More Than Half of CEOs Say AI Has Delivered No Revenue or Cost Benefit. So Where Is the Trillion-Dollar Payoff

Artificial Intelligence 08 Sep 2026

More Than Half of CEOs Say AI Has Delivered No Revenue or Cost Benefit. So Where Is the Trillion-Dollar Payoff?

Imrul • 26 min read

Artificial intelligence has been sold to corporate leaders as almost everything at once.

A productivity revolution.

A cost-cutting machine.

A new revenue engine.

A threat to companies that move too slowly.

And, increasingly, something executives feel they cannot afford not to buy.

Yet after several years of aggressive enterprise investment, one of the world's largest surveys of chief executives has delivered an uncomfortable reality check.

According to PwC's 29th Global CEO Survey, 56% of CEOs said their organizations had experienced neither higher revenue nor lower costs from AI during the previous 12 months. Only 12% reported receiving both benefits simultaneously.

The survey is substantial: PwC collected responses from 4,454 CEOs across 95 countries and territories.

The numbers immediately generated a provocative interpretation:

After all the excitement, more than half of corporate leaders are getting zero financial return from AI.

That captures something real, but it needs an important qualification.

PwC did not ask CEOs whether every dollar spent on AI had produced a mathematically calculated return on investment of exactly zero.

It asked about two tangible business outcomes:

whether AI had increased revenue;
whether AI had reduced costs.

And 56% said they had achieved neither.

A company could therefore have experienced employee productivity improvements, faster research, better customer service, reduced risk or future strategic value without yet seeing those gains appear as higher revenue or lower cost.

Still, the message from the survey is difficult to ignore.

For most companies, AI activity has not yet become obvious bottom-line performance.

And that may be the most important phase of the AI boom so far.

The conversation is shifting from:

"Are we using AI?"

to:

"What, exactly, are we getting for the money?"

What PwC's 2026 CEO Survey Actually Found

PwC's results reveal a much more uneven AI economy than the hype might suggest.

Among surveyed CEOs:

30% said AI had produced additional revenue during the previous 12 months.
26% said AI had reduced costs.
22% said costs had actually increased because of AI.
56% said AI had produced neither higher revenue nor lower costs.
Only 12% said AI had simultaneously increased revenue and reduced costs.

PwC's accompanying press release summarized the situation another way:

Only around 33% of CEOs reported a financial gain in either revenue or cost performance, while the majority had yet to see significant financial benefit.

This does not describe an AI industry where nothing works.

It describes one where value is highly uneven.

A minority of companies appear to be extracting substantial business results.

Many others are still experimenting.

And some are spending more money on AI without yet recovering those costs elsewhere.

The 56% Figure Does Not Mean AI Has “Failed”

This distinction is crucial.

A technology can be useful before it becomes visibly profitable.

Consider a company that gives employees AI coding assistants.

Developers may produce code more quickly.

Documentation may improve.

Testing may accelerate.

But unless the company:

ships more products;
reduces headcount or contractor costs;
increases customer retention;
improves pricing;
shortens time-to-market;
expands capacity without equivalent hiring,

the productivity improvement may never appear as a clearly measurable financial return.

Employees simply accomplish more while the company's cost structure remains largely unchanged.

That may be valuable.

But it is not necessarily ROI in the form a CFO can point to on an income statement.

PwC's data therefore exposes a fundamental difference between AI usefulness and AI economics.

Those concepts have frequently been treated as though they were identical.

They are not.

A Worker Saving 30 Minutes Is Not Automatically a Corporate Profit

Imagine an AI tool saves an employee half an hour each day.

That sounds valuable.

Now ask what happens to the saved time.

Perhaps the employee:

completes more useful work;
speaks with more customers;
writes more software;
solves a backlog;
develops a new product.

Those outcomes could eventually produce economic value.

But perhaps the employee simply fills the recovered time with other routine tasks.

The organization still pays the same salary.

It now also pays for the AI license.

From the employee's perspective, AI may be helpful.

From the company's financial statements, costs have increased.

That gap is one reason AI adoption statistics can look spectacular while financial-return statistics look much weaker.

Buying AI Is Easy. Changing a Company Is Hard.

This is perhaps the central lesson emerging from PwC's research.

Buying access to AI has become remarkably simple.

A company can purchase:

enterprise chatbot licenses;
coding assistants;
AI search tools;
customer-support copilots;
meeting transcription;
document-generation software.

Within weeks, thousands of employees can technically have access to generative AI.

That creates adoption.

It does not necessarily create transformation.

PwC's CEO survey found that organizations achieving both revenue and cost benefits were two to three times more likely to report extensive AI deployment in products and services, demand generation and strategic decision-making.

The difference is not simply:

"Employees at successful companies use ChatGPT more."

The stronger companies are changing how work actually happens.

The Companies Winning With AI Are Redesigning Workflows

PwC followed its CEO survey with a separate 2026 Global AI Performance Study involving 1,217 senior executives across 25 sectors.

The results make the implementation gap even clearer.

The top-performing AI organizations were twice as likely to redesign workflows around AI rather than simply adding AI tools onto existing processes.

That difference sounds subtle.

It is enormous.

Imagine a customer-service department.

Company A: AI as an Add-On

An employee receives an AI assistant.

They:

read a customer complaint;
copy information into an AI tool;
ask it to draft a response;
edit the answer;
manually search another system;
update customer records;
escalate the case if necessary.

AI makes one step faster.

Company B: AI as Workflow Redesign

The system:

reads the incoming request;
identifies the customer;
retrieves account history;
categorizes the problem;
checks applicable policy;
recommends or executes an approved resolution;
updates internal systems;
drafts the response;
sends only unusual cases to a human.

The first company has AI adoption.

The second has process transformation.

The economics can be completely different.

The Most Successful Companies Are Not Merely “Doing More AI”

PwC's AI Performance Study found that the top 20% of companies captured 74% of measured AI-driven value.

Its most "AI-fit" companies generated AI-driven revenue and efficiency gains 7.2 times greater than the rest.

That is an extraordinary concentration.

It suggests that AI's benefits may not spread evenly simply because model access becomes cheaper.

The advantage increasingly comes from what organizations build around the model.

PwC identified foundations including:

strategy;
investment;
data and technology;
workforce capabilities;
governance and risk;
innovation practices.

Companies with strong foundations were much better at translating greater AI usage into measurable performance.

In the CEO survey itself, organizations with strong AI foundations were three times more likely to report meaningful financial returns.

Data Is Often the Real Bottleneck

Large language models can be extraordinarily capable.

But enterprises rarely need only general knowledge.

They need AI to understand:

internal prices;
inventory;
customer histories;
company policies;
contracts;
product catalogs;
supply chains;
proprietary research;
previous transactions;
operational metrics.

That information frequently lives across incompatible systems.

Some data is outdated.

Some has no clear owner.

Some exists only in PDFs.

Some cannot legally be exposed to particular employees.

Some is duplicated.

Some contains inconsistent definitions.

Some lives in databases built 20 years ago.

Buying an advanced AI model does not solve any of those problems automatically.

If the AI cannot safely access reliable enterprise data, it may remain an impressive conversational layer floating above the actual business.

AI Can Make Bad Processes Faster

Companies have spent decades learning a painful lesson about automation:

Automating a broken process does not necessarily fix it.

It can simply accelerate the broken process.

Suppose a company has a purchasing workflow requiring seven unnecessary approvals.

Adding AI that writes the purchasing request faster does not remove the six unnecessary approvals.

Suppose employees repeatedly enter the same customer information into three disconnected systems.

An AI assistant might help fill those forms.

But the more transformative solution may be eliminating the duplicate entry entirely.

This is why workflow redesign matters.

AI creates its largest economic value when leaders reconsider the process itself rather than placing a chatbot on top of it.

Productivity Gains Can Disappear Into Organizational Slack

Economists and technology executives have long struggled with the difference between local productivity gains and enterprise-wide financial improvement.

An individual worker can become faster without the corporation becoming more profitable.

There are several ways AI-generated time savings can disappear.

Employees may:

increase output without increasing sales;
spend more time checking AI errors;
use AI inconsistently;
save minutes across tasks too fragmented to monetize;
perform more work that was never economically valuable;
recreate outputs that managers do not need.

Organizations may then report high employee adoption while the CFO sees little change in:

operating margin;
revenue;
labor expense;
throughput;
customer retention.

This helps explain why "hours saved" is an imperfect AI metric.

A company cannot deposit saved hours into a bank.

Those hours must be converted into a business outcome.

License Counts Became the Wrong Metric

During the early generative-AI boom, organizations often measured success using metrics such as:

number of AI accounts activated;
percentage of employees using the tool;
prompts submitted;
AI training sessions completed;
pilots launched.

Those metrics are understandable during experimentation.

They show whether people are engaging with the technology.

But they say almost nothing about economic impact.

A corporation can have 90% AI adoption and still lose money on AI.

The increasingly relevant questions are:

Did sales conversion improve?
Did customer churn decline?
Did software ship faster?
Did support cost per case fall?
Did inventory accuracy improve?
Did fraud losses decline?
Did product development accelerate?
Did revenue per employee increase?
Did margins improve?

AI maturity is beginning to mean proving those outcomes rather than merely proving people logged in.

The Pilot Trap

The word pilot became almost synonymous with enterprise AI after the launch of modern generative models.

Pilots are attractive because they are low risk.

A department can test an idea with:

20 employees;
one dataset;
a limited budget;
minimal systems integration.

If it works, executives can show an impressive demonstration.

But pilots often exist inside artificially favorable conditions.

Scaling introduces:

security;
permissions;
compliance;
reliability;
latency;
integration;
training;
organizational politics;
data quality;
monitoring;
procurement;
support.

A prototype that saves one analyst two hours may be easy.

Transforming the workflow of 20,000 employees is an organizational redesign project.

That is where many AI initiatives stall.

“Pilot Purgatory” Is Becoming Expensive

Companies can accumulate dozens or even hundreds of AI experiments.

Each looks promising individually.

But they may use:

different vendors;
duplicate data pipelines;
separate governance systems;
incompatible models;
independent budgets.

Eventually, executives realize they have built an impressive collection of demonstrations without changing the operating model.

PwC's Global Chairman Mohamed Kande described 2026 as a decisive year, warning that while a small group is producing measurable financial returns, many organizations remain stuck beyond the pilot stage.

That divide is increasingly visible.

AI Costs More Than the Model Subscription

Another reason ROI is difficult is that businesses often underestimate the total cost.

The visible price may be a software license or API bill.

The actual cost can include:

cloud infrastructure;
model inference;
data engineering;
cybersecurity;
governance;
legal review;
system integration;
consultants;
model evaluation;
employee training;
monitoring;
AI-specific hires;
process redesign.

Some companies also build internal AI platforms that duplicate capabilities already available commercially.

If an AI assistant saves $2 million worth of employee time but costs $3 million to deploy and operate, the organization has improved productivity while destroying financial value.

Those distinctions become important as experimental budgets move under greater scrutiny.

Sometimes AI Actually Raises Costs

PwC's finding that 22% of CEOs reported higher costs due to AI deserves almost as much attention as the 56% figure.

Early technology adoption frequently behaves this way.

Companies must maintain the old system while paying for the new one.

Employees still perform existing work while also:

learning new tools;
checking AI output;
developing policies;
running experiments.

The organization has temporarily duplicated effort.

This does not necessarily mean the investment is bad.

But it means a transition phase can be financially painful.

The challenge for executives is distinguishing:

temporary implementation cost

from:

a permanently uneconomic AI project.

The 12% Matter More Than the 56%

The most interesting group in PwC's survey may not be the companies receiving no measurable benefit.

It is the one in eight achieving both:

additional revenue;
lower costs.

If AI were fundamentally incapable of producing economic value, that group should not exist.

Instead, PwC found meaningful differences between them and everyone else.

They deploy AI more deeply.

They have stronger technology and governance foundations.

They integrate it into actual business functions.

That points toward an implementation problem rather than simply a technology problem.

Top Performers Use AI for Growth, Not Only Cost Cutting

Much of the early corporate AI narrative focused on efficiency.

Write the email faster.

Summarize the document.

Reduce support workload.

Automate administrative work.

Those uses can matter.

PwC's performance research suggests the strongest companies go further.

They are approximately two to three times more likely to use AI to pursue growth opportunities and reinvent their business models.

That can mean:

creating new products;
improving personalization;
entering adjacent markets;
developing AI-powered services;
improving pricing;
accelerating product discovery;
changing how customers interact with the company.

Cost reduction has a ceiling.

You can only eliminate so much expense.

Revenue growth has much more room.

AI Leaders Are Also Automating Decisions

PwC found that leading organizations were roughly 2.8 times more likely to have increased the number of decisions made without direct human intervention.

This is another important transition.

A traditional AI copilot might say:

"Here is my recommendation."

A more integrated system may:

approve a low-risk transaction;
reorder inventory;
route a customer request;
modify a marketing campaign;
detect fraud;
assign work.

That produces more leverage.

It also creates more risk.

The companies pushing furthest into autonomy therefore need stronger:

governance;
auditability;
monitoring;
human escalation;
security controls.

Interestingly, PwC found that its AI leaders tended to advance both automation and governance rather than choosing between them.

Governance Is Not Just a Brake on AI

AI governance is sometimes portrayed as bureaucracy preventing companies from moving quickly.

Poorly designed governance can certainly do that.

But organizations cannot scale AI into valuable processes if executives do not trust it.

Imagine allowing AI to:

approve loans;
price insurance;
screen job applicants;
refund customers;
change production schedules;

without confidence in:

data quality;
accuracy;
fairness;
security;
accountability.

Executives will naturally restrict the system to low-risk demonstrations.

PwC found that only 51% of companies had formalized their approach to AI risk, suggesting weak governance may itself be limiting broader deployment.

Good governance can therefore enable adoption rather than merely constrain it.

The Same AI Model Can Produce Different Business Outcomes

This is perhaps the most important strategic implication.

Competitors increasingly have access to similar foundational AI technologies.

A bank can buy access to the same model another bank uses.

A retailer can purchase similar cloud AI tools as a rival.

A software company can use the same coding assistant as another developer.

Model access is becoming commoditized.

If everyone can buy similar intelligence, the competitive advantage shifts toward:

proprietary data;
workflow design;
customer relationships;
distribution;
integration;
organizational speed;
employee capability.

The model may be impressive.

The business system around it determines whether the impression becomes money.

AI May Be Following a Familiar Technology Pattern

Major general-purpose technologies often require organizational changes before their full economic value becomes visible.

Electrification is a famous historical example.

Factories initially replaced centralized steam power with electric motors without redesigning the factory around electricity's unique advantages.

The largest productivity gains came later when factory layouts and processes changed.

Computers followed a similar pattern.

Buying PCs did not automatically transform productivity.

Companies had to redesign information flows and work practices around them.

AI may be entering an equivalent stage.

Installing an AI assistant into an unchanged organization may generate incremental improvements.

Redesigning the organization around cheap machine intelligence could produce something much larger.

PwC's data is consistent with that possibility, although the long-term magnitude remains uncertain.

This Does Not Guarantee That Every AI Investment Will Eventually Pay Off

Technology optimism can create another dangerous assumption:

"If there is no return yet, just wait."

That is not necessarily true.

Some AI projects are likely bad investments.

They may solve problems that:

are not economically important;
already have cheap solutions;
occur too infrequently;
require too much human verification;
introduce excessive risk.

Not every process benefits from AI.

The challenge for executives is becoming more willing to kill unsuccessful projects.

A pilot should answer a question.

If the answer is:

"This does not create enough value,"

the correct response may be stopping rather than scaling.

ROI Discipline Could Be Healthy for the AI Industry

The shift toward financial accountability may sound like bad news for AI vendors.

In the long term, it could strengthen the market.

The first phase of a technology boom rewards:

novelty;
announcements;
experimentation.

The second phase rewards:

reliability;
integration;
economics.

That forces vendors to prove that their products create measurable value.

Buyers become more sophisticated.

Weak products disappear.

Strong applications become embedded in real operations.

This is how a technology moves from hype cycle to infrastructure.

CEOs Are Increasingly Worried About Transformation Speed

Despite the disappointing financial-return numbers, CEOs are not becoming indifferent to AI.

Quite the opposite.

PwC asked CEOs which question concerned them most.

The leading answer, selected by 42%, was whether their organizations were transforming quickly enough to keep pace with technological change, including AI.

That creates a difficult psychological environment.

Executives simultaneously fear:

We're spending too much on AI without returns.

and:

We're not moving fast enough on AI.

Those fears pull organizations in opposite directions.

One encourages caution.

The other encourages spending.

The strongest companies will likely need to do both:

move aggressively where value is visible and ruthlessly stop projects where it is not.

AI Is Not Yet Deeply Embedded Across Most Enterprises

Another PwC finding helps explain the weak financial results.

Only a relatively small share of CEOs said AI had been deployed to a large or very large extent in important business areas:

demand generation: 22%
support services: 20%
products, services and experiences: 19%
direction setting: 15%
demand fulfillment: 13%.

So although almost every major company may now talk about AI, enterprise deployment remains much shallower than the public narrative sometimes suggests.

The technology may be everywhere culturally while still being relatively thin operationally.

Another 2026 Survey Offers a Different-Looking Result

In July 2026, Boston Consulting Group published research finding that nearly nine in ten CEOs reported some cost or revenue benefits from AI in targeted areas.

At first glance, that seems to contradict PwC's 56% figure.

It does not necessarily.

The surveys asked different questions, used different samples and measured different levels of impact.

BCG surveyed 152 CEOs of large companies and emphasized benefits in targeted areas, while also finding that most organizations struggled to translate those localized gains into scaled financial impact. More than half of BCG's CEOs said there was a missing connection between AI and the company's profit-and-loss results.

The two studies therefore point toward a similar underlying pattern:

AI can produce local wins without yet materially changing enterprise economics.

The Next AI Metric Is P&L

Profit-and-loss accountability changes how companies evaluate technology.

Instead of asking:

"Did employees like the AI tool?"

leaders increasingly ask:

"Which P&L line changed?"

Revenue?

Gross margin?

Sales expense?

Customer-support cost?

Inventory?

Capital spending?

That forces clarity.

A company claiming that AI has saved 500,000 employee hours may now be asked:

What happened to those hours?

Did headcount growth slow?

Did output rise?

Did sales improve?

If nothing measurable changed, the claimed savings may be theoretical rather than economic.

CFOs Are Likely to Become More Powerful in AI Strategy

The first phase of generative AI was driven heavily by:

technology teams;
innovation groups;
CEOs;
venture-style experimentation.

As projects mature, finance departments become more important.

CFOs naturally ask:

What is the baseline?
What is the implementation cost?
What KPI changes?
Over what period?
Compared with what alternative?
Who owns the result?

That may feel less exciting than watching a generative AI demo.

It is also how enterprise software becomes sustainable.

Companies Need to Measure AI Before and After

One surprisingly common ROI problem is the absence of a baseline.

Suppose an organization launches an AI support tool and discovers that average case resolution now takes eight minutes.

Is that good?

It depends.

If the old average was 12 minutes, probably.

If it was seven, the AI may have made performance worse.

Reliable AI measurement therefore requires metrics established before deployment.

Examples include:

cost per transaction;
conversion rate;
resolution time;
defect rate;
developer cycle time;
customer satisfaction;
revenue per sales representative.

Without a baseline, organizations can demonstrate activity without demonstrating improvement.

AI ROI Should Include Error Costs

Speed is not the only dimension that matters.

An AI system may process tasks 50% faster while producing more mistakes.

If humans then spend significant time checking the output, apparent productivity disappears.

Worse, undetected errors may create:

refunds;
legal exposure;
incorrect financial decisions;
customer churn;
security incidents.

A serious ROI model must therefore include:

gross productivity benefit minus verification, error and risk cost.

This is especially important for generative models because fluent output can make mistakes less obvious.

Human Work May Need to Change Alongside AI

PwC's workforce analysis emphasizes that enterprise value depends on redesigning roles, skills and work—not merely deploying technology.

If AI can complete the first draft of an analyst's work, the analyst's job should eventually change.

Perhaps the human spends less time producing material and more time:

validating assumptions;
making decisions;
communicating findings;
handling exceptions.

If management keeps the old job description completely intact, AI may simply create duplicate effort.

Transformation therefore becomes a workforce problem as much as a software problem.

Reskilling Is Not Optional

Companies often speak about AI training as:

"Teach employees how to prompt."

That is only one small part.

A genuinely AI-enabled employee may need to understand:

when AI is useful;
when it is unreliable;
how to validate outputs;
how to redesign a process;
how to work with automated agents;
when human escalation is required.

Managers also need new skills.

They must learn how to manage systems in which work is divided between humans and machines.

The more capable the AI becomes, the more important those organizational questions become.

Agentic AI Could Increase Both Value and Risk

The next stage of enterprise AI is increasingly described as agentic.

Instead of only generating text, agents can:

retrieve information;
use software tools;
execute workflows;
make decisions;
interact with other agents.

This may solve one of the current ROI problems because agents can automate entire processes rather than merely helping humans complete individual steps.

But it dramatically raises stakes.

A chatbot producing a bad recommendation is one thing.

An autonomous system executing the recommendation is another.

Companies pursuing deeper AI-driven returns will therefore need stronger controls at the same time that they grant AI more autonomy.

The Winners May Spend More, Not Less

The finding that many AI investments are failing to produce returns does not necessarily imply that successful companies will minimize AI spending.

PwC's broader benchmarking suggests the opposite can be true.

Organizations obtaining stronger outcomes tend to invest seriously in the foundations needed to scale AI.

The relevant distinction may not be:

high spending versus low spending.

It may be:

integrated investment versus fragmented spending.

A company spending heavily on:

data modernization;
workflow redesign;
governance;
employee skills;
integration

may ultimately outperform one purchasing thousands of standalone AI subscriptions.

There Is a Difference Between AI Cost Savings and Layoffs

Whenever AI efficiency is discussed, attention quickly turns to jobs.

AI may allow organizations to accomplish more with fewer people in particular processes.

But cost reduction can emerge in many ways besides immediate layoffs.

A company could:

slow future hiring;
reduce contractor spending;
process more transactions with existing staff;
eliminate overtime;
reduce mistakes;
shorten project cycles;
automate outsourced work.

This is important when interpreting CEO reports of cost benefits.

"AI reduced costs" does not automatically mean:

"AI replaced employees."

The economics can emerge through multiple channels.

The AI Boom Is Becoming Less About Models

During the early generative-AI race, business attention centered on model capability.

Which model scores highest?

Which has the largest context window?

Which reasons better?

Those differences still matter.

But enterprise economics increasingly depend on everything surrounding the model.

The model may be only one layer within a system containing:

data;
APIs;
permissions;
workflow engines;
human review;
monitoring;
analytics.

That means the next era of AI competition inside corporations could look less glamorous.

The decisive work may involve process maps and databases rather than flashy chatbot demonstrations.

What Companies Seeing AI Returns Appear to Do Differently

Taken together, PwC's 2026 research suggests a recognizable pattern.

Companies generating stronger financial value tend to:

Target meaningful business outcomes.
They connect AI initiatives to revenue, efficiency or strategic growth rather than merely launching experiments.
Redesign workflows.
They change processes around AI instead of dropping an AI tool into an unchanged process.
Deploy at scale.
AI becomes embedded across important functions rather than isolated in a few pilots.
Build strong foundations.
Data, technology, workforce, strategy and governance are treated as essential infrastructure.
Pursue revenue as well as efficiency.
Leading companies use AI to create offerings and reshape business models, not only save labor.
Measure actual performance.
They focus on financial and operational outcomes rather than adoption statistics.
Increase automation responsibly.
High performers permit more autonomous decisions while simultaneously strengthening governance.

What Companies Should Stop Doing

PwC's findings also imply several practices that deserve skepticism.

Buying AI Because Competitors Bought AI

Fear of missing out is not a business case.

Counting Licenses as Success

Usage is valuable only if it contributes to a meaningful outcome.

Launching Endless Pilots

Experiments should eventually be scaled, modified or killed.

Automating Broken Processes

Fix the process before automating unnecessary steps.

Treating Data Problems as Someone Else's Problem

AI output cannot consistently exceed the quality of the information the system can access.

Assuming Productivity Automatically Means Savings

Time savings require an economic mechanism that captures them.

Measuring Everything Except Money

Eventually, significant investments need financial accountability.

Is the AI Bubble Bursting?

PwC's results do not prove that.

A bubble implies prices or investment have become disconnected from sustainable economic value.

The survey demonstrates that many corporate AI programs have not yet produced visible financial returns.

Those are different claims.

In fact, PwC also found a minority of companies generating substantial measurable value.

That suggests AI itself is not merely imaginary economic activity.

The harder question is whether the enormous amount of investment across:

models;
data centers;
enterprise software;
consulting;
internal transformation

will ultimately produce returns large enough to justify total spending.

That question remains open.

2026 May Be the Year AI Has to Prove Itself

The early years of generative AI benefited from extraordinary tolerance for experimentation.

Executives could reasonably say:

"We need to learn."

By 2026, boards increasingly want the next sentence.

"What did we learn, and what did it earn?"

This does not mean experimentation ends.

It means experimentation acquires a deadline.

The technology has become too expensive and strategically important to remain indefinitely outside ordinary financial discipline.

The Numbers Are a Warning to Vendors Too

AI companies frequently sell value through stories about employee productivity.

Corporate buyers are becoming more demanding.

They will increasingly want vendors to demonstrate outcomes such as:

increased revenue;
reduced operating costs;
faster product release;
lower customer-support costs;
fewer errors.

Vendors that cannot connect their technology to these outcomes may face harder renewals.

That shift could fundamentally reshape the enterprise AI market.

The easiest software to purchase is not necessarily the easiest to justify a year later.

The 56% May Fall Rapidly—or Remain Stubbornly High

There are two plausible interpretations of the survey.

The Optimistic Interpretation

Enterprise AI is still early.

Most companies are learning.

Infrastructure and workflows take time to change.

As companies mature, many of today's failed pilots will become profitable systems.

The 56% figure could fall dramatically.

The Skeptical Interpretation

AI is genuinely useful, but corporations have overestimated how much general-purpose generative AI translates into measurable economic value.

Some productivity gains may never be captured financially.

AI budgets may be substantially cut once experimentation ends.

Reality could contain both.

AI can be transformative in some processes while commercially disappointing in others.

What PwC's Survey Does Not Tell Us

The CEO survey is valuable, but it has limits.

It relies on executive responses rather than audited financial statements proving causation.

CEOs may have incomplete visibility into every AI benefit.

Companies may define AI differently.

It can also be difficult to isolate the impact of AI from:

pricing changes;
economic conditions;
workforce restructuring;
other technology investments.

The survey therefore should not be treated as precise accounting evidence that 56% of AI investments produced mathematically zero ROI.

Its importance lies in executive perception.

More than half of thousands of CEOs could not point to either of the two most obvious financial benefits of AI:

higher revenue or lower costs.

That alone is significant.

The Bottom Line

Artificial intelligence has reached an important transition.

The question is no longer whether companies are interested in AI.

They clearly are.

The question is whether that enthusiasm is turning into measurable financial performance.

PwC's 29th Global CEO Survey, based on responses from 4,454 CEOs across 95 countries and territories, found that 56% said their organizations had experienced neither increased revenue nor reduced costs from AI during the previous 12 months.

Only:

30% reported higher revenue;
26% reported lower costs;
12% achieved both.

Those statistics are sobering.

But they do not prove that AI itself lacks economic value.

PwC's separate 2026 performance research found the opposite at the top end of the market.

The strongest 20% of companies captured 74% of AI-driven value, while the most AI-fit organizations produced revenue and efficiency gains 7.2 times greater than their peers.

The difference appears to be less about access to the latest model and more about organizational execution.

AI leaders are more likely to:

redesign workflows;
integrate AI deeply;
strengthen data and technology infrastructure;
develop workforce capabilities;
build governance;
pursue revenue growth as well as productivity.

PwC found that top performers were twice as likely to redesign workflows around AI rather than simply adding AI tools to existing processes.

That may be the defining lesson of the current AI boom.

A chatbot license is easy to buy.

A demo is easy to produce.

An impressive pilot is increasingly easy to build.

Rewiring an organization so that intelligence—human and artificial—flows through the business differently is much harder.

And that is where the money appears to be.

The corporate AI debate is therefore entering a healthier, more demanding phase.

Executives are moving beyond:

How many employees are using AI?

toward:

Which revenue increased? Which cost disappeared? Which process changed?

For vendors, employees and corporate leaders, that shift changes everything.

AI no longer gets credit merely for being impressive.

Increasingly, it has to show the receipts.

Frequently Asked Questions

Did 56% of CEOs say AI has produced zero ROI?

PwC found that 56% of CEOs said their organizations had realized neither higher revenues nor lower costs from AI during the previous 12 months. Calling that "zero ROI" is understandable shorthand, but it is not identical to a formal accounting calculation showing exactly 0% return on every AI investment.

What survey did the 56% AI figure come from?

It came from PwC's 29th Global CEO Survey, published in January 2026.

How many CEOs participated in PwC's survey?

PwC surveyed 4,454 CEOs across 95 countries and territories.

How many CEOs say AI has increased revenue?

Approximately 30% said AI had generated additional revenues during the previous 12 months.

How many CEOs say AI has reduced costs?

Approximately 26% reported that AI had lowered costs.

How many companies achieved both higher revenue and lower costs from AI?

Only 12%, or roughly one in eight CEOs surveyed, reported both benefits.

Did AI increase costs for some companies?

Yes. PwC reported that 22% of CEOs said costs had increased due to AI.

Does PwC's survey prove AI is a financial failure?

No. It demonstrates that the majority of surveyed CEOs had not yet seen two major categories of financial benefit. A smaller group was producing substantial measurable returns.

Why are companies struggling to make money from AI?

Common issues include shallow deployment, poor data foundations, fragmented pilots, limited workflow redesign, training challenges, governance concerns and difficulty translating worker productivity into financial outcomes.

What are successful AI companies doing differently?

PwC found that stronger performers deploy AI more deeply, strengthen organizational foundations, redesign workflows and use AI for growth and business-model reinvention rather than focusing only on isolated productivity improvements.

Are successful companies simply buying better AI models?

PwC's research suggests organizational factors are extremely important. Companies with stronger AI foundations were much more likely to report meaningful financial benefits even as access to advanced models becomes widely available.

What percentage of AI value is captured by leading companies?

PwC's separate 2026 AI Performance Study found that the top 20% of companies captured 74% of measured AI-driven returns.

How much better are the strongest AI companies performing?

PwC says its most AI-fit organizations achieved 7.2 times higher AI-driven revenue and efficiency gains than other companies in its study.

Do AI leaders redesign their workflows?

Yes. PwC found leading companies were twice as likely to redesign workflows around AI instead of simply adding AI tools to existing processes.

Why doesn't giving everyone an AI assistant automatically create ROI?

An assistant can make individual tasks faster without changing headcount, output, revenue or operating costs. Unless time savings are converted into economically useful outcomes, productivity does not automatically become profit.

What is an AI workflow redesign?

It means restructuring an entire business process around AI capabilities rather than merely inserting an AI assistant into one step of an unchanged process.

Why are AI pilots often difficult to scale?

Large-scale deployment introduces issues involving security, data permissions, reliability, compliance, integration, workforce training, governance and organizational change that may not appear in a small pilot.

What is “pilot purgatory” in AI?

It refers to organizations repeatedly launching successful-looking AI experiments without converting them into scaled, financially meaningful business systems.

Why is enterprise data important to AI ROI?

Most valuable corporate AI applications require accurate access to internal customer, product, financial or operational information. Poor or fragmented data limits what AI can reliably automate.

Can AI increase productivity without reducing costs?

Yes. Employees may accomplish more while salaries and staffing remain unchanged. In that case productivity can rise even though the company's immediate costs do not fall.

Can AI save time without producing revenue?

Yes. Saving employee hours creates potential capacity, but the company must use that capacity to increase output, improve service, reduce hiring or achieve another measurable business outcome.

Are AI adoption metrics misleading?

They can be. License activation, prompt volume and user counts show usage but do not establish revenue growth, margin improvement or cost reduction.

What AI metrics should companies track instead?

Useful measures can include revenue growth, conversion rates, cycle time, cost per transaction, support cost, defect rates, customer retention and revenue per employee.

Why is workflow redesign more important than AI access?

Advanced models are increasingly available to many competitors. Organizations can differentiate themselves through how effectively they integrate those models into unique business processes and data.

Do AI leaders focus only on cost cutting?

No. PwC found leading companies were more likely to use AI to pursue growth opportunities and reinvent business models.

Can AI create new revenue?

Yes. PwC's CEO survey found 30% of respondents reporting additional revenue associated with AI, while its performance research identified companies using AI for new offerings and business-model innovation.

Are companies using AI extensively across their businesses?

Not most of them. Only 22% of CEOs reported extensive AI deployment in demand generation, 20% in support services, 19% in products and experiences, 15% in direction setting and 13% in demand fulfillment.

Are AI risk and governance preventing adoption?

They may be one factor. PwC reported that only 51% of companies had formalized their approach to AI risk, which it suggested could constrain broader deployment.

Does governance reduce AI innovation?

Not necessarily. Strong governance can increase executive confidence in allowing AI to participate in higher-value and higher-risk workflows.

Are AI leaders allowing more autonomous decisions?

Yes. PwC's AI Performance Study found leading companies were about 2.8 times more likely to have increased decisions made without human intervention.

Why are CFOs becoming more important to AI adoption?

As AI projects move beyond experimentation, companies increasingly need formal financial measurement, baselines, budgets and accountability for outcomes.

What is the difference between AI productivity and AI ROI?

Productivity measures how much work can be produced with given resources. ROI compares the economic value generated with the money invested. Productivity can improve without producing positive financial ROI.

Why should AI projects establish a baseline before deployment?

Without knowing performance before AI was introduced, companies cannot reliably determine whether the technology improved cost, speed, quality or revenue.

Why should AI error costs be included in ROI?

AI-generated mistakes may require human review or create financial, legal and customer-service costs. Gross time savings can therefore overstate actual economic value.

Does Boston Consulting Group agree that companies are struggling with AI ROI?

BCG's July 2026 CEO research found much broader reports of benefits in targeted areas but similarly identified difficulty converting those local improvements into scaled P&L impact. More than half of CEOs cited a missing connection between AI and P&L.

Why are BCG's AI numbers different from PwC's?

The surveys used different samples and questions. BCG emphasized benefits in targeted applications, whereas PwC asked CEOs about organization-level revenue and cost effects. The findings can therefore coexist.

Is AI currently overhyped?

Certain expectations may have moved faster than measurable enterprise returns, but PwC's data also shows that some companies are generating significant benefits. The evidence supports an execution gap more clearly than a conclusion that AI has no economic value.

Is the AI bubble bursting?

PwC's survey alone cannot establish whether there is an investment bubble. It shows that many companies have yet to translate AI spending into revenue gains or cost reductions.

Could companies simply need more time?

Some likely do. Large technological transformations often require changes to data, systems and work processes before financial benefits become visible. Other projects may never produce adequate returns and should be discontinued.

Should companies stop investing in AI if they have not seen ROI?

Not automatically. They should determine why returns are missing, establish measurable objectives and decide whether to redesign, scale or stop each initiative rather than continuing investment solely because AI is fashionable.

What should executives ask before approving an AI project?

They should identify the business problem, baseline performance, expected financial outcome, total deployment cost, responsible owner and measurable success criteria.

What is the biggest mistake companies are making with AI?

One major mistake is treating AI as a software purchase rather than an operating-model change. PwC's research suggests that simply adding tools to existing workflows produces much weaker outcomes than redesigning the work itself.

What does PwC mean by strong AI foundations?

PwC groups foundations around areas including strategy, investment, data and technology, workforce capabilities, governance and risk, and innovation.

Are CEOs worried that their companies are moving too slowly on AI?

Yes. PwC found 42% of CEOs identified keeping pace with technological transformation, including AI, as the question that concerned them most.

What is the key lesson from PwC's 2026 AI findings?

Access to powerful AI is becoming common.

Profitable implementation is not.

The organizations gaining the most appear to be changing workflows, data systems, decision processes and business models around AI rather than simply purchasing AI tools and counting how many employees use them.