How Many Companies Use AI?

Current Global Rates of Business AI Adoption

Estimating how many companies use AI requires distinguishing between regular experimentation, employee-level software utilization, and formal enterprise-wide deployment. Across global enterprise surveys, between 70% and 78% of large organizations report using artificial intelligence in at least one business function. However, when examining the broader macroeconomic landscape—including small and medium-sized enterprises (SMEs)—formal production adoption ranges from 5% to 35%, depending on the country, sector, and definition of technology.

Technological adoption accelerated dramatically following the public release of accessible large language models (LLMs) and generative tools in late 2022. Prior to this shift, corporate AI usage was predominantly concentrated in predictive analytics, machine learning (ML) for fraud detection, and automated logistics. Today, corporate adoption encompasses both embedded software features (such as automated summaries in customer relationship management tools) and custom internal infrastructure.

Macroeconomic Adoption Spectrum (Approximate)

[ All Registered Businesses (Including Micro/SMEs) ] ---> ~5% - 15% Core Production Use
[ Mid-Market Enterprises ($50M - $500M Revenue) ]     ---> ~30% - 55% Active Deployment
[ Global Enterprises ($500M+ Revenue) ]           ---> ~70% - 78% Regular Operational Use

Methodological Differences in Measuring AI Adoption

Discrepancies in the reported percentage of businesses that use AI stem from varying survey scopes, definitions of "use," and the structural composition of the surveyed samples.

Enterprise Surveys vs. Representative Economic Data

Most widely cited figures originate from global management consultancies and enterprise software vendors. These studies primarily sample senior technology executives, Fortune 500 leadership, or enterprise clients. Because these entities possess large IT budgets, dedicated research teams, and scalable digital architectures, their adoption metrics skew high (typically between 60% and 80%).

Conversely, national statistical agencies (such as the U.S. Census Bureau and Eurostat) survey representative cross-sections of the entire economy. Because micro-enterprises and non-technical service providers (e.g., local retail, trades, single-location hospitality) make up the vast majority of registered firms, national averages appear significantly lower. In these datasets, formal AI utilization generally tracks below 15%.

Defining "Active Use"

Survey results vary based on how "using AI" is defined:

• Passive / Embedded Utilization: Employees using everyday commercial SaaS tools (e.g., Microsoft 365 Copilot, Google Workspace, Canva, Zoom AI summaries) where AI is integrated by default.
• Ad-Hoc / Employee-Led Use: Staff independently querying public web tools (e.g., ChatGPT, Claude) for drafting emails, debugging code, or summarizing documents without formal corporate procurement.
• Active Pilot / Proof of Concept (PoC): Organizations testing custom models, fine-tuned LLMs, or predictive algorithms in isolated sandbox environments without connecting them to live customer data or core production pipelines.
• Full Production Deployment: Automated, end-to-end integration where machine learning models, computer vision systems, or generative APIs actively process production data, make autonomous decisions, or interface directly with end users.

   Level 0: No AI Usage (Legacy workflows, manual processes)
      │
   Level 1: Passive/Embedded (SaaS auto-features, automated transcribing)
      │
   Level 2: Ad-Hoc Employee Use (Shadow IT, individual prompt engineering)
      │
   Level 3: Proof-of-Concept / Sandboxing (Departmental trials, custom prototypes)
      │
   Level 4: Integrated Production (Connected to data lakes, core APIs, live workflows)
      │
   Level 5: Autonomous / Core Infrastructure (AI-first architectures, continuous training loops)

AI Adoption by Industry and Sector

Adoption rates vary widely across economic sectors. Industries characterized by digital workflows, large data volumes, and high-margin knowledge work lead the market, while asset-heavy, physically constrained, or strictly regulated sectors adopt the technology more deliberately.

High Adoption  [████████████████████] Technology, Software, Telecommunications
               [████████████████    ] Financial Services, Insurance, Asset Management
               [█████████████       ] Marketing, Media, Professional Services
Moderate       [██████████          ] Healthcare Diagnostics, Life Sciences, Pharma
               [████████            ] Retail, E-Commerce, Consumer Goods
Low to Medium  [██████              ] Advanced Manufacturing, Automotive, Logistics
               [████                ] Construction, Agriculture, Extraction, Real Estate

Technology, Software, and Telecommunications

The technology sector exhibits the highest concentration of active deployments, with over 80% of companies utilizing machine learning or generative AI. Common applications include:

• Automated code generation, code review, and synthetic testing.
• Intelligent network traffic routing and predictive server maintenance.
• Product feature enhancement via native natural language interfaces.

Financial Services and Insurance

Banking and financial institutions have used machine learning for decades. The financial sector shows roughly 65% to 75% institutional adoption, focusing on:

• Algorithmic trading and real-time fraud detection engines.
• Automated credit scoring, underwriting, and risk modeling.
• Conversational virtual assistants for tier-one consumer banking support.
• Regulatory compliance scanning and anti-money laundering (AML) transaction monitoring.

Healthcare, Pharmaceuticals, and Life Sciences

Adoption in healthcare is bifurcated between operational administration and clinical applications:

• Administrative Integration (~50%–60%): Automated medical billing, clinical documentation transcription (ambient clinical intelligence), and patient scheduling.
• Clinical & Research (~25%–35%): Deep learning in diagnostic imaging (radiology, oncology), protein folding prediction, targeted drug discovery, and clinical trial cohort selection.

Manufacturing, Supply Chain, and Logistics

Industrial adoption focuses primarily on predictive analytics, physical computer vision, and operational optimization rather than generative text tools:

• Predictive maintenance of machinery via IoT sensor analysis.
• Computer-vision-based automated defect detection on assembly lines.
• Dynamic supply chain forecasting, inventory optimization, and route planning.

Company Size and Geographic Disparities

The Enterprise vs. SME Divide

Organizational scale is one of the strongest predictors of whether a company deploys proprietary or customized AI systems.

Enterprise ($500M+ Revenue)
├── Dedicated Machine Learning & MLOps Teams
├── Custom Data Warehousing (Snowflake, Databricks)
├── Proprietary Fine-Tuning & Self-Hosted Infrastructure
└── Formal AI Governance & Compliance Frameworks

Small-to-Medium Enterprise (SME)
├── Rely on Off-the-Shelf SaaS Embeddings
├── Low/No-Code Integrations (Zapier, OpenAI APIs)
├── Limited or Centralized IT Management
└── Informal or Absent Governance Policies

• Large Enterprises (1,000+ Employees): More than 70% deploy formal AI solutions. They possess the capital required to build dedicated data platforms, hire machine learning engineers, clean unstructured data, and manage regulatory obligations.
• Small and Mid-Sized Businesses (<250 Employees): Formal custom adoption ranges from 15% to 30%. SMEs predominantly access advanced machine learning through third-party platforms (e.g., accounting software with built-in predictive reconciliation or automated marketing platforms) rather than direct API or model engineering.

Geographic Distribution

Global Adoption Landscape
┌──────────────────┬────────────────────────────────────────────────────────┐
│ Region           │ Characteristics & Drivers                              │
├──────────────────┼────────────────────────────────────────────────────────┤
│ North America    │ Leads in total enterprise spending, foundation model   │
│                  │ development, and venture-backed SaaS integration.      │
├──────────────────┼────────────────────────────────────────────────────────┤
│ Asia-Pacific     │ High implementation rates in China, India, and ASEAN;  │
│                  │ focus on automation, manufacturing, and mobile scale.  │
├──────────────────┼────────────────────────────────────────────────────────┤
│ Europe           │ Strong adoption in Northern/Western Europe; higher     │
│                  │ regulatory compliance focus (EU AI Act, GDPR).         │
├──────────────────┼────────────────────────────────────────────────────────┤
│ Latin America &  │ Rapidly growing cloud-based adoption; concentrated in  │
│ MEA              │ fintech, telecom hubs, and sovereign AI initiatives.   │
└──────────────────┴────────────────────────────────────────────────────────┘

• North America: High rates of generative AI and enterprise ML integration, driven by concentrated technology capital, cloud hyperscaler ecosystems, and major software vendor hubs.
• Asia-Pacific: High overall enterprise adoption reported in markets like India and China, heavily weighted toward manufacturing automation, smart logistics, telecommunications, and financial technology.
• Europe: High variation between countries. While Northern European nations (e.g., Sweden, Finland, Denmark) report enterprise adoption rates exceeding 30% to 40%, other regions remain closer to the single-digit EU baseline. European organizations often spend more resources on data privacy, copyright verification, and alignment with the EU Artificial Intelligence Act.
• Latin America and Middle East/Africa: Driven by leapfrogging initiatives in banking, telecommunications, and national modernization agendas (e.g., sovereign AI infrastructure programs in the UAE and Saudi Arabia).

Primary Functional Areas of Deployment

Within companies that use AI, deployment is rarely uniform across the entire organization. Businesses generally target functional units where automation can directly increase revenue or reduce operational overhead.

Percentage of AI-Adopting Companies Deploying in Specific Functions

Marketing & Content Creation  [████████████████████████████████] ~55% - 65%
Customer Service & Support    [████████████████████████████    ] ~50% - 58%
Software Engineering / IT     [██████████████████████          ] ~45% - 52%
Operations & Supply Chain     [████████████████                ] ~30% - 38%
Human Resources & Recruiting  [████████████                    ] ~20% - 26%
Legal, Risk & Compliance      [██████████                      ] ~18% - 22%

1. Marketing, Sales, and Content Operations

Marketing departments exhibit the fastest implementation timelines due to the low technical barrier of generative text and image models:

• Drafting search-engine-optimized content, ad copy, and social campaigns.
• Personalized email outreach and automated lead qualification.
• Image generation, asset variation, and localized translation at scale.

2. Customer Support and Experience

Customer service departments frequently convert conversational AI into a direct cost-reduction mechanism:

• First-line resolution via natural language processing (NLP) chat and voice agents.
• Real-time agent assistance, including automated ticket categorization and retrieval of internal knowledge base articles.
• Post-call sentiment analysis, summarization, and CRM data entry.

3. Software Engineering and IT Management

Software development teams have adopted AI as a core productivity enhancement:

• In-line code completion and automated test script generation (e.g., GitHub Copilot).
• Legacy code refactoring and migration (e.g., COBOL to Java translation).
• Automated infrastructure monitoring, anomaly detection, and synthetic data generation.

4. Human Resources and Corporate Operations

Internal operations deploy machine learning to manage operational tasks:

• Initial resume screening and applicant ranking.
• Internal HR and IT helpdesk automation.
• Automated contract abstraction, document retrieval, and expense auditing.

Operational Barriers Preventing Universal Adoption

While interest in artificial intelligence is broad, the gap between initial experimentation and sustained, enterprise-wide production remains significant. Many organizations encounter operational, technical, and financial barriers when moving beyond the pilot phase.

  [ Experimental Phase ] ───▶ ( Proof-of-Concept ) 
                                      │
                                      ▼
                  ╔═══════════════════════════════════╗
                  ║       THE DEPLOYMENT GAP          ║
                  ║  - Fragmented / Siloed Data       ║
                  ║  - Unclear Return on Investment   ║
                  ║  - Governance & Legal Liabilities ║
                  ║  - Talent & MLOps Shortages       ║
                  ╚═══════════════════════════════════╝
                                      │
                                      ▼
                         [ Scaled Production System ]

Data Infrastructure and Silos

Machine learning models require clean, accessible, and well-labeled data pipelines. Many legacy enterprises operate across fragmented databases, on-premises systems, and unstructured document stores. Without modern data lakehouses, semantic layers, and real-time streaming architectures, implementing advanced AI models often results in unreliable performance or model hallucination.

Demonstrating Return on Investment (ROI)

Training, fine-tuning, and running inference on large neural networks or proprietary foundation models entails high compute, infrastructure, and licensing costs. Organizations frequently struggle to quantify the productivity gains of generative tools relative to their integration expenses, leading some leadership teams to stall projects in the proof-of-concept phase.

Governance, Security, and Intellectual Property Risks

Enterprises must navigate severe operational and regulatory risks before releasing AI systems into production:

• Data Leakage: Accidental exposure of proprietary intellectual property, trade secrets, or protected customer data to third-party public models.
• Regulatory Compliance: Adhering to evolving international frameworks, such as the EU AI Act, which imposes strict risk tiers, transparency mandates, and technical audits for high-risk applications.
• Algorithmic Bias and Hallucination: Defending against biased decision-making in lending or hiring, and preventing inaccurate generative outputs in consumer-facing environments.

Technical Talent Constraints

Building robust, production-grade applications requires multidisciplinary expertise across data engineering, cloud architecture, machine learning operations (MLOps), prompt engineering, and cybersecurity. The market demand for specialized data scientists and machine learning engineers continues to outpace the available workforce, concentrating advanced development in top-tier technology firms and specialized consultancies.

Summary of the Adoption Trajectory

Business AI adoption has transitioned from a niche, research-heavy domain into an essential infrastructure layer. Today, while nearly three out of four large corporations have deployed AI within at least one functional unit, the broader macroeconomic figure across all global businesses remains closer to one in five.

As major enterprise software ecosystems (enterprise resource planning, customer relationship management, and productivity suites) continue to embed machine learning models natively into their core products, the distinction between using software and using AI will gradually disappear. Future adoption metrics will focus less on whether an organization uses AI, and more on how deeply proprietary data, autonomous agents, and automated decision-making are integrated into its operational core.

The short answer

There is no single authoritative number for how many companies use AI. The answer depends on what counts as “use,” which businesses are included, and whether the measurement concerns artificial intelligence generally or generative AI specifically.

The most defensible summary is that a substantial minority to a large majority of surveyed businesses report some form of AI use, depending on the survey definition. Recent international surveys of larger organizations have often found that roughly half to three-quarters report using AI in at least one business function. Surveys designed to represent all businesses, including small firms and businesses that have not adopted advanced software, generally produce lower figures—often single-digit or low-double-digit percentages for active AI use.

These results are not contradictory. A survey of large, digitally mature organizations measures a different population from a government survey of every business, including small retailers, tradespeople, farms, and companies with no dedicated information-technology staff.

Why the percentage varies so much

“Using AI” is not a single, consistently measured activity. It can refer to anything from an algorithm embedded in ordinary business software to an employee occasionally asking a chatbot to draft text. A company may also use AI without describing its software that way.

Common interpretations include:

• Any AI-enabled software: fraud detection, search ranking, recommendation systems, automated translation, demand forecasting, spam filtering, or route optimization.
• AI developed or deployed by the business: an internal machine-learning model, an AI-powered product, or a system connected to company data.
• Purchased AI services: a cloud platform, customer-service tool, recruitment system, marketing application, or enterprise software with an AI feature.
• Generative AI use: tools that produce text, images, audio, video, software code, summaries, or other content.
• Regular operational use: AI integrated into a repeatable business process rather than used experimentally.
• Any employee use: an employee or team uses an AI tool, even if the organization has not formally approved or managed it.

A company counted under the broadest definition may only use a built-in feature supplied by another vendor. Under a narrower definition, it may not count unless AI is formally deployed in production. Consequently, percentages should not be compared unless the definitions, sample, date, and question wording are understood.

Artificial intelligence and generative AI are different measurements

Artificial intelligence is the broader category. It includes systems that classify, predict, optimize, recognize patterns, or make recommendations. Generative AI is a subset that creates new content in response to an input, or “prompt.”

A company can therefore:

1. use traditional AI but not generative AI;
2. use generative AI but not build its own AI systems;
3. use both; or
4. report no formal AI deployment while employees use public AI tools informally.

A survey reporting that 70% of organizations use AI does not necessarily mean that 70% use chatbots, large language models, or image generators. Similarly, a survey about generative AI should not be treated as a measure of all business automation.

What major surveys indicate

Large business surveys provide useful evidence about adoption among organizations that are generally more digitally mature, internationally active, or large enough to be visible to business research firms. Their results commonly show rapid growth in reported AI use, especially since the public release of widely accessible generative-AI tools.

For example, McKinsey’s global surveys have reported a marked increase in the share of respondents saying their organizations use AI in at least one business function. Its 2023 survey reported approximately 55%, while its 2024 survey reported approximately 72%. These figures describe the organizations represented in those surveys and the respondents’ interpretation of “use”; they are not a census of every company in the world. The 2024 result also included generative AI within the broader AI measurement.

Other surveys have produced lower or differently defined figures. IBM’s 2023 Global AI Adoption Index reported that 42% of surveyed enterprise-scale organizations had actively deployed AI, while another group reported that it was exploring AI. “Actively deployed” is a narrower status than “an employee has used an AI tool” and is not directly comparable with a question asking whether AI is used in any function.

Research focused on workers rather than companies has sometimes found even higher rates of individual use. For example, surveys of knowledge workers have reported that many employees use generative-AI tools at work, including in organizations that have not formally adopted them. This measures worker behavior, not the proportion of businesses with a governed, organization-wide AI deployment.

The broad pattern is therefore clear even though the precise percentage is not:

The table is a guide to the structure of the evidence, not a universal range that applies to every country or year.

Why a business-population survey can report a much lower figure

A representative survey of all businesses typically includes many organizations that are not the usual respondents to executive technology surveys. Small companies may have few employees, limited software budgets, little access to technical expertise, and no obvious need to build a machine-learning system. A family-owned shop, a local construction firm, and a multinational bank all count as companies, but their opportunities and incentives for AI adoption are very different.

Official business surveys also tend to ask concrete questions, such as whether the firm used AI during a specified reference period or whether it used a named technology for production, sales, administration, or logistics. These questions may exclude experimentation, informal employee use, and AI features that the respondent does not recognize as AI.

For this reason, a representative survey may report that only a small share of all businesses use AI, while a survey of executives at large organizations reports that most organizations do. Both may be accurate within their respective populations.

A further complication is the denominator. “How many companies use AI?” could mean:

• the percentage of all registered businesses;
• the percentage of businesses with employees;
• the percentage of surveyed companies;
• the percentage of large companies above a particular revenue or employee threshold;
• the percentage of firms in a specific industry; or
• the percentage of organizations that have heard of or tested AI.

These denominators can differ substantially. A percentage based on firms may also count a group of companies equally even though one has two employees and another has hundreds of thousands.

Turning a percentage into a number of companies

The basic calculation is straightforward:

estimated number of AI-using companies = number of companies in the defined population × adoption percentage

For example, if a properly defined population contains 100,000 businesses and a survey estimates that 12% use AI under its stated definition, the implied estimate is:

100,000 × 0.12 = 12,000 businesses

This is an estimate for that population, not a count verified company by company. To use the calculation responsibly, the population and the adoption percentage must refer to the same country, period, business-size range, and definition of AI use.

It would be misleading to multiply a survey of large companies by the total number of businesses worldwide. The survey may overrepresent companies that are already technology-oriented, may exclude microbusinesses, and may use a different interpretation of AI. The resulting number would look precise but would not be statistically meaningful.

There is also no stable global denominator. Business registers differ by country, and some count legal entities while others count enterprises, establishments, or employer firms. A single corporate group can contain many legal companies, while a franchise network may appear as many businesses or as one system depending on the database. Dormant firms, sole proprietorships, and firms without employees may or may not be included.

The sectors where adoption is most visible

AI adoption is uneven across industries because the value of automation depends on data availability, workflow structure, regulatory requirements, and the cost of mistakes.

Higher reported adoption is common in sectors such as:

• information technology and software;
• financial services and insurance;
• professional and technical services;
• telecommunications;
• media and online commerce;
• advanced manufacturing and logistics; and
• larger healthcare or life-science organizations, where deployment is subject to specialized requirements.

These sectors often have large stores of digital data, repeatable processes, and employees whose work is already mediated through software. AI can be integrated into forecasting, document processing, customer support, coding, quality inspection, cybersecurity, or research.

Adoption may be slower or harder to detect in agriculture, construction, hospitality, small retail, personal services, and some parts of traditional manufacturing. That does not mean AI is absent. Businesses in these sectors may use AI indirectly through accounting platforms, advertising systems, payment processors, navigation software, scheduling tools, or equipment supplied by vendors. They may simply not regard those features as an AI deployment.

Industry comparisons should therefore distinguish between direct adoption—a business intentionally selecting or managing an AI system—and embedded adoption, where AI is part of a product or service purchased from someone else.

What “using AI” looks like in practice

Business AI use generally falls into several layers rather than one event of adoption.

Embedded functions in ordinary software

Many companies use AI through software that does not require them to train or operate a model. Examples include email filtering, optical character recognition, automatic transcription, search, recommendations, invoice extraction, and demand estimates. These uses may have been present for years and may not be included when a survey asks specifically about newer generative AI.

Department-level tools

A marketing, sales, customer-service, human-resources, finance, or engineering team may adopt an AI application without the entire company changing its operating model. Examples include drafting campaign material, summarizing calls, answering routine customer questions, reviewing documents, or assisting with code. Department-level use is often what broad surveys mean by use in “at least one function.”

Enterprise deployments

A more mature deployment connects AI to internal data, business systems, permissions, monitoring, and human review. A company may use a model to forecast inventory, prioritize service cases, identify unusual transactions, or support a regulated decision process. This is more consequential than occasional experimentation and normally requires security, privacy, procurement, and governance controls.

AI as a product or core capability

Some firms sell AI-enabled products or make AI central to their business model. This is a much narrower population than companies that merely use an AI feature in a back-office application. Surveys should not treat “develops AI” and “uses AI” as interchangeable categories.

How to judge a claimed adoption statistic

When a source says that a particular percentage of businesses use AI, examine the following points before applying it to a broader question:

1. Date: Adoption has changed quickly, especially for generative AI. A result from one year may not describe another.
2. Geography: A figure for one country or economic region is not automatically a global estimate.
3. Business size: Ask whether the sample includes microbusinesses, small firms, and sole proprietors, or mainly large enterprises.
4. Industry: Technology and finance surveys tend to report higher adoption than economy-wide surveys.
5. Respondent: An executive, an IT manager, and an individual employee may answer different questions about the same company.
6. Definition: “Experimenting,” “using,” “deploying,” “regularly using,” and “using in production” represent different stages.
7. Sampling method: A probability sample designed to represent the business population differs from an opt-in survey of interested technology professionals.
8. Question wording: “Does your organization use AI?” can produce a different result from “Has your organization deployed an AI system in a production business process during the reference period?”
9. Multiple responses and uncertainty: A survey percentage is an estimate with sampling error and possible nonresponse or reporting bias.

A strong source should disclose the population, fieldwork period, sample size, weighting, and operational definition. Even then, survey answers measure reported adoption rather than independently audited usage.

Why reported adoption is not the same as successful adoption

A company may report AI use while a project remains a pilot, is used by only a few employees, or has not produced measurable business value. Conversely, a company may receive substantial AI functionality from a vendor without identifying itself as an AI user.

Adoption statistics therefore do not answer several important questions:

• How many companies use AI every day?
• How many use it in revenue-generating operations?
• How many have achieved cost savings or higher productivity?
• How many have replaced a previous process with AI?
• How many have effective oversight, security, and privacy controls?
• How many use internally developed models rather than third-party tools?

Those are outcome, intensity, and governance questions. They require more detailed research than a single adoption percentage.

Organizations also face practical limitations. AI systems can produce inaccurate or fabricated outputs, expose confidential information if used improperly, reproduce bias in data or decisions, and create compliance or intellectual-property concerns. Human review, access controls, testing, monitoring, documentation, and clear rules for sensitive information may be necessary, particularly in areas such as employment, credit, healthcare, education, safety, and public services. The applicable obligations depend on the jurisdiction, sector, use case, and nature of the data.

The most accurate way to state the answer

For a general audience, the answer should be expressed with both a range and its qualification:

There is no reliable worldwide count of companies using AI. Surveys of large and digitally mature organizations commonly find that around half to three-quarters report AI use in at least one function, while representative surveys covering businesses of all sizes often find much lower rates, frequently in the single digits or low double digits for concrete AI adoption. The exact percentage depends on the country, year, business size, industry, and definition of “use.”

The frequently cited figures of about 55% and 72% from recent McKinsey surveys illustrate broad reported use among surveyed organizations, while the approximately 42% figure from IBM’s 2023 enterprise survey illustrates active deployment under a narrower enterprise-oriented definition. None of these numbers should be presented as the percentage of every company in the world.

Thus, the answer to “what percentage of businesses use AI?” is not one fixed global percentage. The responsible answer is a survey-specific estimate, and the source’s population and definition matter as much as the number itself.