The Productivity Mirage: Why AI Gains Aren't Reaching The P&L
Janaki Jayachandran is Global Head of Software & Technology at Aspire System.
gettyThree years into the generative AI era, the boardroom question has evolved from “What is our AI strategy?” to “What has our AI strategy earned us?”
The evidence is uncomfortable. McKinsey’s 2026 “State of AI survey,” conducted with 1,719 executives and published in August, found that only 37% attribute any measurable earnings before interest and taxes (EBIT) to AI. Just 6% qualify as high performers, attributing 5% or more of EBIT to AI. Yet, 80% say AI has made them more productive.
That gap is the central problem in enterprise AI. The technology works for individuals but has not yet reached the income statement.
Other research reports agree. BCG’s 2025 study, “The Widening AI Value Gap,” found that 60% of companies see minimal revenue or profits from AI, while only 5% are extracting value at scale. A preliminary 2025 report from MIT’s Project NANDA reported that 95% of enterprise pilots had produced no measurable return from generative AI.
The shortfall is mostly a failure of management, measurement and operating design, not technology. Seven challenges explain most of it.
Most AI business cases promise saved hours. A copilot increases coding speed by over 30%. A support assistant shortens handling time by two minutes per call. These gains are real, but they rarely reach the margin because nobody decides what happens to the freed capacity.
Converting productivity into margin requires a deliberate decision to remove capacity through slower hiring or lower external spend, redeploy it to revenue-generating work, or shorten cycle time customers will pay for. Without that decision, organization will become busier without improving margins.
Many deployments add AI to existing processes and leave the same approvals, handoffs and exception queues in place. But, to see returns, leaders need to ask how an AI-led process should run.
McKinsey’s survey found that 73% of high performers redesigned workflows around AI.
Enterprise-wide chatbots and copilot licenses are easy to buy, deploy and report. However, their value is hard to measure and capture.
So, while BCG estimates that about 70% of AI’s potential value sits in core functions such as sales and marketing, manufacturing, supply chain and pricing, McKinsey’s 2025 study says 90% of function-level use cases remain stuck at the pilot stage.
Pilots prove that something can work. Production must prove that it works reliably, at scale and inside real systems of record. A successful proof of concept on clean data is followed by months of integration, security review, data access, governance and change management. Budget or patience often runs out first, leaving many promising experiments and few operational systems.
Many enterprises track counts: use cases launched, licenses deployed, prompts per user, adoption rates. They say little about whether cost per transaction fell, days sales outstanding improved, or churn declined. The baseline is never recorded, so the value cannot be proved even when the program works.
CFOs are right to insist that every AI initiative above a set threshold carries a named financial metric, a recorded baseline and a date by which the effect should appear. Without these, AI spending cannot be defended, and budgets will face growing scrutiny in the next planning cycle.
Business cases focus on benefits and underestimate running costs. Inference charges that look trivial in a pilot can climb sharply at production volume, alongside data engineering, monitoring, security controls, vendor-license stacking and the specialist staff.
Agentic systems can make many model calls per task, so an initiative that pays off at 10,000 transactions a month can erode margin at a million. Organizations therefore need to assess total cost of ownership with the same rigor as the benefits case.
In many organizations, the technology team leads AI implementation while business leaders are responsible for the P&L impact. Technology is measured on delivery, the business on results, and nobody is accountable for connecting the two.
The companies pulling ahead have stopped treating AI as a technology initiative. They treat it as a business accountability issue. The executive who owns the margin owns the AI agenda that is expected to improve it. Technology doesn’t own the value, it enables the business to create it.
Leaders who treat AI as a business transformation discipline concentrate deep investment on a few high-value domains instead of scattered experiments. They redesign before they automate, building AI around the real constraint. They decide in advance what happens to freed capacity and write it into the business case. They measure outcomes and unit economics, with a financial metric, baseline, timeline and cost per outcome for every initiative. And they give business leaders ownership, with shared accountability alongside technology.
In my own work leading software and technology practices, I’ve found that AI gains are not secured at the point of adoption, but through disciplined execution in the months that follow. Early experiments in software engineering showed promising productivity improvements, but sustaining those gains required us to connect AI adoption to measurable outcomes, governed usage and changes in how teams work.
We have also learned that the people and financial dimensions matter as much as the technology. Teams need to adapt roles and practices around human-AI collaboration, while AI consumption and token costs need the same discipline applied to other technology spend.
None of this is new. It is a discipline long applied to capital investment, and AI is no exception.
• Which three AI initiatives have the largest measured effect on EBIT, and what was the baseline?
• For each productivity deployment, where did the freed capacity go?
• What share of AI spend goes to core value-chain processes versus general-purpose tools?
• What is the fully loaded cost per outcome for our largest AI workloads, and how does it scale?
• Who is the named business owner accountable for the return on each major initiative?
Organizations that can answer these clearly are likely already among the minority seeing results.
AI’s productivity potential is not in doubt. The open question is whether enterprises will do the less glamorous work of turning that potential into margin by redesigning processes, harvesting capacity, controlling costs and assigning accountability.
The gap between leaders and laggards is widening, and operating discipline, not access to technology, will decide it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
