The AI Spend-ROI Gap: Why Bigger Budgets Keep Buying Smaller Returns

Every budget cycle, the AI line item gets bigger. Every budget cycle, the returns don’t keep pace. That’s not a hunch — it’s now the consistent finding of three separate research efforts run by three different firms, using three different methodologies, in the same twelve-month window. When Bain, McKinsey, and MIT all land on the same conclusion, it stops being an anecdote and starts being a pattern CIOs have to answer for.

The numbers, stacked up

Start with Bain & Company’s Automation and AI Pathfinder Survey, published this June, which polled 951 global companies. The headline finding: 37% of companies targeted cost reductions of 11% to 20% from their automation and AI investments. Nearly 40% of the companies that actually measured their results landed in the 0%-to-10% range instead. The technology shipped. The savings didn’t show up. And despite that gap, 90% of those same companies are increasing their AI budgets again for the next wave — agentic AI — which carries even more complexity and consequence than the automation that came before it.

McKinsey’s State of AI in 2025 survey, released in November and drawing on nearly 2,000 respondents across 105 countries, tells a similar story from a different angle. Eighty-eight percent of organizations now use AI in at least one business function, up from 78% a year earlier. But only 39% attribute any measurable EBIT impact to that AI use — and most of that 39% say the impact is under 5% of EBIT. Only about a third of companies have even begun scaling AI past the pilot stage. Fully scaled deployment: 7%.

Then there’s MIT’s NANDA initiative, whose “GenAI Divide: State of AI in Business 2025” report produced the number that’s been quoted in every boardroom this fall: 95% of enterprise generative AI pilots deliver no measurable P&L impact. The study — built on interviews with roughly 150 executives, hundreds of employee surveys, and an analysis of 300 public AI deployments — found that of the $30 to $40 billion in enterprise GenAI spend it tracked, only about 5% of pilots produced real, bankable value.

Three firms. Three surveys. One conclusion: most companies are spending real money on AI and getting very little of it back.

Why this keeps happening

The most useful thing about this year’s research isn’t the failure rate — it’s that all three reports point to the same root causes, and none of them are “the model wasn’t good enough.”

Companies are automating broken processes. Bain’s researchers call this “workflow debt” — the redundant approvals, unnecessary handoffs, and workarounds every organization accumulates over time. AI doesn’t clean that up. It speeds it up and locks it in, making it more expensive to unwind later. The right question before any AI project gets funded isn’t “where can we apply AI?” It’s “if we designed this process from scratch today, what would it look like?”

Budgets are chasing the wrong use cases. MIT’s research found that most GenAI spend concentrates in sales and marketing pilots — flashy, visible, easy to greenlight — while the highest actual returns show up in unglamorous back-office work: document processing, compliance, internal reporting. Amazon’s finance team, for example, used a generative AI tool to cut the time spent tracking global VAT regulatory updates from 26 minutes per update to 2 — a 92% reduction, with 80% of AI-generated summaries accepted without edits. That’s not a moonshot. That’s a bounded, high-volume, low-glamour workflow where the data already existed.

The autonomy gap between the business case and reality is real. Only 7% of companies in Bain’s survey are running fully autonomous AI agents in production. The dominant model — 38% of respondents — still requires human approval on every action. If the investment case that got board sign-off assumed full-automation economics, and what’s actually running routes a large share of decisions to a human review queue, the CFO approved one set of numbers and the business is living inside a different one.

Companies are self-funding the next wave from savings that never fully materialized. Bain found 44% of companies plan to fund their next round of AI investment from savings generated by the last round — savings that, per the same survey, consistently came in below target. That’s not fiscal discipline. It’s compounding risk on top of an already-optimistic forecast.

Data access is still the wall, and waiting for it to disappear is the wrong move. All three reports flag data access and integration as the top operational barrier to AI value — ahead of budget, skills, and even compliance. But here’s the twist Bain’s data surfaces: companies that hit their targets cite data as a bigger obstacle than companies that missed. The winners aren’t waiting for clean data. They’re running into the data problem harder because they’re actually deploying at scale, and they’re using AI itself to close the gap rather than treating perfect data as a prerequisite.

There’s no named owner for AI accountability. Across the survey base, governance responsibility splits almost evenly between IT, individual business functions, and central teams — meaning, in practice, nobody owns it. When an agent makes a costly decision in production, that accountability question can’t be worked out after the fact.

What this means for the people reading this

None of this is an argument against AI investment. It’s an argument against unexamined AI investment. The companies pulling ahead in all three studies didn’t get there with better models or bigger budgets — they got there by treating data governance, process redesign, and accountability as executive-level decisions made before deployment, not IT cleanup done after a pilot stalls.

If your organization’s AI spend keeps growing while the ROI conversation keeps getting deferred to “next quarter,” that’s worth a hard look now — not after the next budget cycle locks in another round of the same math.

Sources: Bain & Company, “Your AI Budget Is Growing. Your Returns Aren’t. Here’s Why” (June 2026); McKinsey & Company, “The State of AI in 2025: Agents, Innovation, and Transformation” (November 2025); MIT NANDA Initiative, “The GenAI Divide: State of AI in Business 2025” (2025).
Photo by Hitesh Choudhary on Unsplash