The AI boom is forcing firms to confront their data chaos

Most companies believe the biggest challenge in artificial intelligence is building smarter models. Now, many of the executives actually deploying AI across global enterprises are discovering something less glamorous but far more consequential: the real battle lies in cleaning up decades of messy, fragmented and inconsistent corporate data.

Inside large organisations, information is often scattered across ageing ERP systems, customised databases, spreadsheets, emails and disconnected software environments accumulated over decades. One division may label a product one way, another may describe the same item differently, while yet another stores it in an entirely separate system. AI systems, executives say, inherit every one of those problems.

“We are not struggling with the amount of data,” said Bejoy John, senior director, enterprise data management at Wesco. “We are sitting on volumes and volumes of it. But the data lacks quality, ownership, observability and accountability. That’s why many AI systems become brittle and people stop trusting them.”

As companies rush to adopt generative AI and autonomous AI agents, executives increasingly argue that data quality — rather than algorithms themselves — is becoming the defining factor separating successful AI deployments from expensive failures.

“Everyone has data,” said Kamlesh Solanki, director of data engineering at HP. “Supply chain, finance, customer data — every organisation has it. But the real differentiator is not who has the most data. It is who has the most trusted, governed and contextual data.”

Solanki believes many companies still misunderstand the famous phrase “data is the new oil”.

“Oil is just a commodity,” he said. “Data is context. And context is what separates organisations experimenting with AI from organisations actually winning with AI.”

That distinction matters because many companies remain trapped in endless AI pilots that never scale into meaningful business systems. According to executives, the problem is rarely the sophistication of the AI model itself. More often, it is the inability to trust the data flowing into those systems.

At HP, Solanki said AI-powered predictive support systems now monitor millions of PCs and printers globally, detecting signs of hardware failures before customers even notice a problem.

“We have an AI that knows what is going to break before it breaks,” he said.

The system analyses telemetry data continuously, monitoring battery degradation, storage health and thermal performance. If problems are detected, automated tickets are generated and replacement parts can be dispatched proactively.

“But if the telemetry data is noisy or inconsistent, the AI starts recommending the wrong parts and you are back to square one,” Solanki said. “Then you have wasted logistics, multiple repair visits and frustrated customers.”

The challenge becomes even more complicated in sprawling industrial organisations.

Naveen Kamat, chief digital & AI officer at Larsen & Toubro, said the company deals with enormous volumes of multimodal data — from ERP systems and procurement platforms to drone feeds, digital twins and complex engineering drawings.

For the company, AI is increasingly being embedded into construction sites, factories and project execution systems.

“We are deploying drones, robotics and physical AI systems at sites,” Kamat said. “If a worker is in a hazardous environment or not wearing the right safety harness, we need to detect that in near real time and notify safety managers immediately.”

That requires reliable, real-time data pipelines. “If the data is not available within the window where it can make an impact, it may no longer be relevant,” he said.

Padmashree Shagrithaya, executive VP and head of insights & data for India at Capgemini, said the rise of autonomous AI agents is making data governance vastly more important than it was during earlier analytics eras.

“What’s happening today is AI is becoming more autonomous in its behaviour,” she said. “If the underlying data is wrong, it’s no longer just garbage in, garbage out. It becomes a business risk.”

She pointed to the example of AIpowered HR policy assistants. In many companies, policies evolve over decades, creating multiple versions and overlapping documentation. Humans instinctively understand timelines and context. AI systems do not.

“If the data is not versioned properly, the AI agent may pull the wrong policy or wrong context,” she said.

The issue becomes more serious when AI systems begin taking actions independently rather than merely making recommendations. Autonomous systems increasingly handle pricing decisions, inventory forecasting, maintenance planning and operational workflows in real time.

That is forcing organisations to rethink data governance as a continuous, real-time discipline rather than a periodic compliance exercise.

John described data lineage — the ability to trace how data moved through systems — as following “breadcrumbs”.
“You should be able to explain exactly how a number travelled from a source system all the way into a financial report or AI model,” he said.