Every data leader I talk to is using some sort of AI analytics. Almost none of them fully trust it.
In our new CDO report, Meet the modern data team, we surveyed over 200 enterprise data, AI, and analytics leaders. The results confirm that less than 1 in 5 data leaders are confident in their AI-generated answers.
When AI accuracy falls short, most data leaders (60%) look to increase trust with cleaner data. It’s a logical instinct: cleaner data gives teams a specific metric to measure, not an abstract trust goal.
But as our research and in-depth conversations with customers show, clean data is not the right lever. Enterprises that wish to increase AI accuracy need to prioritize context instead.
Why cleaner data fails to fix the AI accuracy problem
Theoretically, clean data fixes values and connections. I use this preface because perfectly clean data is more of a fallacy than a reality. It’s a fix data leaders have strived for long before the AI era, but few have managed to achieve. Not because of their incompetence, but because it is genuinely a moving target.
But for the sake of the argument, let’s imagine you have perfectly clean data. Here’s the next problem you have to solve: AI’s misinterpretation of business meaning, which happens to be the real cause of inaccuracy. Simply put, an analytics agent that doesn’t understand what a metric means, how it’s calculated, or when it’s changed can’t be trusted to return the right answer.
Consider, for example, you ask AI to show you “revenue by region.” Any LLM can surface a generic definition to run this analysis with connected data sources. However, it won’t inherently know what the phrase means at your company:
Which tables have the right data
Whether you calculate “revenue” is net or gross
By revenue, do you mean ARR, ACV, TCV, or some special combination
By region, do you mean territory, country, or customer HQ location
If your organization changes its internal definition of the metric after a board meeting and follow-up Slack conversation, how will the LLM know? If one team calculates that metric differently than another, or if only certain teams should have privileged access to that data, how is that governed? If a user realizes an answer is incorrect, how is that feedback delivered and managed?
This is enterprise context i.e. the meaning behind your data. Without it, accuracy tanks. According to our benchmarks, AI gets those answers wrong in over 70% of interactions. That’s a reality clean data doesn’t fix.
Prioritizing context over data perfection
Everyone wants cleaner data, and it’s not a bad goal for a data team to track. But when clean data is your north star, proof of concepts drag on longer than any high-performing team can afford to wait.
As part of our study, we talked to several customers, including Tom Smith, Procurement Governance, Policy, and Reporting Lead at Arm, who reinforced this notion:
By prioritizing context-aware AI analytics, Arm’s procurement team was able to immediately cut their reporting time in half. Ironically, prioritizing context, not chasing clean data first, is what’s given Arm’s team the bandwidth to continue improving their data. “It’s freeing up the team’s time to improve data quality,” Smith explains.
This pattern shows up repeatedly. To improve AI accuracy, teams chase a laundry list of separate fixes including better AI visibility, data catalogs, and semantic layers — when they actually need a Context Development Lifecycle and an equally matched Adaptive Context Engine that enables teams to treat context as the essential enterprise IP it is.
Data teams are at a context crossroads
Smith’s approach isn’t the norm. In fact, 74% of enterprises still don’t have a shared definition of context across their organization. Of those who have at least somewhat defined “enterprise context,” about 40% say they manage it on an ad hoc, case-by-case basis.
Making matters worse, 53% say that context is undocumented, sitting in employees’ heads and scattered Slack threads instead of systems of record. This is what I hear most often from leaders who have started trying to address this problem.
Even the tools built to solve the context problem in the pre-GenAI era aren’t widely used. Just 37% of enterprises use data catalogs and only 31% use semantic layers. In the survey, leaders noted that this fragmented tooling is one of leaders’ primary challenges with their existing BI stacks.
Despite, or perhaps because of, these challenges, leaders are beginning to make practical investments in context infrastructure.
Our research found that 94% of leaders plan to change how they store and manage enterprise context within the next 12–18 months. And 75% say they have roles on their team dedicated to owning and managing AI context. Of those that don’t, 33% have open positions, and the rest plan to hire within the next two years.
Translation: the AI Context Engineer is now top of mind for enterprise data teams.
Hear more from Tom Smith on what AI transformation looks like for Arm’s Procurement team:
What comes next
AI accuracy was never about clean data. To scale AI analytics across the enterprise and deliver reliable answers that provide tangible business value, context is key.
Our full report exposes how AI is reshaping data teams, offering you a blueprint for delivering trustworthy, actionable Agentic Analytics at scale.
Looking for tools to support your Context Development Lifecycle? WisdomAI delivers over 95% accuracy in real enterprise environments. See how we can help you. Schedule a demo today.
