What the 2026 evidence shows about AI adoption and productivity - and why use alone does
not establish a return.
AI now appears in customer service, sales, software development and everyday office work. Yet many
companies still struggle to answer a basic question: what changed in revenue, cost, time, quality or customer outcomes?
PwC's 29th Global CEO Survey found that 12% of CEOs reported both cost savings and revenue growth
from AI over the prior year; 56% reported neither. In a separate 2026 study, PwC found that 20% of organizations captured 74% of the AI economic value measured in its sample. The question for an AI strategy is whether use produces measurable value.
Adoption is not the same as scale
McKinsey's 2025 survey found that 88% of respondents reported regular AI use in at least one business
function, but roughly one-third said their companies had begun scaling AI programs. Twenty-three percent said they were scaling an agentic AI system in at least one function. Its newer 2026 survey puts enterprise-wide AI scaling at 44%: progress, but still less than half of respondents.
MIT Project NANDA's 2025 report found that roughly 95% of the enterprise generative AI initiatives it
studied showed no measurable profit-and-loss impact. This is a finding from its sample, not a failure rate for every AI project. In that sample, external partnerships reached deployment about twice as often as internal builds; the researchers caution that this correlation does not prove the partnership caused the difference.
Giving more people access to tools is not enough. Value depends on choosing a workflow, integrating AI
into it and measuring the full result.
Productivity depends on the work
In a 2025 trial, METR found that experienced developers using AI tools took 19% longer to finish assigned
work, although they believed they had worked faster. A later study using newer tools estimated an 18% speedup for returning participants, but its uncertainty range included no speedup. These results concern particular developers and tasks, not all software work.
Routine, well-defined tasks may benefit quickly. Work requiring context, judgment or careful review may
deliver smaller gains once verification time is counted. Productivity claims should measure the complete workflow, including corrections and quality checks.
Work, infrastructure and trust
AI can change how people learn, how digital services scale and how audiences judge information.
Jobs: watch the entry point
Stanford Digital Economy Lab research using payroll data covering millions of US workers finds
employment changes concentrated among workers aged 22-25 in highly AI-exposed occupations. Its August 2026 revision describes a widened employment gap for younger workers, while stressing that the patterns are early, descriptive indicators rather than proof that AI caused them. It does not find widespread displacement across the labour market.
For leaders, a practical question follows: if AI takes on tasks that once trained junior employees, how will
new hires build the experience and judgment those tasks provided?
Infrastructure has a real cost
The International Energy Agency reports that worldwide data centre electricity demand grew 17% in
2025; electricity use from AI-focused data centres grew 50%. Its updated central projection puts total data centre electricity use at about 950 terawatt-hours in 2030, roughly double the 2025 level.
Power costs, grid capacity and approval to build facilities can influence how quickly AI infrastructure and
dependent services scale. They belong in longer-term plans alongside software and staffing costs.
Trust needs attention
A March 2026 Sprout Social survey found that 88% of respondents said the rise of AI video generation
tools had reduced their trust in news on social media. Fifty-six percent said they saw low-quality AI-generated content often or very often in their feeds. These are survey responses about social platforms, not a measure of trust in every AI product.
For a business, speed of content production is only one measure. Customer communications, marketing
and information used to make decisions still need accuracy, relevance and review.
Governance cannot wait for one deadline
The EU AI Act takes effect in stages. Article 50 transparency obligations generally apply from 2 August 2026
A limited grace period until 2 December 2026 covers certain marking and detection duties for AI
systems already on the market. Rules for specified stand-alone high-risk systems apply from 2 December 2027; those for high-risk AI embedded in regulated products apply from 2 August 2028.
Businesses should track the provisions relevant to their systems while building documentation, risk
assessment, clear accountability and human oversight for consequential decisions.
From AI use to AI value
Specific workflows, honest measurement and clear ownership make an AI strategy testable.
Know where you stand
Experimentation. What are we testing, and what result would tell us it failed? Adoption. Where is AI part of daily work, rather than merely available? Integration. Is it connected to the systems and data people use? Economics. What has it saved or generated after tool, maintenance and review costs? Governance. Who is accountable when an output is wrong? Workforce impact. Which roles and training paths need to change? Readiness. Are data, processes and capacity for change ready to support scale?
What leaders should do next
Measure outcomes, not activity. Establish a baseline before deployment; track cost, time, quality,
revenue or customer experience after it. Seat counts and query volumes show use, not value.
Choose the workflow before the tool. Begin with a defined problem, a clear owner and a result that can
be measured.
Budget for process redesign and review. Integration, training, quality checks and changes to daily work
can determine whether an investment pays off.
Plan for people and governance. Document risks, retain oversight where decisions matter and redesign
how junior employees gain experience.
Know when to stop. Agree on a measurement window and an end criterion before starting a pilot. Revisit
the economics regularly as tools, costs and usage change.
The real dividing line
AI has moved beyond the question, "Do we have it?" The more useful question is, "Can we show what it
changed?" Businesses that close the gap between adoption and value will focus on specific workflows, honest measurement, clear ownership and the discipline to stop initiatives that do not deliver.
