For the first time in any enterprise technology survey I can recall, the number is 100%. Not 94%. Not 87%. One hundred percent of chief information officers are now allocating budget to AI and large language model projects, according to a June 2026 survey of more than 100 CIOs by RBC Capital Markets. And yet — somehow — the ROI picture looks nothing like the investment picture.
This is the central paradox of enterprise AI in 2026. Universal buy-in at the boardroom level. Widespread disappointment at the bottom line.
Gartner's most recent data puts a hard number on the gap: only 28% of enterprise AI use cases are meeting ROI expectations. PwC's 2026 Global CEO Survey found that 56% of CEOs have not yet realized measurable revenue or cost benefits from AI. And less than 1% of executives — across thousands surveyed — report what anyone would call significant ROI, defined as a 20% or greater improvement in profitability or cost savings.
So what exactly is going on? And more importantly, what separates the 28% who are winning from the 72% who are still waiting?
The 100% Moment
The RBC survey is worth pausing on because it's genuinely unprecedented. Rishi Jaluria and the RBC Capital Markets tech team survey over 100 CIOs and senior technology leaders twice a year. Earlier iterations raised real concerns about whether enterprise AI was hype or real. This one landed differently.
"We came away encouraged by broad-based enterprise spending momentum into 2H 2026, with AI adoption beginning to transition from pilot to production," Jaluria wrote in the report.
More than half of respondents said AI is already in production at their organizations. Another 35% expect to reach production status within six months. That means by year's end, roughly 85% of these enterprises will have moved past experimentation.
And the money is following fast. Ninety-one percent of survey respondents said they are creating entirely new AI budgets — not reshuffling existing IT spend. AI now represents 1.7% of company revenue on average, more than double the level from 2025. Eighty-six percent of enterprises expect their AI spending to increase further.
The token budget fears? Overblown. Nearly nine in ten CIOs said token budgets are manageable — even though almost half have already exceeded their original spending plans. Rather than pulling back, most plan to spend more.
The ROI Gap Is Real, Not a Narrative
Here's where things get complicated.
The investment is very real. The returns, at scale, are not yet materializing the way the industry promised. Gartner's figure — 28% of AI use cases meeting ROI expectations — isn't a pessimistic outlier. It tracks closely with what PwC, Futurum Research, and multiple other independent surveys are finding.
One data point that stopped me: less than 1% of executives across major surveys report a 20% or greater improvement in profitability or cost reduction from AI. The majority of "positive ROI" being reported is in the 1-5% range — and much of that is measured as productivity gains rather than hard financial impact hitting the P&L.
This matters because productivity gains are notoriously hard to convert to actual cost savings at scale. A developer who can write code 30% faster doesn't necessarily generate 30% more revenue. The business has to be reorganized around that productivity gain for it to show up as profit.
That's exactly what most enterprises aren't doing.
Why the 72% Are Stuck
In conversations with technology and finance leaders across industries, I keep hearing the same underlying issue: companies are adding AI to existing workflows instead of redesigning workflows around AI.
The distinction sounds subtle. The results are not.
Adding AI to an existing customer support operation might reduce average handle time by 15%. That's a real improvement. But if the headcount, tooling, and management structure all stay the same, the savings never materialize. You've made a workflow faster without changing the cost structure.
Companies that are actually seeing measurable ROI are doing something structurally different. They're not asking "where can we use AI?" They're asking "which processes could be fundamentally different if AI were native to them?" Then they redesign the process — headcount, tooling, and all — before deploying the AI.
That's a much harder organizational change. And it's exactly the change that's being skipped.
Three other patterns that correlate with ROI failure:
Poor data foundations. AI produces better outputs when connected to clean, well-structured organizational data. Most enterprises have years of technical debt in their data infrastructure. Deploying AI on top of siloed, inconsistent data doesn't unlock potential — it amplifies the mess.
Missing governance structures. The RBC survey found that hybrid pricing models (seat licenses plus usage-based pricing) have quickly become the preferred enterprise procurement approach. But pricing model sophistication hasn't been matched by AI governance sophistication. Many organizations don't have clear policies on which models to use for which use cases, how to handle sensitive data in prompts, or how to measure outputs. Without governance, you can't optimize.
Measuring the wrong things. The vast majority of AI ROI is currently being tracked as productivity uplift rather than financial outcome. A talking point for strategy decks, but not the P&L impact that justifies further investment. Until enterprises tie AI usage directly to revenue, margin, or cost line items — not just time-savings estimates — they'll continue to undercount successes and struggle to justify scale.
The agentic AI Shift
There's a meaningful inflection point coming that could change the ROI math significantly: agentic AI.
Where generative AI largely augments individual human tasks, agentic AI can orchestrate multi-step workflows autonomously. The difference isn't incremental. Financial services analysts are projecting that agentic AI could deliver a 20% operational efficiency gain for banks. Retailers are already using AI agents for contract negotiation. HR teams are running full candidate screening workflows without human intervention in the first three rounds.
The key difference is that agentic AI makes it structurally harder to "add it to an existing workflow." By definition, agents replace workflow steps. That forces the organizational redesign that most companies are avoiding with standard generative AI.
AWS drove this home at their 2026 New York Summit, with their entire keynote from VP of Agentic AI Swami Sivasubramanian focused on Amazon Bedrock AgentCore — a platform specifically designed to connect AI agents to organizational data, manage production failures, and enforce governance controls as agents scale. The message from one of the world's largest cloud providers: agentic AI isn't the future, it's the current product roadmap.
For enterprises serious about moving from the 72% to the 28%, agentic AI represents the clearest architectural path to doing so.
What the 28% Are Actually Doing
Looking across the data and peer conversations, the pattern among enterprises with strong AI ROI isn't mysterious. It's just disciplined.
They start with the P&L, not the technology. The winning enterprises aren't asking "where can we use AI?" They're identifying specific line items — a cost center, a revenue inefficiency, a customer churn problem — and then working backward to which AI capabilities could move that specific number.
They restructure processes, not just tools. A change management program accompanies every significant AI deployment. If the headcount or workflow structure doesn't change, the savings don't materialize.
They connect AI to trusted data. Before deploying, they've cleaned and structured the data that AI will operate on. This often means 3-6 months of data infrastructure work before any model is deployed. Unglamorous. Absolutely essential.
They measure outcomes, not activities. The metric is never "prompts submitted" or "hours saved by AI tools." It's "customer support cost per ticket," "sales cycle days," "time to close a financial period." Hard financial outcomes, not AI activity metrics.
They govern before they scale. Clear policies on model selection, data handling, and output review are in place before broad deployment. This reduces rework, compliance risk, and the cost of fixing mistakes at scale.
The Market Context
A few additional data points from the RBC survey that matter for enterprise leaders:
OpenAI continues to dominate enterprise adoption. Fifty-seven percent of CIOs named ChatGPT as their most-used AI model-based service. Anthropic's Claude comes in at 12%. On performance perception, 44% name OpenAI as the highest-performing provider versus 24% for Anthropic. The gap is large enough that it has real procurement implications for any enterprise standardizing on a single provider.
The "SaaSpocalypse" narrative — the idea that AI spending would cannibalize traditional software budgets — is not materializing. Zero respondents expect to cut software spending. The enterprises spending more on AI are largely funding it through net-new budget creation, not by eliminating existing software investments.
Hybrid pricing (per-seat plus usage-based) has rapidly become the dominant enterprise preference. This is a structurally important shift because usage-based pricing gives finance teams visibility into actual consumption and enables tighter ROI measurement per use case.
What Leaders Should Do Now
For CIOs and CTOs building their H2 2026 plans, the data suggests a clear set of priorities:
Move one process to full agentic redesign before the end of Q3. Not an experiment — a production deployment where agents own a workflow step. The learning from that deployment will be worth more than six months of generative AI pilots.
Audit your current AI spend against P&L outcomes, not productivity estimates. If you can't map each AI deployment to a specific financial metric that's moved, you're in the 72% and you need to know it.
Fix data infrastructure in parallel with AI deployment. Every dollar spent cleaning and structuring organizational data for AI use has higher leverage than the same dollar spent on model licenses.
Build governance now, before scale forces it. Hybrid pricing means you'll have detailed usage data. Build the policies and monitoring to use that data proactively.
The Bottom Line
One hundred percent of CIOs funding AI is genuinely remarkable. Six months ago, the narrative was still about whether enterprise AI investment would sustain. That question is settled.
The new question — the one that will separate enterprise winners from laggards over the next 18 months — is which companies convert that investment into the 28% that actually hit ROI targets.
The gap isn't a technology problem. The technology is real, it's in production, and it's capable of delivering the results being promised. The gap is an organizational change management problem. The companies that treat AI deployment the same way they'd treat any major business transformation — with process redesign, data infrastructure, governance, and outcome measurement — are going to be the ones writing the case studies in 2027.
The other 72% will still be talking about productivity gains.
Sources: RBC Capital Markets CIO Survey (June 2026, 100+ respondents); Gartner Enterprise AI ROI Analysis 2026; PwC Global CEO Survey 2026; NVIDIA State of AI Report 2026; AWS Summit New York 2026 keynote
Rajesh Beri is an enterprise AI leader and founder of THE D*AI*LY BRIEF, a twice-weekly newsletter on Enterprise AI for technical and business leaders. Connect on LinkedIn or X/Twitter.
