AI-Ready Data

Why data quality for AI is now a board-level issue

Key Takeaways

  • AI agents remove the human review step that once absorbed data quality problems, so errors people used to catch become automated decisions at scale.
  • Organizations that act on data quality anchor to one owned use case with a deadline, rather than a vague company-wide initiative.
  • Certified data products with visible quality and governance scores give business users and agents a way to judge whether data is safe to use, and they create a measurable ROI story.

Agentic AI adoption sits at 41% in the 2026 State of Data Integrity and AI Readiness report, published by Precisely in partnership with Drexel University’s LeBow College of Business. This survey of 505 data and analytics leaders revealed that only 55% of respondents are executing on initiatives to improve the quality of data used for AI, and 16% haven’t started.

That gap is data quality debt, and it is what’s made data quality for AI, and agentic AI in particular, a board-level problem.

I started my career as a practitioner building data warehouses, so I’ve watched this debt build for decades. I recently moderated a Precisely forum on this topic to dig in deeper, and was joined by Cam Ogden, SVP of Product Management at Precisely, and Franck Carassus, CEO of Huwise, a Precisely partner that provides a secure data product marketplace. Here are the highlights.

Why does data quality debt matter more now that agents are acting?

Data quality debt goes back to the earliest data warehouses. For years, a human in the loop checked whether data was complete, accurate, valid, and timely, so organizations could defer the fix.

Agents change that. They act on data directly, and nobody reviews it first. The report puts numbers on the pressure:

  • 51% of leaders name data quality as their most common priority
  • 43% cite data readiness as the top barrier to aligning AI with business goals.

Cam described how the problem hides. “On average, there’s one data quality issue for every 10 tables,” he said. Once an agent replaces the analyst or engineer who would have caught those hairline fractures, nobody notices them, because the agent has no strong signal about which data is reliable and which isn’t.

Franck was blunt about what that means. “With AI, I would say the safety net is gone,” he said. When an agent acts on data with no review step, “dynamic verification of data fitness stops being a ‘nice to have’ and becomes essential to avoid a real operational risk.”

The difference between teams that fix data quality for AI and those stuck in planning

The report found 55% of organizations executing on data quality for AI, 30% planning or evaluating solutions, and 16% not yet started or still in discovery. Our live audience polls skewed toward planning, a little behind the survey.

I asked Franck what separates the two groups, and his answer had nothing to do with money. “It’s not budget. It’s not awareness. Everyone’s aware by now,” he said. The difference is whether one specific, owned use case is forcing the issue. Companies that execute anchor to that use case, give it a named owner, and set a deadline to get it unstuck. Teams that stall treat data quality as a company-wide initiative where everyone agrees it matters and nobody is accountable for results.

Franck also described a second pattern. Companies that execute are moving toward data democratization, with self-service data for business users organized around data products rather than datasets. Once you commit to that model, data quality stops being optional.

What does a trust layer for AI agents look like in practice?

Cam laid out the picture most organizations share: an agent layer sits on top, and a data landscape of legacy systems, lakehouses, and warehouses sit underneath.

However, the layer in between is missing. Agents return answers, but leaders can’t tell where the information came from, whether it’s governed or a raw table on the side, or what quality level it carries. If an auditor walked in, could you show the decision an agent made and the signals behind it?

Cam’s demo showed how the Precisely and Huwise partnership fills that gap, pairing Precisely data governance and data quality capabilities with the Huwise data product marketplace so that shared data products stay trusted and fit for use. A data steward defines a semantic definition for a data product, runs it through a certification workflow, flags personally identifiable information, and sets scoring thresholds for completeness, validity, and governance.

Once a data product reaches a governance score of 100% and is certified, it appears in the marketplace. When Cam asked the Huwise agent to find data products related to customer analytics, the result carried the quality and governance scores from Precisely, and agents can reach the same metadata through MCP.

The stakes are rising inside Precisely, too. Since February, we’ve gone from about 60 or 70 Tableau developers building reports to 2,000 people accessing the same data through Claude and Copilot. At that scale, trust in the data underneath matters for every decision, including the ones we ask agents to make in sales.

How do you show ROI on data quality?

Proving ROI around data quality and data products can be elusive for most organizations. The Lebow report found 32% of organizations expect positive ROI from AI within 6 to 11 months, yet only 31% have well-established metrics tied to business KPIs. This is because many focus on technical performance instead of how they impact business outcomes, which makes it hard to know whether the investment keeps paying off.

Franck shared results from a Forrester joint study with Huwise, based on customer interviews. Business users cut the time they spend searching for data by half, and data teams cut prep time by half.

Teams also reduced duplicate data product creation by 75%, which means fewer rebuilds and more reuse of trusted products. Forrester estimated a 474% ROI from deploying a data marketplace, with payback in under six months. One large US customer retired more than 300 dashboards and expects roughly $10 million in cost reduction.

The next steps in your trusted data journey

When asked how organizations should plan next steps, both panelists said to start small. Franck’s advice is to build a foundation of 50 to 100 certified data products that anyone in the company can trust. The winners, he said, are not the companies with the most data. They are the ones who make their best data easy to find, trust, and reuse.

Cam offered three questions:

  • Can you name the datasets your agents are touching right now?
  • If an agent gave you a wrong answer today, could you trace it to that dataset in under an hour?
  • Does every team that touches the data agree on what trustworthy means for it?

The answer to the third question is almost never yes, and that’s often the real gap.

Cam’s warning for anyone tempted to launch another program: “The instinct will be that we need more pilots, more policies. And I can tell you that that’s usually backwards.” The companies that broke through picked one use case, gave it one owner, answered those three questions for that use case alone, and expanded from there.

The data quality debt is due

Data quality debt built up over decades because people covered for it. Agents don’t provide that coverage. That’s why data quality for AI initiatives can’t wait.

The organizations pulling ahead pick one use case, name an owner, and make the data behind it AI-ready. Unified data management gives that work a shared foundation, so definitions, quality rules, and policies stay consistent as more agents come online.

Download the 2026 State of Data Integrity and AI Readiness report to see the full findings and benchmark your own progress. And for more insights from our discussion, watch the full panel on demand.

 

Steve Varsolona

Vice President, Sales

Read More from the Precisely Blog

View All Blog Posts

For AI, Context Isn’t Optional: What Data and Analytics Leaders are Saying
AI-Ready Data

Enterprise AI Needs a Deterministic Layer

AI-Ready Data

Three questions that will tell you if your data is ready for AI

AI-Ready Data

AI-ready data is the foundation for successful enterprise transformation

Let’s talk

Integrate, improve, govern, and contextualize your data with one powerful solution.

Get in touch