The numbers are large enough to invite skepticism. McKinsey's 2025 Global AI Survey reported that 65% of organizations are now regularly using generative AI in at least one business function, up from 33% in 2023. Total AI investment — including infrastructure, software, and services — is estimated at over $200 billion globally in 2025 by IDC. The natural question is what that investment is actually producing.
The most consistent finding from multiple surveys and implementation audits is that AI adoption is concentrated in a relatively small set of use cases: content generation and editing, code generation and review, customer support automation, and internal knowledge management. These are functions where AI's capabilities are well-matched to the task requirements, where the cost of errors is manageable, and where there are clear productivity metrics to measure against.
Marketing and communications organizations have the highest AI adoption rates in the surveys. Generating first drafts of marketing copy, localizing content across languages, personalizing email campaigns, and producing social media content at scale have all moved from early adoption to standard practice at large marketing teams. Jasper, Writer, and built-in AI features in tools like HubSpot and Salesforce Einstein have collectively embedded AI into the day-to-day work of millions of marketers.
Software engineering follows marketing in adoption rate. GitHub Copilot, Cursor, and similar tools have become standard in many engineering teams. McKinsey's research found that software teams using AI coding assistants reported a 20-30% reduction in time to first working code, though they noted that the quality and reliability of AI-generated code varied significantly by task complexity and that acceptance rates for multi-file changes were lower than for single-function completions.
Where adoption has been slower than anticipated: strategic decision-making, financial planning, and functions requiring institutional knowledge or judgment about organizational relationships. These are the tasks where AI's tendency to generate plausible-sounding but potentially incorrect outputs creates real risk. Finance teams that have deployed AI report using it primarily for data aggregation and report generation, not for the judgment calls about what the data means.
The ROI picture is clearer in some domains than others. Customer support automation with AI can show clear cost reduction when the AI successfully handles queries that previously required human attention. The measurement challenge is that human quality often differs from AI quality in dimensions that are hard to quantify — a human support agent notices when a customer seems upset about something beyond the stated issue; an AI generally does not.
The $200 billion investment is producing real productivity in specific, well-defined functions. The next challenge is extending those gains to the less structured, judgment-intensive work that constitutes much of what knowledge workers actually do.