The productivity story is real. Workers are genuinely faster. Drafting takes half the time. Research tasks that used to eat a morning are done before the first coffee. Support tickets get resolved before the queue backs up. The tools are working. McKinsey's data confirms this — 80% of workers say AI has made them more productive, a number that has climbed consistently as the tools have improved.
The earnings story is also real. Of the 1,719 executives McKinsey surveyed, only 6% qualify as what they call "AI high performers" — meaning they can attribute at least 5% of their EBIT directly to AI use. Worldwide AI spending is forecast to hit roughly $2.5 trillion in 2026. Enterprise generative AI spending alone reached $37 billion in 2025, tripling the prior year. You would expect that kind of investment to show up in earnings results. It largely hasn't.
This is not a contradiction in the data. It's a description of exactly where the value is going and where it stops. Between "worker is more productive" and "company earns more money" are three specific gaps, and most businesses — from 10-person startups to 10,000-person enterprises — are falling into all three of them simultaneously.
Gap One: Efficiency That Has Nowhere to Go
When an individual worker becomes 30% more productive, one of two things happens: they use that time to do more value-generating work, or they fill it with more of the same work at a slower pace. In most organizations, the second scenario is the default — not because people are lazy, but because there's no structural mechanism to redirect the freed capacity toward anything that moves revenue.
A sales rep who can draft a follow-up email in two minutes instead of ten doesn't automatically close more deals. They send better emails, faster. But if the bottleneck in the sales process is actually in demo scheduling, or in the sales leader's ability to review pipeline, or in the product's ability to convert trials — speeding up email drafting changes almost nothing downstream. The efficiency gain evaporates into the ambient busyness of the role.
PwC's 2026 AI Performance Study, which surveyed over 1,200 senior executives, found that the companies actually capturing AI value were roughly twice as likely to redesign workflows around AI rather than simply add AI to existing ones. That distinction sounds subtle. It's not. Adding AI to an existing workflow makes the existing workflow faster. Redesigning a workflow around AI means asking what the workflow is actually trying to accomplish and whether AI changes how you should structure the work to accomplish it.
The companies that redesigned their sales process for AI — not just giving reps an AI writing tool, but rethinking how leads are qualified, how demos are structured, how follow-up is triaged — are the ones seeing it in their pipeline. The ones who handed out ChatGPT access and called it an AI strategy are not.
Gap Two: AI Pointed at the Wrong Problems
The second gap is more uncomfortable to name: most AI deployment in businesses is pointed at tasks that don't directly move revenue. It's focused on admin reduction, content creation speed, internal communication, and documentation. These are real costs worth cutting. But they're not the core of what makes a business grow.
QuickBooks and Intuit's 2026 AI Impact Report, built on survey responses from over 34,000 business owners and anonymized data from 5.3 million businesses, found that 43% of US businesses report AI has increased their revenue — not just cut costs. That's a meaningful number. It also implies that 57% of businesses using AI are getting efficiency gains but not revenue growth. The businesses reporting revenue increases are the ones that have pointed AI at customer-facing processes: acquisition, conversion, retention, upsell.
McKinsey's data on "AI leaders" — the 6% with significant earnings impact — confirms this directional difference. Leaders in McKinsey's survey pointed AI at growth, not just efficiency. They used AI to expand into new customer segments, accelerate product development cycles, and improve customer experience in ways that reduced churn. The laggards used AI to make internal operations leaner. Both groups are using the same tools. One group is using them on the part of the business that generates revenue; the other is using them on the part that supports it.
For a founder-led business, this translates to a diagnostic question: of the AI workflows you've built or bought in the last 12 months, how many are reducing internal overhead, and how many are directly improving what a customer experiences or what a sales process looks like? The answer tells you almost everything about why you're or aren't seeing it in your results.
Gap Three: AI Accelerating Broken Processes
The third gap is the one that's hardest to see from inside: AI is being used to run broken processes faster, not to fix them. A customer support workflow with unclear escalation paths doesn't get better when you add an AI triage bot — it gets faster at routing customers to the wrong place. A sales process with an undefined ideal customer profile doesn't improve when AI helps reps draft more outreach — it scales the inefficiency of reaching out to bad-fit prospects.
This is what PwC means when it says the companies getting AI value are focused on "redesign and foundations, not just tools." The foundation — the underlying process logic, the data quality, the decision rules — has to be sound before AI acceleration has anywhere productive to go. Layering AI on a process that was already struggling produces a faster version of the same struggle.
McKinsey's data includes a finding that illustrates this precisely: the 6% of companies with significant earnings impact are 2.8 times more likely than average to have increased the number of decisions made without human intervention. That's not about removing humans from the loop for its own sake. It's about identifying the decisions where the process logic is clear enough, and the data reliable enough, that AI can own the decision outright — which forces you to articulate the process clearly enough to encode it. Companies that can automate decisions have, by definition, done the work of making those decisions explicit and reliable. Companies that can't automate decisions usually have processes that don't hold up to that scrutiny.
What the Top 6% Are Actually Doing
The companies capturing meaningful earnings impact from AI share a pattern that McKinsey's study makes legible. It's not about which AI tools they're using — the tools are largely the same across the spectrum. It's about how they deploy them.
They identify two or three high-leverage workflows and put focused investment there. Not 15 AI tools running in parallel across the organization, but a small number of workflows where the process is well-understood, the data is clean, and the connection to revenue is direct. The Gartner prediction that 40% of enterprise applications will feature task-specific AI agents by end of 2026 — up from less than 5% in 2025 — reflects this shift. The move is toward specialized AI that does one thing reliably, not general AI tools that can theoretically do everything but are pointed at nothing specific.
They measure outcome, not activity. Most companies measure AI adoption by how many employees are using the tools and how frequently. That's activity measurement. The companies with earnings impact measure whether the workflows they've built AI into are producing better outcomes: more deals closed, lower support cost per ticket, faster time to close, higher customer retention. The measurement discipline forces the question of whether the AI deployment is actually connected to a business outcome — and surfaces quickly when it isn't.
They redesign before they automate. The sequence matters more than most organizations realize. Adding AI to an existing workflow is much faster than redesigning the workflow first. It's also much less likely to produce earnings impact. The companies getting results are doing the slower, less exciting work of asking what the workflow should look like before they add AI to it — and often finding that the workflow was poorly designed to begin with, in ways that weren't obvious until they tried to encode it.
They target growth, not just efficiency. The 7.2x revenue and efficiency multiple that PwC's study found for AI leaders over laggards isn't primarily explained by cost reduction. Leaders are using AI to do things they couldn't do before: respond to every inbound lead within five minutes (automated triage and initial response), personalize outreach at a scale that would have required a team three times larger, run pricing experiments continuously, identify at-risk customers before they churn. These are expansion moves, not cost-cutting moves. They show up in revenue growth, not just margin improvement.
The Founder Version of This Problem
Most of what McKinsey found in large enterprises applies at founder scale, with one important difference: the correction is much faster to make.
A 500-person company that has been running AI tools for 18 months in fragmented, individual-use patterns has a change management problem that takes quarters to solve. A 10-person founder-led business that's been doing the same thing can audit its AI stack in an afternoon, identify the two workflows worth redirecting toward revenue, and start the redesign next week. The structural advantages of small size — faster decisions, no internal politics, direct founder ownership of every workflow — are precisely the advantages that make the correction from gap one, two, and three achievable without a multi-quarter initiative.
The QuickBooks data showing 43% of small businesses reporting AI-driven revenue increases suggests that a meaningful portion of founder-led businesses have already found the pattern that works. The ones that haven't share a common diagnosis: they have AI tools deployed on tasks that don't move the needle, in processes that weren't clean to begin with, with no measurement in place to surface the problem.
The honest version of what to do with McKinsey's new data is not to spend more on AI. Enterprise spending at $186 million average budgets isn't producing earnings for 94% of companies. The answer isn't more spend; it's better targeting of what you already have. Map the AI workflows you're running. For each one, write down what business outcome it's connected to and how you'd know if it was working. The ones where you can't answer that question clearly are the ones burning budget without producing results. Fix those first, and don't add new tools until you can.
The One Number Worth Watching
McKinsey's 6% figure — the share of companies attributing significant earnings impact to AI — is a useful benchmark, but it's also a moving target. In 2024, when most organizations were still in pilot mode, the gap between individual productivity gains and business earnings impact was widely accepted as temporary: once adoption scaled, the earnings would follow. It's now mid-2026. 88% of organizations use AI in some capacity. The earnings impact hasn't scaled with adoption. That's the finding that matters.
The implication is that adoption alone was never going to close the gap. The gap is structural — it lives in the workflow design, the targeting decisions, and the measurement discipline, not in the adoption rate. Organizations that close it won't do so by adding more AI tools. They'll do so by doing the less glamorous work of making their processes explicit, pointing AI at the right problems, and measuring what actually changes in the business as a result.
For founders, that work is an advantage if you do it and a significant competitive liability if you don't. The businesses that are quietly closing the gap between productivity and profit right now are building an operational lead that will compound. The ones waiting for the next AI product announcement to solve the problem for them are falling further behind every quarter.
If you're not sure where your AI spending is going or why it's not showing up in your results, that's the conversation we're built for. We'd rather spend an hour diagnosing your stack than sell you more tools. Start here.
Related: Your Team Uses AI Every Day. Your Business Doesn't. Atlassian Just Measured the Gap. | The AI Performance Gap Is Real — Here's Which Side of It You're On