Here is the most uncomfortable statistic in enterprise AI right now: 62% of business leaders use AI multiple times every day — yet only 17% can point to clear ROI from their AI investments. That is not a rounding error. That is a structural failure. And it is happening inside organizations that are convinced they are winning.
Zip's State of AI in Spend 2026 — a survey of 1,050 global enterprise leaders across procurement, finance, IT, and operations — landed today with a finding that should reframe every AI budget conversation you have this quarter. The companies pulling ahead are not doing more AI. They are doing it differently. And the gap between those two groups is widening at exactly the wrong moment.
The 17% Problem
Over 80% of enterprise leaders say they expect to succeed with AI. Most also admit they lack the skills to back that up. That contradiction is not a minor inconsistency — it is the defining tension of enterprise AI in 2026.
The Zip report frames the divide cleanly: Builders versus Bystanders.
Builders are leaning hard into AI transformation. They are not running one proof-of-concept per quarter and calling it a strategy. They are deploying broadly, integrating deeply, and treating AI readiness as a core operational capability — not an IT initiative.
Bystanders know change is coming. They are genuinely interested. They have approved the pilots. But they are still in pilot mode twelve months later, waiting for some signal that it is safe to commit.
The ROI gap between these two groups? 6x. Not 20% better. Not 2x. Six times the return from the same underlying technology — the difference being depth of commitment, not size of investment.
What "Depth, Not Dollars" Actually Means
The headline of the Zip report is worth repeating: AI's real payoff comes from depth, not dollars. That cuts against how most enterprises are thinking about this.
The typical enterprise AI conversation goes like this: How much should we budget? Which vendors should we select? Which department gets the pilot first? These are real questions, but they are the wrong starting point. They optimize for spend, not for transformation.
Builders are asking different questions. They are mapping AI to specific business outcomes — not productivity in general, but this process, this decision, this bottleneck. They are measuring before and after. They are building feedback loops between AI performance and business metrics. They are not waiting for the technology to prove itself; they are engineering the conditions under which it can.
Bystanders are doing the opposite. They are measuring AI adoption (how many people use it) rather than AI impact (what changed because of it). They are running pilots that never reach decision quality. They are treating AI as a feature of their existing software stack rather than a capability that requires organizational rewiring.
This distinction matters because it explains why the ROI gap exists. The 6x advantage Builders have is not because they bought better tools. It is because they built the operational context in which those tools can deliver.
The Shadow AI Crisis Nobody Is Admitting
Here is the finding from the Zip report that should concern every CIO and CISO reading this: 57% of enterprise leaders are using AI tools their company has not approved.
More than half. At the leader level. Not interns experimenting on their laptops — VPs and directors running procurement, finance, and operations decisions through unauthorized AI tooling.
This is what governance failure looks like in practice. It is not a malicious act. These leaders are trying to do their jobs better. They found a tool that helps. They used it. Nobody asked them to stop.
The risk is real and layered. Proprietary financial data, vendor contracts, customer information, competitive strategy — all potentially flowing into systems that have not been vetted by legal, security, or compliance. And because it is happening at the leadership level, the usual data-loss-prevention controls often do not apply.
The more interesting question is not "how do we stop this" but "what does this tell us?" The Zip report frames shadow AI as a potential leading indicator of demand — people reaching for unauthorized tools because the authorized options are too slow, too limited, or too hard to access. That is an IT governance failure, not a user discipline failure.
For technical leaders, the 57% number is a forcing function. If your employees are finding workarounds faster than you are building guardrails, your AI governance strategy is at least six months behind where it needs to be.
AI Has Become a Hiring Gate
One of the subtler findings in the Zip report deserves its own analysis: nearly three in four organizations now factor AI proficiency into hiring decisions.
That is not surprising. What is surprising is the direction the pressure is coming from. 17% of organizations now require managers to prove that AI cannot do a job before they are permitted to hire a human for it.
Read that again. The default assumption is shifting from "hire a human unless we have AI" to "use AI unless we can prove a human is necessary."
This is not hypothetical future policy. It is happening today, at enterprise scale, across procurement, finance, IT, and operations. It has direct implications for headcount planning, skills investment, and organizational design.
For CFOs, this is immediately relevant to workforce cost modeling. If AI-first hiring becomes the norm across your peer group, the financial assumptions underlying your current headcount plans are outdated. The question is not whether AI will displace certain roles — that debate is over. The question is how fast it will happen in your sector and whether your talent strategy accounts for it.
For CIOs and heads of AI, this creates both opportunity and obligation. Opportunity because AI-first hiring requires AI infrastructure that actually works — which means IT has a seat at the workforce planning table it has never had before. Obligation because if the infrastructure fails, the hiring strategy fails with it.
The Technical Leader's Lens
For CIOs, CTOs, and heads of engineering, the Zip data connects to a broader pattern visible across multiple 2026 surveys.
Gartner recently estimated that $234 billion in enterprise software spending is at risk from agentic AI by 2030 — describing the phenomenon as "agentic arbitrage," where AI agents completing tasks across systems reduce the need for human interaction with traditional software interfaces. This breaks the seat-license model that underpins most enterprise SaaS procurement.
At the same time, Gartner estimates that 40% of enterprise agentic AI projects are at risk of cancellation by 2027 due to governance gaps, unclear ROI, and escalating compute costs. These two forecasts are not contradictory — they describe the same split between Builders and Bystanders playing out at the vendor and deployment level simultaneously.
For technical leaders, this creates a specific set of decisions to get right in the next 12 months:
Architecture choices matter more than they did last year. AI agents built on proprietary vendor stacks lock you into seat-license economics at exactly the moment those economics are being disrupted. Building on open, composable infrastructure gives you the flexibility to shift as the market evolves.
Governance is not a compliance checkbox. The 57% shadow AI figure means your governance strategy is already playing catch-up. The goal is not to restrict access — it is to make the approved path easier than the workaround. That requires investment in tooling, training, and approval workflows that actually move at the speed of the business.
Production is different from pilot. BCG and Forrester data shows a median 5.1-month payback on AI agent deployments — but that clock starts when you ship to production, not when you start the pilot. Every month spent in evaluation is a month of compounding ROI you are not capturing. The Builders in the Zip survey understand this. The Bystanders do not.
The Business Leader's Lens
For CFOs, CMOs, COOs, and functional leaders, the Zip findings translate directly into budget and strategy decisions.
The "AI confidence" problem is a measurement problem. Over 80% of executives believe they will succeed with AI, but only 17% can demonstrate ROI today. That gap is not caused by bad technology or bad vendors — it is caused by organizations that are measuring the wrong things or measuring nothing at all.
ROI from AI is not the same as ROI from traditional software. Traditional software automates a fixed process. AI improves decision quality over time — which means the ROI compounds as the system learns your business context. Measuring AI ROI like you measure an ERP implementation will always produce misleading results.
The leaders getting 6x returns are almost certainly measuring differently. They are tracking decision quality, not just process speed. They are measuring error rates before and after. They are calculating the cost of decisions that AI helped avoid — supplier disputes, compliance violations, procurement fraud — not just the cost of tasks AI completed faster.
The 12-18 month window is real. The Zip report notes that a procurement and finance window is opening as AI governance shifts ownership across the organization. The same window applies broadly: every function that establishes its AI governance framework and measurement approach in the next 12-18 months will hold institutional knowledge and process ownership that latecomers will struggle to replicate.
This is not a technology advantage. It is an organizational learning advantage. And it compounds.
What Separates Builders from Bystanders
Synthesizing the Zip data with the broader 2026 enterprise AI landscape, the pattern is clear. Builders share five characteristics that Bystanders consistently lack:
1. They measure outcomes, not adoption. Active users and hours spent in AI tools are vanity metrics. Builders track decisions improved, errors prevented, cycle times shortened, and dollars saved or earned.
2. They treat governance as infrastructure. Shadow AI is not a user problem in Builder organizations — because the approved path is better than the workaround. They invested in making compliance easy.
3. They commit to production faster. Pilots have 90-day windows, not 12-month evaluation cycles. Builders accept some risk of imperfect deployment in exchange for the learning that only production provides.
4. They connect AI to specific business problems. Not "we are deploying AI to procurement" but "we are using AI to reduce our invoice processing cycle from 14 days to 2 days, which frees up $X in working capital." The specificity is the point.
5. They make AI proficiency a hiring and promotion criterion. This is not about culture. It is about incentive alignment. When AI capability affects career outcomes, adoption and skill development follow without mandates.
What to Do This Quarter
The data from the Zip report is clear enough to act on immediately. Three moves that separate Builders from Bystanders at the organizational level:
Audit your measurement approach. If you cannot articulate the specific business outcome AI is improving — with a before and after number — you are a Bystander regardless of how many AI tools you have licensed. Pick one process, define the metric, and measure.
Map your shadow AI exposure. Survey your leadership team (anonymously if needed) on what unauthorized AI tools they are using and for what. What you find will be uncomfortable. It will also tell you exactly where your governance gaps are and where the authorized alternatives need to catch up.
Set a production deadline. If you have a pilot that has been running for more than 90 days without a go/no-go decision, schedule the decision. The cost of staying in pilot mode is not zero — it is the compounding ROI you are not capturing.
The Bottom Line
The 83/17 split is not a technology story. The Builders and Bystanders in the Zip survey have access to the same AI platforms, the same vendors, and the same market intelligence. What they do not share is organizational commitment, measurement discipline, and governance infrastructure.
The 6x ROI gap will not close by waiting. Every quarter in pilot mode is a quarter the Builders are compounding their advantage — in institutional knowledge, in process optimization, in the talent they are attracting with AI-forward culture.
The uncomfortable truth is that most enterprise AI programs are structured to produce exactly the 17% outcome they are getting: broad deployment of tools, shallow integration into workflows, and measurement frameworks designed to prove activity rather than impact.
Changing that requires decisions at the leadership level, not the technology level. The technology is ready. The question is whether the organization is.
Source: Zip's State of AI in Spend 2026, survey of 1,050 global enterprise leaders across procurement, finance, IT, and operations. Published July 23, 2026. Additional data from Gartner Agentic AI Enterprise Forecast (July 2026), BCG/Forrester AI Agent ROI Survey (2026).
Connect with Rajesh Beri on LinkedIn or follow on X/Twitter for daily enterprise AI insights.
