“AI cannot fix what incomplete data creates”: Deepu Chacko on India’s agentic AI challenge

As Indian enterprises accelerate their artificial intelligence (AI) deployments, the biggest bottleneck isn’t only compute power or model sophistication, it’s also data.

According to Salesforce India’s Vice-President, Solution Engineering, Deepu Chacko, disconnected and untrustworthy data is undermining AI ambitions across organisations.

“AI cannot fix what incomplete data creates. For India to truly unlock the promise of agentic AI, leaders must treat data as a strategic asset, unified, governed, and contextual,” Chacko told Business Today. “The companies that modernise their data foundations today will be the ones that scale AI responsibly and lead the economy tomorrow.”

Disconnected systems, disconnected intelligence

A decade ago, Indian enterprises were focused on digitisation. Today, they generate vast amounts of digital exhaust, from customer engagement data to backend transaction records. But much of it remains siloed.

“Your orders would be in ERP (enterprise resource planning). Your sales orders would be in Salesforce. Your other engagement data is sitting on a different application,” Chacko said.

This fragmentation weakens AI performance.

“AI outcomes usually weaken when customer data is spread across disconnected systems,” he explained. A customer’s website behaviour may be tagged to an anonymous visitor ID, while the same customer’s verified B2B profile, complete with identifiers such as a PAN number, sits separately in an ERP database.

“How do you actually tie it all together? Because only if you tie it together, then the AI becomes a lot more efficient,” he said.

When 25% of data is ‘untrustworthy’

The challenge isn’t merely fragmentation, it’s trust.

A recent Salesforce report found that 89% of data and analytics leaders agree that AI outputs are only as good as their data inputs. Yet those same leaders estimate that over a quarter, 25%, of their organisational data is untrustworthy.

Chacko distilled the problem bluntly: “Bad data is equal to bad AI.”

India’s rapid digital acceleration has enriched customer data over time. “Ten or fifteen years ago, it was just a mobile number and name. Today, customers share email IDs, addresses and detailed preferences,” he noted.

The implication is clear, richer data demands stronger governance. Enterprises that fail to modernise their data architecture risk scaling AI on unstable foundations.

Reducing hallucinations: Prompt vs context

As concerns about AI hallucinations grow, Chacko argues that enterprises often focus on the wrong lever.

“Prompt is where you tell GPT (Generative Pre-trained Transformer), ‘play the role of a financial advisor.’ That’s prompt engineering,” he said.

But prompt engineering alone is insufficient.

“Context engineering is about feeding enterprise data that already exists in backend systems.”

An AI assistant that understands a customer’s past loan responses, spending behaviour or engagement patterns can personalise interactions far more accurately.

“That is called context engineering. Now, how do you actually feed all of this? You have to get access to that data, put it into the context, and give it to him,” Chacko said.

Beyond data access, he stressed the importance of observability and auditability.

“You should be able to go into each conversation and say, why did the AI respond like that? What context did it have?”

The rise of the ‘agentic enterprise’

Looking ahead, Chacko believes the next frontier is agentic AI, systems that act autonomously and interact with other AI agents.

“In about 10 to 15 years, you will actually have an AI agent for you,” he said, predicting a world where personal AI agents transact directly with brand agents, booking travel, negotiating purchases or managing services.

An “agentic enterprise,” he explained, is one that retools its systems, workflows and workforce for this AI-to-AI interaction model.

“Agentic AI isn’t the next technology, it’s the next revolution. AI agents handle routine tasks so humans can focus on creativity, relationships and impact.”

For India Inc., that means preparing not just infrastructure, but culture and capability.

Reskilling without an age divide

Contrary to the perception that AI disproportionately benefits younger, digital-native workers, Chacko sees no age-based displacement.

“We are not seeing this age shift,” he said.

Instead, automation is freeing senior professionals from repetitive tasks. “They are now able to focus on relationships and brand value rather than doing all the mundane tasks.”

In fact, he argued, senior leaders may gain an edge, provided they have access to unified data. Strategic conversations with customers require context, and context depends on integrated systems.