The Productivity Lie
Companies spent $37 billion on generative AI in 2025 — 3.2 times more than 2024. Two-thirds report productivity gains. Sounds like success, right? Except Deloitte found something that should make every executive pause: these productivity gains aren't translating into revenue growth. Companies are scaling AI deployments without actually making more money.
Let that sink in. Billions spent. Productivity up. Revenue... flat.
The Vanity Metric Trap
The problem is that most companies measure AI success with vanity metrics. "We're 30% faster at generating reports." Great — but were those reports producing revenue? "Our team processes 50% more documents." Wonderful — but did processing more documents create more business value?
Being faster at the wrong thing is still doing the wrong thing. Being more efficient at a process that shouldn't exist is just organized waste. And this is where the majority of AI investment is going: making existing processes faster, rather than asking whether those processes should exist at all.
MIT's research is blunt about this: "Organizations must move beyond substituting technology for current employee work and embrace fundamental redesign of work." The companies that succeed with AI don't bolt it onto existing workflows. They redesign the workflow from scratch, with AI as a core component.
The Three Ways Companies Waste AI Money
The Spray-and-Pray: "Let's give everyone access to AI and see what happens." What happens is that some people use it for trivial tasks, most people forget about it after a week, and nobody achieves the transformative results that justified the investment. Without clear use cases and measurement, AI access becomes the corporate equivalent of a gym membership in February — purchased with great enthusiasm, abandoned by March.
The Ivory Tower: "Let's build an AI Center of Excellence and have them figure it out." A team of data scientists builds impressive demos that work in controlled environments and fail in the real world. The demos impress leadership. The production deployment encounters dirty data, legacy system incompatibilities, and employee resistance. The center of excellence becomes a center of expensive experiments that never reach the people who could actually benefit.
The Compliance Theater: "We need AI governance before we do anything." The organization spends months writing AI policies, forming ethics committees, and creating approval processes. By the time the governance framework is in place, the competitive window has closed. The company has a beautiful AI policy and zero AI-generated revenue. Only 1 in 5 companies have mature AI governance — but the answer isn't to wait for perfect governance. It's to build governance alongside deployment.
What Actually Works
The companies succeeding with AI share four habits:
They start with money, not technology. Every successful AI project begins with a specific financial target: "reduce inventory waste by $X" or "cut document review time by Y hours." If you can't put a number on the expected outcome, don't start the project.
They give it to the people who do the work. Not a separate AI team — the actual workers who will use it daily. The best AI implementations happen when accountants build their own AI workflows, marketers design their own AI prompts, and operations managers configure their own automations. The people closest to the work know the nuances that an outside team will miss.
They expect the dip. MIT documented a temporary productivity decline of 1.33 percentage points during AI adoption — before gains materialize. Companies that don't expect this dip panic at month 3 and kill the project. Companies that budget for it push through and reach the gains.
They measure weekly. Not quarterly. Weekly. What's working? What's not? What do we adjust? The companies that measure frequently iterate faster and course-correct before small problems become expensive failures.
Your Move
In your next meeting about AI, ask one question: "What specific revenue or cost outcome will this produce in 6 months, and how will we measure it?" If nobody can answer, you've just identified why the project will likely fail — and potentially saved your company from joining the 95% that waste their AI investment. If you can answer it, you're already in the 5%.
The companies that win with AI don't have better technology. They have better questions and the discipline to answer them before writing the first check.