For years, organizations treated technical debt as a technology problem. We modernized applications, upgraded infrastructure, migrated to the cloud, and invested heavily in digital transformation. But as artificial intelligence moves from experimentation into day-to-day decisions, a different liability is surfacing. The central question is no longer only whether a model performs well. It is whether leaders trust the data behind it. Much of that data debt has been building quietly for years.

Why Data Trust Matters for AI

Technical-debt conversations used to focus on aging applications, unsupported technologies, fragile integrations, and infrastructure that had remained in service too long. The remedies were equally familiar, typically involving replacing, upgrading, migrating, or retiring outdated systems. Meanwhile, organizations kept expanding analytics, automation, and AI. The technology advanced faster than confidence in the information underneath it.

Most organizations now have more data and more powerful tools than ever. Yet chief data officers (CDOs) and other business leaders cannot always confidently answer whether they can trust the information well enough to act on it.

That uncertainty is becoming one of the defining business challenges of the AI era. Generating information is easy. Establishing trust is not.

I have seen how this debt forms through practical decisions made under pressure, such as building a report to meet a deadline, using a spreadsheet to close a process gap, or completing an acquisition that introduces overlapping customer records and another version of the truth. Even within the same organization, departments may define a familiar metric in slightly different ways. Each choice can make sense at the time, but the accumulated cost becomes clear later, when two reports are both technically correct yet tell different stories.

How Data Debt Accumulates

Traditional technical debt is often easier to identify through application crashes, server failures, or security alerts. Data debt, however, can go unnoticed for years, accumulating through duplicate customer records, undocumented transformations, and inconsistent metric calculations across dashboards.

In my experience leading data and architecture teams, I’ve seen organizations rely heavily on experienced analysts to compensate for gaps in data quality. These analysts know which sources are reliable, which fields require additional interpretation, and which spreadsheets need to be reconciled before information can be shared. While their expertise is valuable, this reliance highlights a larger problem. Organizations are depending on individual knowledge and manual workarounds instead of addressing the underlying data issues.

During one modernization effort, I encountered a report that appeared straightforward but relied on several systems and business definitions that had evolved separately over time. It also depended on undocumented transformations, manual checks, and knowledge spread across multiple teams. Moving the technology proved easier than determining why people trusted the numbers, who owned the definitions, and how exceptions were handled.

How AI Exposes Existing Data Quality Issues

That arrangement can work while people remain the buffer between flawed data and an important decision. AI changes the equation because it takes time out of the process. When a system can recommend or act in seconds, there may be no analyst standing between an unclear definition and a decision made at scale.

In my experience, the most challenging AI discussions rarely focus on the model itself. Instead, they arise when a team identifies a promising use case and begins questioning where the data came from, who owns it, and whether the results can be trusted.

Data Quality in Regulated Environments

Banking offers another example. Following the 2008 financial crisis, regulators introduced BCBS 239 to improve risk data aggregation and reporting. Although financial institutions already had vast amounts of data, they struggled to turn that information into accurate, timely, and reliable insights for decision-making. My experience in regulated environments has reinforced the importance of being able to explain not just the numbers themselves but also their sources, definitions, controls, and who is accountable for them.

A flawed report can cause confusion, but when automated decisions rely on inaccurate information, the consequences can be much greater. Errors can be repeated at scale and potentially damage an organization’s reputation. As organizations automate more decisions, reliable data becomes increasingly important.

With trusted data, teams can spend less time validating information and more time using it to make decisions. Leaders can act with greater confidence, AI initiatives can progress with fewer unresolved questions, and regulatory discussions can become more straightforward.

“Technology can accelerate a decision, but it cannot create trust the organization never established.” — Chris Yates

From Debating Numbers to Making Decisions

The strongest data leaders I have worked with prioritize trust in data before pursuing AI initiatives, and they recognize that governance goes beyond compliance and helps ensure information is reliable enough for leaders to make decisions without having to question the numbers.

For executives, the work comes down to four disciplines: 

1. Define critical business concepts consistently
2. Govern information at its source
3. Measure data quality with the same seriousness as system performance
4. Assign clear ownership 

None of this can be delegated entirely to the data team. Business teams must establish definitions, technology teams must provide visibility into lineage and controls, and leaders must resolve conflicts involving definitions or processes.

A practical starting point is to focus on a single important decision rather than the entire data estate. Trace the information supporting that decision, identify handoffs and workarounds, agree on the relevant business definitions, and establish ownership of exceptions. This approach makes governance more tangible by showing how it supports day-to-day operations.

Build Trust Before Scaling AI

As organizations expand their use of AI, leaders need to look beyond adoption speed and model performance. They must also consider whether their data is reliable enough to support decisions at scale.

Before AI can create a competitive advantage, data quality has to create confidence. Leaders should identify the decisions that matter most, trace the data supporting them, establish accountable ownership, and measure whether the information is accurate, timely, complete, and explainable. Progress becomes real when a leadership team can answer those questions without relying on the one person who simply knows how the process works.

The organizations that succeed in the next decade may not be the first to deploy AI or the ones with the most sophisticated models. Success will depend on their ability to use trusted information to make reliable decisions at scale.

That is the lesson I keep coming back to: technology can accelerate a decision, but it cannot create trust the organization never established. Artificial intelligence can only move at the speed of trust.