Technological innovation inspires awe and concern. When innovation seems fully formed and widespread overnight, both responses amplify. Generative AI’s fall 2022 arrival is the latest example due to rapid adoption and the ensuing scramble to deploy, integrate, and experiment.
We are at the start of understanding this technology’s power, reach, and capabilities. If the past eight months are a guide, the next years will be a roller-coaster of rapid innovation and breakthroughs forcing recalibration of AI’s work and life impact. It is important to understand and anticipate this phenomenon. Given generative AI’s deployment speed, digital transformation and labor force reskilling are critical.
These tools can create enormous global economic value when adapting to and mitigating climate change is costly. They also have potential to be more destabilizing than previous AI generations. They are capable of language, a fundamental human ability for expertise and knowledge work, but also a skill used for harm, misunderstanding, obscuring truth, and inciting violence.
We hope this research contributes to understanding generative AI’s capacity to add value and fuel economic growth, and its potential to transform work and societal purpose. Companies, policymakers, consumers, and citizens can work together to ensure generative AI delivers promised value while limiting potential for disruption. The time to act is now.11The research, analysis, and writing in this report was entirely done by humans.
