Here’s a confession that keeps appearing, quietly, in enterprise AI post-mortems: the model wasn’t the problem. The data was the problem. The model just made the problem embarrassingly visible.
A survey of data professionals puts this number on the table: 67% of organisations say they don’t fully trust their own data for decision-making. That’s up from 55% just two years ago, moving in the wrong direction at exactly the moment when more AI systems are being built on top of that data.
There’s a certain dark comedy to it. Billions of dollars flowing into AI infrastructure, and the thing quietly undermining it all is a customer table with three different spellings of the same company name.
The Chain Nobody Wants to Follow Backwards
The logic of AI’s dependency on data quality is straightforward — almost too straightforward for how often it gets overlooked. Better data creates better-trained models. Better models make better inferences. Better inferences enable better decisions. Better decisions drive better business outcomes.
Follow that chain backwards, and every link depends on the one before it. A decision-support AI built on incomplete, inconsistent, or ungoverned data doesn’t just underperform — it performs confidently while being wrong, which is considerably worse than not performing at all.
Gartner has predicted that through 2026, organisations will abandon 60% of their AI projects primarily due to insufficient data quality. A separate study found that companies with strong data integration achieve 10.3 times the ROI from AI initiatives compared to those with poor data connectivity. That’s not a marginal difference. That’s the difference between an AI strategy and an AI expense.
McKinsey found something particularly concrete in the manufacturing context: predictive maintenance programmes with upfront data governance investments delivered 1.8 times more ROI than those that skipped the master data foundation. The headline stat is always about AI enabling predictive maintenance. The footnote — quietly crucial — is that it only works when the underlying equipment hierarchy and spare parts data are clean enough for the AI to reason with.
The Awareness–Action Gap Is Where Competitive Advantage Lives
Here’s the uncomfortable part of the data quality story. Almost everyone knows this. And almost no one has fixed it.
The 2025 Dataversity Trends in Data Management survey found that 61% of organisations list data quality as a top challenge, and the vast majority have at least some form of data governance programme in place or planned. Yet only 4% of organisations have high maturity in both data governance and AI governance simultaneously. Awareness is widespread. Execution is rare.
This gap isn’t primarily technical. The tools exist. Databricks and Snowflake for modern data infrastructure. Collibra and Alation for governance. Informatica for data fabric. The platforms are mature, accessible, and increasingly cloud-native. What’s missing is more often organisational will, executive prioritisation, and a cultural shift from treating data as IT’s problem to treating it as a shared business asset.
The organisations that have made that shift — and built genuine data quality programmes rather than compliance-adjacent documentation — are now experiencing a compounding return. Their AI performs better. Their models generalise more reliably. Their governance teams have audit trails that satisfy regulators without months of scrambling. The investment made two years ago in unglamorous master data management is now showing up in AI performance metrics that competitors can’t match.
The Infrastructure Investment Hiding in Plain Sight
One of the more striking findings from Deloitte’s 2025 technology value survey is the budget dynamic. AI and generative AI topped the investment list by a significant margin — 74% of surveyed organisations invested in those capabilities. What’s eroding? Foundational capabilities: data management, integration, infrastructure. The very things that determine whether the AI investment pays off.
There’s a version of this that ends predictably. Organisations pour resources into AI tooling on a shaky data foundation, see inconsistent results, and conclude that AI doesn’t work as advertised. The diagnosis is wrong, but the investment follows the wrong diagnosis.
The organisations avoiding this trap are taking a deliberately unsexy approach: treating data quality as a product, not a project. Rather than one-time data cleansing initiatives that degrade almost immediately, they’re building continuous observability — automated anomaly detection, active metadata management, data SLAs with real ownership. The data stack becomes something that maintains itself rather than something that requires periodic heroics to clean up before an important AI launch.
Organisations with enterprise-wide master data management frameworks in place show roughly 287% ROI over three years, primarily through reduced rework, accelerated decisions, and the downstream AI performance improvement. That figure doesn’t travel well in executive conversations — it doesn’t land the same way as a flashy AI demo. But it holds up considerably better when the CFO asks why the AI project isn’t delivering what was promised.
Data as Strategy, Not Infrastructure
The framing shift that’s worth noticing is this: organisations that are winning competitive battles with data aren’t primarily winning on volume. They’re winning on quality, accessibility, and governance.
Raw data abundance is increasingly democratised. Cloud platforms have made storage cheap. APIs have made external data accessible. The advantage isn’t in having more data — it’s in having data that can actually be trusted, traversed, and acted upon without a team of data engineers standing between every AI query and a usable answer.
This is what “data as a strategic asset” actually looks like in practice. Not a corporate vision statement about being data-driven. A boring, rigorous, ongoing programme of data quality management, federated governance, master data alignment, and active metadata management. The strategy is visible only in the results — in AI systems that perform reliably, decisions that hold up under scrutiny, and the ability to build new AI capabilities quickly because the foundation is solid enough to build on.
The data governance market is projected to grow from roughly $5.4 billion today to over $18 billion by 2032. That’s not a speculative bet on future regulation. It’s the market catching up to a truth that data-mature organisations already know: governance isn’t the cost of having data. It’s the thing that makes data worth having.
The Founder Angle: Show the Chain
For founders building in the data and AI infrastructure space, the conversation worth having with enterprise customers isn’t about features. It’s about the chain.
Show them what their AI ROI looks like with their current data quality. Show them what it looks like with the problem fixed. The gap between those two numbers is your value proposition — and in most enterprises, it is not a small gap.
The organisations most receptive right now are the ones that have already gone through one cycle of AI investment and emerged with mixed results they can’t fully explain. They know something is wrong. They suspect it’s data. They’re ready for the conversation that helps them see it clearly.
The irony is that the highest-leverage infrastructure investment in enterprise AI right now might not be a better model. It might be a better way to manage customer IDs.
In your organisation, if an AI system started making decisions with full access to all your data right now — how confident would you be in the results?
Let’s keep learning — together.
harish.balasubramanian 
Share your thoughts