Technology Trends Outlook 2022: Applied AI
The McKinsey Technology Trends Outlook 2022 report highlights the rapid growth and widespread adoption of applied AI—encompassing machine learning, computer vision, NLP, deep reinforcement learning, and knowledge graphs—to automate and enhance business functions such as risk management, service operations, and product development, with global AI adoption rising from 50% in 2020 to 56% in 2021, driven by increased financial investment, easier access, faster training speeds, and a surge in AI-related patents.
McKinsey & Company
McKinsey
Technology Trends Outlook 2022
Applied AI
August 2022
What is the trend about?
Applied AI uses intelligent application to solve classification, prediction, and control problems to automate, add, or augment real-world business use cases. As AI technologies rapidly push new frontiers of innovation, business adoption continues to grow across use cases
Selected AI technologies<sup>1</sup>
Foundational methods of AI
Machine learning (ML)
Computer vision
Natural-language processing (NLP)
Deep reinforcement learning
Knowledge graphs
Selected use cases<sup>2</sup>
Applications of AI at work
Risk management
Service operations optimization
Product and/or service development
Why should leaders pay attention?
AI adoption has continually increased, enabled by its financial investment and development for easier access<sup>1</sup>
AI adoption by organizations, 2020–21, %
| Category | 2021 (%) | 2020 (%) |
|---|---|---|
| All geographies | 56 | 50 |
| Developed Asia–Pacific | 64 | 61 |
| Developing markets | 57 | 45 |
| Europe | 51 | 47 |
| India | 65 | 57 |
| Latin America | 47 | 41 |
| North America | 55 | 51 |
<sup>1</sup>For details about easing ML development and integration, see “Industrializing machine learning,” McKinsey Technology Trends Outlook 2022, McKinsey, August 2022. <sup>2</sup>Including China, Middle East, and North Africa.
Source: Daniel Zhang et al., The AI Index 2022 annual report, Stanford University, Mar 2022; “The state of AI in 2021,” McKinsey, Dec 8, 2021
Global 56% expansion of AI Share of respondents to a 2021 Easier and more 94.4% Rapidly growing 30× innovation
global survey who said their organizations were adopting AI (up 50% from 2020)
affordable AI implementation
Improvement in training speed for AI models since 2018
Relative number of patents filed in 2021 vs. 2015 (compound annual growth rate of 76.9%)
Investment growth and intensified efforts
$93.5 billion
Private investment in AI-related companies in 2021, accompanied by higher concentration of efforts (doubling vs. 2020)
Why should leaders pay attention? (continued)
The potential value at stake from AI is $10 to $15 trillion ...
Global annual potential, forecast
Value at stake, $ trillion
| Category | Value at stake, $ trillion |
|---|---|
| Service operations | 0.4–0.6 |
| Product development | 0.3 |
| Strategy and corporate finance | 0.2 |
| Finance and IT | 0.2 |
| HR | 0.1–0.2 |
| Risk | 0.5–0.9 |
| Marketing and sales | 3.3–6.0 |
| Supply chain management and manufacturing | 3.6–5.6 |
| Other operations | 0.3–0.4 |
Source: McKinsey Global Institute, 2018 Source: S&P Global, Oct 2020; McKinsey Analytics Quotient data set
… and leaders adopting AI exhibit stronger financial performance
AI maturity and financial performanceOthers Analytics leaders<sup>1</sup> CAGR, %
5-year total return to
shareholders (TRS)<sup>3</sup>
<sup>1</sup>AI leaders are defined as the top quintile of companies that have that taken the McKinsey Analytics Quotient (AQ) assessment.
<sup>2</sup>Includes revenue through fiscal year 2019; during this time, the 5-year revenue CAGR of the S&P 500 index was 4.1%.
<sup>3</sup>Includes TSR through FY 2019; during this time, the 5-year TSR CAGR of the S&P 500 index was 11.7%.
Why should leaders pay attention? (continued)
| Service operations | Revenue increase from AI adoption, by function % of respondents1 34 | Revenue increase from AI adoption, by function % of respondents1 16 | Revenue increase from AI adoption, by function % of respondents1 15 | Revenue increase from AI adoption, by function % of respondents1 65 | Cost decrease from AI adoption, by function % of respondents1 12 | Cost decrease from AI adoption, by function % of respondents1 24 | Cost decrease from AI adoption, by function % of respondents1 51 | Cost decrease from AI adoption, by function % of respondents1 87 |
|---|---|---|---|---|---|---|---|---|
| Manufacturing | 38 | 15 | 10 | 63 | 23 | 27 | 37 | 87 |
| Human resources | 30 | 18 | 15 | 63 | 20 | 26 | 40 | 86 |
| Marketing and sales | 38 | 25 | 11 | 74 | 21 | 35 | 27 | 83 |
| Risk | 26 | 25 | 13 | 64 | 17 | 20 | 41 | 78 |
| Supply chain management | 27 | 15 | 12 | 54 | 15 | 27 | 36 | 78 |
| Product and/or service development | 30 | 25 | 15 | 70 | 22 | 24 | 23 | 69 |
| Strategy and corporate finance | 33 | 32 | 2 | 67 | 10 | 28 | 30 | 68 |
| Average across all activities | 33 | 21 | 13 | 67 | 18 | 28 | 33 | 79 |
<sup>1</sup>Earnings before interest and taxes.
Source: “The state of AI in 2021,” McKinsey, Dec 8, 2021 (for FY 2020)
Across business functions, AI has already made notable financial impact [Image: Image54]
27%
Share of respondents who report at least 5% of EBIT<sup>1</sup> being attributable to AI
67%
Average share of respondents reporting a revenue increase via AI adoption
79%
Average share of respondents reporting a cost decrease via AI adoption
What are the most noteworthy technologies?
AI involves machines exhibiting intelligence,<sup>1</sup> encompassing various interconnected fields of technology<sup>2</sup>
Artificial intelligence
Machine learning
1 Machine learning
Computer vision
Naturallanguage processing
Deep Reinforcement learning
Knowledge graphs
Description
ML: Subfield of AI that uses statistical methods to learn from data
Example use case
Schedule optimization
Computer vision: Subfield of ML using visual data, such as images, videos, and 3-D signals, extracting complex information and gaining rich interpretations
Facial recognition as biometrics
NLP: Subfield of ML that involves processing, generating, and understanding language-based data, such as written text and spoken word
Speech recognition in a virtual voice assistant
Deep reinforcement learning: Combination of deep learning and reinforcement learning, in which an agent makes decisions within an uncertain environment using complex algorithms inspired by brain neural networks
4 Deep reinforcement learning: Combination of deep
Planning robotic-arm motion for the manufacturing line
Knowledge graphs: Collection of data points structured into a network to show complex relationships among themselves
Social-network analysis
<sup>1</sup>AI is nonprogrammatic intelligence exhibited by machines, in which they perform cognitive functions often associated with human minds. Cognitive functions
include all aspects of perceiving, reasoning, learning, and problem solving.
<u><sup>2</sup>Technologies are not exhaustive and are examples that are at the frontiers of innovation and cut across industries.</u>
What industries and functions are leading in the adoption of AI applications?
AI adoption by industry and function, 2021 % of respondents
| Human resources | Manufacturing | Marketing and sales | Product or service development | Risk | Service operations | Strategy and corporate finance | Supply chain management | |
|---|---|---|---|---|---|---|---|---|
| All industries | 9 | 12 | 20 | 23 | 13 | 25 | 9 | 13 |
| Business, legal, and professional | 11 | 26 | 20 | 15 | 4 | 18 | 6 | 17 |
| Consumer goods/retail | 14 | 8 | 28 | 15 | 13 | 26 | 8 | 13 |
| Financial services | 2 | 18 | 22 | 17 | 1 | 15 | 4 | 18 |
| Healthcare systems | 10 | 4 | 24 | 20 | 32 | 40 | 13 | 8 |
| Pharma and medical products | 9 | 11 | 14 | 29 | 13 | 17 | 12 | 9 |
| High tech/ telecom | 12 | 11 | 28 | 45 | 16 | 34 | 10 | 16 |
Technology-centric industries are leading adoption by businesses
industries are leading
Product and service development, service operations, and marketing and sales are the business functions leading adoption of AI
What industries are most affected by the trend?
A diverse set of stakeholders across all industries are experiencing the impact from applied AI, which can include disruption in value chains, better financial outcomes, and improved operations
Industry affected<sup>1</sup>
| Industry affected | Example impact from the trend |
|---|---|
| Information technology and electronics | Pervasive use across the tech industry and constituent sectors, such as software, hardware, and electronic devices (eg, use of generative AI models to create 3-D visuals for software simulations) |
| Telecommunications | Programming AI models to identify recurring customer concerns and deliver solutions before complaints arise |
| Pharmaceuticals and medical products | Exploring relationships across different medical treatments and their combined outcomes for the discovery of new drugs |
| Aerospace and defense | Aiding the design process (eg, through visual simulations of aircraft performance under different conditions) as well as for security and risk mitigation processes |
| Healthcare systems and services | Enhancing healthcare services through functions like automated pathology recognition and diagnosis decision support |
| Financial services | Supporting risk management in financial services, eg, detecting credit card fraud to reduce incidents of loss |
| Retail and consumer packaged goods | Boosting sales by using ML to analyze huge sets of purchasing data, discern patterns, and give shoppers customized recommendations |
| Education | Improving personalized learning based on students' progress |
| Aviation, travel, and logistics | Leveraging multimodal fusion, enabled by AI, to combine inputs from various sensors that can help operate autonomous vehicles2 |
| Not exhaustive and focused on industries where AI has widespread applications with mature adoption. |
<sup>1</sup>Not exhaustive and focused on industries where AI has widespread applications with mature adoption. <sup>2</sup>For more, see “Future of mobility,” McKinsey Technology Trends Outlook 2022, McKinsey, Aug 2022.
What industries are most affected by the trend? (continued)
A diverse set of stakeholders across all industries are experiencing implications from applied AI, which can include disruption in value chains, better financial outcomes , and improved operations
| Industry affected1 | Example impact from the trend |
|---|---|
| Agriculture | Enabling process optimization through capabilities like productivity forecasting and driverless tractor applications |
| Automotive and assembly | Automation of quality testing and manufacturing/assembly processes |
| Chemicals | Optimizing chemical development and production cycles by recognizing molecules, generating chemical compound formulas, and analyzing chemical mixtures |
| Construction and building materials | Using autonomous machinery and robots, computer-vision enhanced safety procedures, and 3-D design optimization software |
| Electric power, natural gas and utilities | Optimizing energy production and scheduling, detecting equipment defects early to minimize downtime, and analyzing consumer energy use data to inform personalized recommendations |
| Metals and mining | Increasing worksite process efficiencies and aiding the development of digital twins that can generate visualizations and models of remote sites |
| Oil and gas | Exploration of site through computer vision to assess the value of holdings and use AI/ML to customize drilling plans for geologically-complex areas and forecast demand |
| Public and social sectors | Leveraging AI/ML to expedite delivery of key services (eg, use of NLP for tax FAQ handling); additionally, AI/ML can be used as a tool to help in audit mechanisms to ensure the proper use of resources (eg, predictive tools to help focus tax auditing) |
| Real estate | Providing personalized customer property recommendations, performing market analyses to help developers manage risk and price volatility, as well as optimizing ROI |
What are some use cases for applied AI?
| Use case1 | Technology2 | Function | Relevant industries3 | Description | Benefits4 |
|---|---|---|---|---|---|
| Generate 3-D models | Computer vision; ML Optional: NLP | Product development | Technology; manufacturing; consumer goods; retail | Apply generative techniques that synthesize 3-D visuals based on singular or multimodal instructions. Examples: Models for animation, furniture models, and apparel re-creations | Decrease cost with improved efficiency through quickly generated 3-D models |
| Prioritize dynamically changing tasks | ML; deep reinforcement learning Optional: Computer vision; NLP | Service operations | Any | Optimize changing workflow through multitask learning to prioritize most relevant tasks. Examples: Schedule-planning and project management tools | Decrease cost with improved productivity |
| Fuse multi-modal sensors | Deep reinforcement learning; ML; computer vision. Optional: NLP | Product development | Transportation; retail; healthcare | Utilize various sensor inputs to perform tasks. Examples: Sales checkout for retail; vehicle sensing for autonomous driving | Decrease cost by automating systems requiring sensor input |
| Recommend products to purchase | ML Optional: Knowledge graphs; NLP; computer vision | Product development | Technology; retail finance; healthcare | Predict and suggest potential products relevant to a customer's interests based on prior customer data (individuals or groups). Examples: Online suggestions of products to purchase; movie recommendations | Improve revenue through increased sales via personalized recommendations |
| Detect fraud | ML Optional: Knowledge graphs; NLP | Risk management | Any | Detect fraudulent behaviors to reduce incidents of loss. Examples: Detection of fraudulent credit card purchases and account log-in | Reduce losses through stronger detection of risky behaviors |
<sup>1</sup>List of use cases is nonexhaustive and highlights those that are at the frontier of innovation and/or rapidly gaining adoption across organizations. <sup>2</sup>Technologies typically used to implement the use case. Optional technologies can be applied but depend on the specific task for the use case.
<sup>4</sup>Nonexhaustive benefits, focusing on major benefits to businesses.
What should a leader consider <u>when engaging with the trend?</u>
Benefits
Cost savings: Up to 90% of survey respondents cited cost decreases in 2020
Overall revenue increase: Up to 75% of survey respondents cited revenue increases in 2020<sup>1</sup>
New use cases: New use cases will unlock new business capabilities and opportunities across automation and acceleration
Increased access to AI and ease of implementation: New technologies and practices, such as ML operations and software automation, should make AI more readily available
<sup>1</sup>For more about development of ML systems and tools, see “Industrializing machine learning,” McKinsey Technology Trends Outlook 2022, McKinsey, Aug 2022.
Risks and uncertainties
High up-front investment in talent and resources:
This creates a high barrier to entry related to developing AI and ML workflows for production<sup>1</sup>
Cybersecurity and privacy concerns: Data risks and vulnerabilities are occurring across the technical AI workflow; 55% of survey respondents cite cybersecurity as a leading risk in their business in 2021 and are actively taking steps to mitigate it
Increasing regulation and compliance: New
legislation will affect the development of AI’s direction
AI ethics: Issues include responsibility, equity, fairness, and explainability
What are some topics of debate related to the trend?
Trustworthiness What does it mean to apply trustworthy and responsible AI? 1
Potential risks and concerns increase as AI use cases expand
According to the EU Commission High-Level Expert Group on AI, responsible and trustworthy AI can be defined by abiding laws, incorporating ethics, and implementing technical and social robustness to mitigate potential harm
The commission has developed 7 requirements for AI responsibility and trust: human agency and oversight; societal and environmental well-being; technical robustness and safety; privacy and data governance; transparency; accountability; and diversity, nondiscrimination, and fairness
</>
###### 2Explainability
###### When is AI explainability needed?
AI explainability looks at how well we can understand an AI model. Interest in this field is rising as models
are growing increasingly complex and high-risk use cases (eg, disease diagnosis) are being explored
According to Stanford University Human-Centered Artificial Intelligence (HAI), there are three types of AI:
engineers’ explainability (technically explains how the AI model works), causal explainability (explains
why a model input leads to its output), and trust-inducing explainability (information that people need to
trust and deploy a model)
Depending on the situation, organizations may use one type of explainability, a combination of types, or
all three types (eg, disease risk evaluation looks at all three types)
###### 3 Applications prioritization
###### How might companies better determine which AI application provide the most benefit?
Across industries and organizations, each applications of AI will impact different stakeholders in a unique
way; understanding how AI impacts each stakeholder, the organization, and the ecosystem will be
particularly important for leaders as they decide which AI applications to leverage
Understanding what impact on a use case an AI application will have will be more essential in
prioritization decisions for leaders as they build the capabilities to deploy and monitor AI at scale
###### Other risks
###### What are other areas of risk that are relevant?
According to Stanford HAI, leading areas of risk for organizations include cybersecurity, regulatory
compliance, explainability, individual privacy, organizational reputation, and equity and fairness
While customers, shareholders, and regulators are calling for increased scrutiny on these topics,
subjective topics (eg, privacy, equity, and fairness) are not high strategic priorities within organizations,
as they lack resources and capabilities to fully understand and address these concerns
---
**Additional resources**
Knowledge center
QuantumBlack, AI by McKinsey
Related reading
##### The state of AI in 2021
The AI Index Report: Measuring trends in artificial intelligence
It’s time for businesses to chart a course for reinforcement learning