Every CEO has, by now, seen the demos. AI summarising documents. AI drafting emails. AI answering customer questions. The temptation, for the last two years, has been to treat these as productivity tools useful additions to the technology stack, somewhere between a better search engine and a cleverer assistant. Worth experimenting with. Worth budgeting for. Not yet worth restructuring the business around.
That view is becoming increasingly difficult to defend. In April 2026, PwC published the results of a global survey of 1,217 senior executives across 25 sectors what the firm called its AI Performance Study. The headline finding was this: nearly three-quarters of AI’s economic value is already being captured by just 20% of companies. And the gap is widening every quarter. That number matters because it says two things at once: AI value is now real enough to measure, but the surplus is not being distributed evenly across the market.
Not a productivity tool. A lever on unit economics.
Productivity is a small word it suggests an extra 5% of output per worker, the kind of improvement a good HR programme might also deliver. That framing has cost organisations real time and real money over the last two years, because it has led them to deploy AI on top of unchanged workflows and then express disappointment when the financial impact looked marginal.
The right framing is different. AI, deployed well, is a lever on unit economics. It does three things to the financial structure of a business that no previous wave of enterprise technology has done at the same scale: it compresses the cost of high-volume customer-facing operations, it lowers the marginal cost of growth in knowledge-intensive functions, and it improves the loss-adjusted economics of decision-heavy functions. The leaders are not using fundamentally different models; they are rebuilding the operating model around them.
Three levers, three case studies
Lever 1: Klarna and the cost of customer operations. Within months of launching its OpenAI-powered customer service agent in early 2024, Klarna was handling two-thirds of all customer service chats with AI a volume the company described as equivalent to the workload of about 700 full-time agents. Average resolution time fell from 11 minutes to under 2. Repeat inquiries dropped 25%. By Q3 2025 the company reported approximately $60 million in cost savings attributable to this single deployment, while revenue grew 26% year over year. That is operating leverage, in the precise sense the term has meant for a century. Critically, Klarna later walked back its initial "AI-only" stance and rehired humans for complex cases the CEO acknowledged the company had "overpivoted" on cost reduction. The savings continued; the architecture that worked turned out to be hybrid, with humans in the loop at the right moments. The leaders are not the organisations that automate the most. They are the ones that automate the right things.
Lever 2: GitHub Copilot and the cost of growth. GitHub’s controlled research found developers completing standardised coding tasks up to 55% faster with Copilot. The tool is now used by more than 20 million developers, including approximately 90% of the Fortune 100, with an average of 46% of the code its users write now AI-generated. This is no longer an early-adopter phenomenon; it is the new baseline of engineering productivity in the largest companies in the world. For a CFO, the translation is sharp: this is not a cost-saving story but a revenue-enabling one a structurally lower cost of growth for any technology-led business.
Lever 3: JPMorgan and decision-heavy functions. CEO Jamie Dimon told Bloomberg in October 2025 that the bank’s ~$2 billion annual AI investment is now generating roughly $2 billion of measurable benefit each year "the tip of the iceberg," in his words. Software engineers 10–20% more productive. Operations staff handling 6% more accounts each. Per-unit fraud handling costs down approximately 11%, with AI-driven fraud detection credited for material loss prevention. At JPMorgan’s scale, these compound year after year into a structural advantage that competitors will struggle to close. The pattern matters more than any single number: AI shows up not as a discrete project but as an operating layer.
And yet most organisations are not on this list
MIT’s 2025 State of AI in Business report, analysing over 300 enterprise AI deployments, found that approximately 95% of pilot programs fail to deliver measurable financial returns. S&P Global reported that 42% of companies scrapped most of their AI initiatives in 2025, up from 17% the year before.
The numbers are not contradictory; they describe different populations. The leaders have deployed and are measuring. The 95% are still attempting and never reach production. The difference between the two is not the technology every serious player uses the same handful of foundation models. AI is a business problem dressed up as a technology problem. The leaders did the operating-model work first. The rest are still treating it as a procurement.
AI is fast becoming a defining factor in enterprise competitiveness. The companies that capture it well will compound advantages in cost, growth and margin over the next two to three years; those that do not risk falling further behind.
For most leadership teams, the question is no longer whether AI delivers value the leading companies have settled that. The question is why capturing it has proved so much harder than the demos suggested, and what the winners understand that the others do not. The answer is the subject of the next article in this series.
By Yiannis Stavrianos, Senior Manager, Advisory services, PwC Cyprus
