Here's a number that should stop every CFO and CTO in their tracks: 93% of enterprises now report improved AI production capability — yet 57% still can't generate returns that outpace their investment. And that 57% figure hasn't moved in two years.
That's not a technology failure. That's a delivery failure. The models work. Getting their value to the people who need it — that's where enterprises are stuck.
This is the central finding of the Fifth Annual Domino Enterprise AI Report, published July 21, 2026. The survey covered 639 senior enterprise AI leaders at organizations with $100M+ in annual revenue across North America, the UK, and continental Europe. It's the clearest data I've seen yet on why AI ROI remains so elusive — and what actually separates the companies pulling ahead from those still spinning in place.
The Paradox Nobody Wants to Say Out Loud
Production capability is genuinely improving. Eighty-eight percent of respondents reported improved production capability in 2025. That number climbed to 93% in 2026. More models are running in production than ever before.
But the ROI plateau is equally real. Fifty-seven percent of enterprises report that AI ROI still fails to outpace spend — the exact same share as in 2025. Independent research from MIT NANDA reached a parallel conclusion: 95% of organizations running AI pilots saw no measurable profit and loss impact, with only 5% of integrated pilots generating meaningful financial value.
This is the paradox: your AI infrastructure is more capable than it's ever been, but the business result hasn't moved.
The Domino report gives this a name: the last-mile gap. Deploying a model and delivering its value to the people who need it are not the same thing. For most enterprises, a wide gap separates the two.
What the Last-Mile Gap Actually Looks Like
When leaders were asked how business users actually access AI-generated insights today, the picture was fragmented:
- 34% of organizations report a mix of AI access methods that varies by business unit — the single most common answer. No consistent delivery layer. No standard experience.
- 40% still rely on at least one mediated access method: a scheduled report from a data science team, or a request submitted to an analyst who runs the analysis and returns results.
Think about what that means in practice. A risk model flags a potential issue at 2pm. By the time that insight travels through a scheduled report, gets reviewed by an analyst, and lands in the hands of the business user who can act on it — it's Wednesday. The moment has passed.
This isn't a model problem. It's a workflow problem. And it's costing enterprises the return they were promised.
The Governance Divide Is More Stark Than You Think
Here's where it gets interesting. The Domino report also surfaced something I've seen play out in conversations with enterprise AI leaders: the organizations that built governance infrastructure first are pulling dramatically ahead.
The data is striking. Among organizations whose governance is fully keeping pace with their AI activity:
- 67.5% have agentic AI running in governed production
- 75% report significantly improved AI delivery velocity
Among organizations where governance is only partially keeping pace:
- Only 17.2% have reached governed agentic deployment
- Only 23% report significantly improved delivery velocity
That makes fully governed organizations 3.9 times more likely to have agentic AI running in governed production — and 3.3 times more likely to see velocity improvements.
The conventional wisdom in enterprise AI has been "move fast, govern later." This data makes a compelling case for the opposite.
Agentic AI Is Already Scaling — Without the Guardrails
Here's the tension that keeps me up at night as someone thinking about enterprise AI deployment: agentic AI adoption is outpacing governance adoption. Fast.
Expanding agentic AI use ranks as the top organizational priority for enterprise AI leaders in 2026, tied with upskilling business users — both at 38.5%. But look at where organizations actually are:
- 43% have agentic AI running in governed production
- 29% are actively scaling agentic AI without governance in place
- 12% are piloting agentic AI without governance
More than two in five organizations are running or scaling autonomous AI systems without the governance infrastructure to manage them. And those actively scaling without governance outnumber those merely piloting by more than two to one.
This isn't a startup problem. These are organizations with $100M+ in revenue. Financial services, life sciences, public sector. The sectors where a governance failure isn't just embarrassing — it's a regulatory event.
The Regional Picture Is Telling
The report expanded globally in 2026, and the regional differences reveal something meaningful.
ROI outcomes:
- North America: 51.1% failing to outpace ROI
- UK: 66.9% failing
- Europe: 67.0% failing
North American enterprises are notably more likely to be generating returns. Whether that's faster enterprise AI adoption, different investment levels, or different governance maturity — the gap is real.
Agentic governance:
- Europe has the lowest rate of fully integrated governance at 42.6%
- Nearly half of European organizations are piloting or scaling agentic AI without governance
- Compared to 40% in North America and 38% in the UK
For anyone building enterprise AI strategy across regions, this is worth noting. Europe's regulatory environment (GDPR, incoming AI Act enforcement) combined with lower governance maturity is a risk combination that shouldn't be ignored.
What Separates the 43% Who Are Getting ROI
Looking across the data, three things differentiate the organizations generating AI returns from those stuck at the plateau:
1. Governance infrastructure built before scale
Financial services, banking, and insurance organizations lead every vertical in both governance maturity and production velocity. These are also the most heavily regulated sectors. Regulation forced them to build governance first — and it turns out that governance is a competitive advantage, not just a compliance checkbox.
The enterprises succeeding here didn't retrofit governance onto running systems. They built it in from the start, then scaled.
2. Direct business user access — not mediated delivery
The 40% of organizations relying on mediated AI access (analyst-in-the-middle, scheduled reports) are structurally limited in what ROI they can generate. The signal degrades by the time it reaches the decision-maker.
The organizations pulling ahead are building interfaces that put AI-generated insights directly in front of the business users who need to act on them — underwriters, sales leaders, finance teams, operations managers — not just data scientists.
3. Treating agents as managed entities, not tools
The governance gap in agentic AI isn't about risk aversion. The organizations leading on agentic deployment treat their AI agents as managed, auditable entities: version controlled, monitored, governed within the same platform their data science teams use for development.
The organizations scaling without governance are treating agents like software tools. That distinction matters enormously when an agent makes an autonomous decision in a regulated process.
What Leaders Should Do With This Data
For technical leaders — CIOs, CTOs, Head of AI:
The last-mile gap is an architecture problem. If business users can't directly access AI-generated insights, you haven't completed the deployment. Build the application layer that bridges models in production to the people who act on their output. This is where the ROI gap closes.
For business leaders — CFOs, COOs, business unit heads:
If your AI investments aren't generating return, ask one question: who is actually using the output, and how? If the answer is "our data science team runs reports and sends them to us weekly," you've found the problem. Push for direct access interfaces. Push for applications your teams can use without an intermediary.
For everyone:
The governance-first organizations are winning on both velocity and ROI. Build governance into your AI roadmap before you scale agentic systems — not after you've already shipped 40 agents into production. The RELVE Q2 research this week also noted that most organizations don't have a full inventory of what their AI agents can do autonomously — start there.
The 2026 Enterprise AI Reality
The technology is no longer the bottleneck. Models work. Production capability is improving year over year. The bottleneck is the gap between what AI can generate and what business users can actually access, in time to act on it.
The organizations that close this gap in 2026 will be the ones generating returns in 2027. The ones that spend another year optimizing model performance while leaving the delivery layer unchanged will be back here in twelve months with the same 57% figure.
Two years of the same plateau is a signal, not a coincidence. The question isn't whether your AI works. The question is whether its output reaches the people who can turn it into revenue.
Sources: Domino Data Lab Fifth Annual Enterprise AI Report (July 2026, 639 senior enterprise AI leaders); MIT NANDA research cited within the report; RELVE Q2 2026 State of AI for SaaS Companies report.
Rajesh Beri writes THE DAILY BRIEF — Enterprise AI insights for technical and business leaders. Follow on LinkedIn | Follow on X
Want to calculate your own AI ROI? Try our AI ROI Calculator — takes 60 seconds and shows projected savings, payback period, and 3-year ROI.
Continue Reading
- Why 57% of Enterprises Waste AI Budgets—And How to Fix It
- 40% of AI Projects Get Canceled by 2027. What Saves Yours?
- AI ROI Is Up 31%—But 73% Still Fail at the Basics
- Agentic AI Will Return $17.6M — Only 3% Are Ready
- Why 69% of Enterprises Can't Govern Their Own AI Agents
- 49% of Jobs Changed. Zero Lost. Anthropic Bets $200M.