Gartner London Data and Analytics Conference 2026: The AI ROI Problem Is a Data Problem

When a room full of 3,000 data executives bursts into applause at the declaration that dashboards are dead, you know the industry has arrived somewhere new.

That was the opening move at this year's Gartner Data & Analytics Summit in London — and it set the tone for three days of unusually candid conversations about where enterprise AI actually stands versus where organizations think it stands. The gap between those two things turned out to be the story of the week.

Why most enterprise AI still shows no ROI

Gartner VP Analyst Adam Ronthal and Director Analyst Georgia O'Callaghan opened Monday's proceedings with a frame that was equal parts rallying cry and warning shot: AI is accelerating into every part of the enterprise, but the infrastructure beneath it is not keeping up. Four out of five organizations have increased their AI investments this year. Only one in five can demonstrate measurable ROI.

The analysts' diagnosis was pointed: the bottleneck is not the model. It is context. Not context in the fluffy sense, but context as critical infrastructure: the governed, organized, semantically enriched layer of business meaning that AI agents need before they can be trusted to act autonomously. Without it, agents make incorrect assumptions, and they hallucinate with confidence. Without it, the more autonomy you give them, the faster they compound your data problems at scale.

AI success isn't about better models, the keynote framing made clear. It's about giving agents governed, contextual access to the right data, at the right time.

Fifty-seven percent of IT leaders, the analysts noted, report being pushed to adopt AI before they are organizationally ready. Only 14% are confident their data is properly secured and governed.

Governance is the engine, not the brake

Here is the thing that kept surfacing across sessions, roundtables, and the expo floor: every organization that has successfully deployed AI at scale did the governance work first.

Gartner Director Analyst Anurag Raj introduced a framework that may be the most useful mental model to come out of the conference: governance has split into three horizons: 

• Governance of AI, which sets policies and constraints for AI systems. 

• Governance by AI, which uses AI agents to automate and accelerate governance tasks themselves. 

• Governance for AI, which prepares data so that agents can use it reliably. 

That third horizon deserves particular attention. Creating AI-ready data has emerged as its own discipline: the structured, ongoing work of ensuring data is semantically enriched, accurately labeled, properly lineaged, and access-controlled before it ever reaches an agent. This is a continuous operational practice, and organizations building dedicated capability around it are pulling ahead of those treating it as a pre-launch checklist item.

According to Gartner's February 2026 press release on the AI governance market, AI governance platform spending will reach $492 million in 2026 and is projected to surpass $1 billion by 2030 — driven by AI regulation expanding to cover 75% of the world's economies. That same analysis projects that organizations deploying specialized governance tools will reduce regulatory compliance costs by 20% through automation alone.

This is much more than a compliance budget; it is a strategic infrastructure budget. Strong governance isn’t a barrier to AI. It’s the foundation for making AI work.

Context without governance is chaos

There is a conversation happening in the data management vendor community right now that the Gartner summit, perhaps inadvertently, threw into sharp relief.

A segment of the market has made a calculated pivot: away from "data governance" (compliance-flavored, slow, unglamorous) and toward "AI context layer" (strategic, forward-looking, exciting). The argument is that governance is yesterday's problem and context is all that matters for enterprise AI success.

The Gartner discourse suggests this is precisely backward.

Context without governance is not a more elegant form of data management. It is an ungoverned data estate with a better product name. The lineage is still missing. The access controls are still undefined. The metadata quality is still inconsistent. The accountability trail that would tell you why your model gave a loan officer the wrong risk score… that is still absent. You have not solved the problem. Data teams that have pursued context and eschewed governance have not solved any problems; they have rebranded them.

Gartner analyst Andrés García-Rodeja made this concrete with a forecast that should make any organization currently betting on a context-only strategy uncomfortable: 60% of agentic analytics projects that rely solely on the Model Context Protocol will fail by 2028. The reason is not a flaw in MCP itself. The reason is that MCP, absent a consistent semantic and governance foundation beneath it, has no stable ground to stand on.

The data from outside the conference wall is equally unambiguous:

• Only 7% of enterprises say their data is completely ready for AI, according to a Cloudera and Harvard Business Review Analytic Services report released in March 2026.

• Eighty-four percent of technical leaders say they need a complete data overhaul for their AI strategies to succeed, per Salesforce's 2026 State of Data and Analytics report. 

• And 78% of executives say they could not pass an independent AI governance audit within 90 days, according to Grant Thornton's 2026 AI Impact Survey.

You cannot skip the homework. You can only defer the consequences.

Agentic AI: Widespread in conversation, rare in production

One of the more honest assessments of the week came from Gartner's examination of where agentic AI actually stands in enterprise deployments. The answer: agents are everywhere in the conversation, and almost nowhere in production.

The primary barrier is not technology. It is cost unpredictability. Unlike human labor, which has established pricing models that organizations know how to budget for, the cost patterns of agentic AI are nonlinear and difficult to forecast across multi-year ROI models. Organizations that have tried to scale agents without solving this modeling problem are discovering that the economics do not close in ways they expected.

The session "AI Agents: Quantifying the Value and Cost" offered a practical corrective: frameworks for modeling total cost of ownership and key value drivers across AI agent use cases. The message was not "agents are not worth it." The message was that agents deployed without a governance and cost framework are a liability waiting to materialize.

The 60% problem and what to do about it

Gartner predicts that 60% of AI projects will be abandoned through 2026 — not because the technology failed, but because the data was not ready. Organizations with successful AI initiatives, a separate Gartner analysis found, invest up to four times more in data quality, governance, and AI-ready foundations compared to organizations with poor AI outcomes.

The key takeaway this year? AI success is not primarily an AI problem. It is a data problem. And the organizations treating it as one: building the governance layer, doing the unstructured data work, closing the context gap — are the ones that will have something to show in 18 months.

Gartner's Day 3 forecast added a significant data point to this picture: by 2027, 40% of all data management technology and services spending will shift toward multistructured data — text, documents, images, the unstructured content that underlies nearly every GenAI use case but that most organizations have not begun to govern. That investment shift is already underway among the organizations moving fastest.

One more signal worth noting

On the final day of the summit, Gartner released the results of a quarterly survey of enterprise risk leaders. Information integrity risk — the risk of AI-enabled decisions being made on untrustworthy or opaque data — has become the number-one concern among risk professionals in Q1 2026, rising above cybersecurity, regulatory risk, and geopolitical uncertainty.

That is a signal worth pausing on. The boardroom has located the problem. The question is whether the data organization gets ahead of it before the audit does.

Where this leaves us

The Gartner Data & Analytics Summit this year was not a conference about what AI can do. It was a conference about what it takes to do AI responsibly, reliably, and at scale. Those are different things, and the distance between them is where most organizations currently find themselves.

The organizations that will close that distance are the ones treating governance not as a prerequisite to skip but as a competitive foundation to build. Context and governance are not a trade-off. Context is the output of governance done well. The semantic layer, the lineage, the trust model, the access controls: these are not the drag on AI delivery. They are the thing that makes AI delivery something you can stand behind.

Alation launched its AI Governance offering at the summit on May 11, a signal of where the company believes the decisive battle for enterprise AI is being fought: not at the model layer, but at the foundation. In a week full of provocative predictions, that may be the most grounded bet of all.

Learn more about how Alation can support your path to AI governance. Book a demo today.