9 Out of 10 Companies Spend on AI.
Fewer Than 4 Prove It Works.

88% of organizations use AI in at least one business function. Only 39% report a significant impact on EBIT.

That means roughly 9 out of 10 companies are spending on AI. Fewer than 4 in 10 can prove it matters to the bottom line. The rest are writing checks to vendors and calling it transformation.

Not adoption. Not capability. Value. The money people want receipts.

This tension is arriving at the worst possible moment for enterprise buyers. Anthropic just surpassed OpenAI in valuation. New models like Claude Opus 4.8 are accelerating deployment pressure. Vendors are shipping faster than buyers can absorb. And the gap between what companies spend on AI and what they earn from it is getting wider, not narrower.

I think this is the most important story in enterprise technology right now. Not which model is biggest. Not which agent framework is coolest. Whether the money works.

The Pilot Purgatory Ratio

Here is a framework for diagnosing the problem. Call it the Pilot Purgatory Ratio: the number of AI initiatives inside your organization divided by the number tied to a specific P&L line item.

Most companies have a ratio north of 10:1. Ten experiments for every one deployment that a CFO can trace to revenue, margin, or cash flow. BCG's September 2025 research found that only 4 to 5% of companies are "future-built" for AI and generating substantial value. The other 95% are running pilots, buying licenses, and hoping.

The Pilot Purgatory Ratio exposes something structural. Adoption is easy. Extraction is hard. Buying a seat on an AI platform takes a credit card. Redesigning a workflow, retraining a team, rebuilding a data pipeline, and measuring the outcome against a baseline takes quarters of unglamorous work.

PwC's 2026 AI predictions report frames it clearly: front-runner companies adopt enterprise-wide AI programs with top-down sponsorship and clear business outcomes. Laggards scatter pilots across departments with no owner, no metric, and no accountability. The technology is identical. The operating model is not.

If your ratio is above 5:1, you are in pilot purgatory. If it is above 10:1, you are funding a science fair.

Why the Finance Function Is Taking the Wheel

The shift happening inside the C-suite right now is not subtle. It is a transfer of authority. AI deployment decisions are moving from the CTO's whiteboard to the CFO's spreadsheet. Builders who do not understand this shift will find their budgets cut before Q3.

Here is what changed. Global AI spending is projected to hit $2.52 trillion in 2026, a 44% year-over-year increase according to market data compiled by The Business Research Company. AI venture funding consumed $225.8 billion in 2025, which was 48% of all venture dollars globally. AI-optimized server spending is expected to jump 49% in 2026. These are not incremental numbers. These are capital-intensive bets that show up on balance sheets and income statements.

When spending reaches this scale, the finance function does what it always does. It asks three questions. What did we spend? What did we get? What should we spend next?

The answers are uncomfortable. Gallagher's 2026 AI Adoption and Risk Survey found that most companies expect meaningful ROI to take two to three years to materialize. KPMG reported that only 15% of executives have even set ROI expectations for generative AI, despite 44% of companies scaling it. You cannot measure what you never defined.

CFOs are now demanding what I call capital-like discipline for AI. That means clear payback periods. Full-cost accounting that includes data engineering, change management, and vendor lock-in, not just license fees. And direct linkage to line-item metrics like days sales outstanding, churn rate, or claims leakage. Not "productivity." Not "efficiency." Specific numbers on specific lines.

This is the Replacement Ladder applied to organizational authority. When AI was a $50,000 experiment, the innovation team owned it. When it became a $50 million line item, the CFO took the chair. The builders who survive this transition are the ones who learn to speak finance, not the ones who complain about it.

Harvard Business School faculty, writing in December 2025 about AI trends for 2026, emphasized that success depends on "change fitness" and the ability to balance trade-offs between speed and control. Professor Tsedal Neeley put it directly: "AI is no longer the experiment on the side; it's rewiring how work gets done." When something rewires how work gets done, the person who signs the checks wants to understand the wiring diagram.

But here is where the story gets complicated. It is genuinely unclear whether CFO-driven accountability will close the value gap or accidentally widen it. The strongest counterargument comes from BCG's own data. The future-built companies generating outsized AI returns invest 2x as much as laggards and reinvest gains into further AI build-out. They operate on portfolio logic: some bets fail, a few generate enormous returns, and the platform capabilities built along the way reduce the marginal cost of every future initiative.

If CFOs impose project-by-project ROI requirements on every AI experiment, they risk killing the portfolio logic that winners exploit. They optimize for local, short-term cost savings and miss the systemic, long-term value. They fund the safe chatbot and starve the supply chain redesign. They measure the easy thing and ignore the important thing.

The right accountability target is not "builders" in isolation. It is the end-to-end value chain from business sponsor to data team to operations to change management. When a pilot fails, the question should not be "why did the model underperform?" It should be "why did we deploy a model without a defined business owner, a baseline metric, and a plan to change the workflow around it?"

The companies stuck in pilot purgatory are not there because their engineers are bad. They are there because nobody connected the engineering to the economics.

2031

Three signals inside the same shift

Zoom out five years. The pattern playing out in AI is not new. It is the same pattern that played out with cloud computing, ERP systems, and the commercial internet. Broad adoption. Uneven returns. A shakeout. Then a new equilibrium where the winners are structurally different from the losers.

By 2031, I expect the AI landscape to look like this. A small cohort of companies, maybe 10 to 15%, will have rebuilt their operating models around AI in ways that create compounding advantages. They will have centralized data platforms, reusable model architectures, and cross-functional teams where finance, engineering, and operations share accountability for outcomes. Their Pilot Purgatory Ratio will be close to 1:1 because they stopped running experiments and started running systems.

The rest will have consolidated their AI spending into a handful of vendor-managed tools. They will get moderate productivity gains. They will not achieve transformation. The gap between the two groups will be structural, not tactical. You cannot close a five-year operating model gap with a better prompt.

The asymmetric bet here is not on any specific model or vendor. Anthropic might lead today. Someone else might lead in 2028. The asymmetric bet is on organizational capability. The companies that learn to connect AI deployment to financial outcomes, that build the boring infrastructure of measurement and change management, that treat AI as a business discipline rather than a technology experiment, those companies will compound their advantage every year.

The generative AI market is projected to grow from $91.6 billion in 2026 to $400 billion by 2030. That is a 34.3% compound annual growth rate. The money flowing into this space is not slowing down. The question is whether it flows toward value or toward purgatory.

Think of it like Costco's hot dog. The $1.50 price has not changed since 1985. That is not a pricing decision. That is a strategic commitment to a specific kind of customer relationship, maintained through decades of operational discipline. The companies that win the AI era will make a similar commitment: not to a specific tool, but to the discipline of connecting every AI dollar to a measurable outcome. Impermanence applies to models and vendors. Discipline compounds.

What to Build This Weekend

You do not need a $2 trillion budget to start closing your own value gap. You need a spreadsheet and two hours.

Step one. List every AI tool your team pays for. Include subscriptions, API costs, and any internal compute. Most teams cannot do this from memory, which is itself a diagnostic.

Step two. Next to each tool, write the specific business metric it is supposed to improve. Not "productivity." Not "efficiency." A number. Revenue per rep. Support tickets resolved per hour. Time to first response. If you cannot name the metric, flag that tool in red.

Step three. For every red-flagged tool, decide within 30 days: define a metric and a baseline, or cancel the subscription. This is not about cutting AI spending. It is about making AI spending legible to the person who controls your budget.

Step four. Pick one green-flagged tool, the one with the clearest metric, and build a simple tracking dashboard. If you want a no-code option, try a basic Notion database or Google Sheet that pulls weekly data. The point is not sophistication. The point is visibility.

Step five. Share the dashboard with your finance counterpart. Not as a defense. As an invitation. The builders who bring the CFO into the conversation early are the ones who keep their budgets when the rationalization wave hits.

If you want to practice building something small and visual this weekend, Pixley AI is a fun, low-stakes tool that turns kids' drawings into animated cartoons in minutes. It has nothing to do with enterprise ROI. But it will remind you what building feels like when the stakes are low and the feedback is immediate. Sometimes you need that before Monday.

The value gap is real. The CFO pressure is real. But the fix is not mysterious. Define what value means. Measure it. Connect every deployment to a number someone in finance cares about. Simple scales. Complex fails. Get your reps in.