Enterprises should cut chaos off their data to make it predictive, insightful Business News & Hub

Artificial Intelligence (AI) has become inevitable, however, the technology’s success entirely depends on whether enterprises can turn their messy, fragmented, unstructured data that lies in diverse silos into something reliable, clean and trustworthy. Yes, no doubt data is gold, but there’s also glass inside, so it is critical to isolate the glass before the gold is put to use.

AI dazzles with its promise of predictive insights, personalised experiences and automated intelligent decision-making. But these magical offerings fizzle into chaos if the quality of the data it feeds on suffers.

These were the key takeaways from an industry roundtable on Data Without Chaos: Building the Foundations for AI@Work, organised by The Hindu Group and powered by IBM, the New York-based tech giant. The roundtable, first in the series, recently held in Bengaluru saw the participation of a 10-member panel comprising CXOs, technologists, data scientists, AI strategists, advisers, consultants and educators from a cross section of industries ranging from petroleum infrastructure, e-commerce, insurance to health-tech, education and the media.

AI from research labs to mainstream markets

Well, until a few years ago, businesses were still debating whether AI was necessary at all, but in the last couple of years, that question has been quickly hijacked by a sense of urgency backed by a flurry of activities in a rush to embrace AI as it offered predictable outcomes and unmatchable customer experiences.

Moderator Mr. Nagaraj Nagabushanam, Vice-President, Data and Analytics and also Designated AI Officer, The Hindu set the tone for a true-blue B2B AI conversation by saying: “Nobody is asking why AI, anymore. The question now is what outcomes will it deliver.”

Yes, nobody asked why, but only plunged deep into discussions as for everyone it was crystal clear that AI was increasingly becoming mainstream, was getting into frontline operations, becoming part of corporate policies, supply chain decisions, customer relations, and certainly AI now is not something that is  ‘just out of a research lab.’

But the challenge is: for AI to deliver desired outcomes, the data must be clean, AI-ready and beyond.

According to one of the panelists, Harish Ramarao, Senior VP-Product Engineering at ACKO Technologies, the biggest issue is timeliness, correctness or freshness of the data.

“For example, a customer, when he or she bought an insurance policy, would have told he/she was not married. Two years down the line, he or she gets married and that information comes from somewhere, maybe via a chatbot interface to the company. So, the consumer’s latest data is different the original data.’‘

The customer would expect the system to know this latest change when he or she called the company. So, correctness of the data or freshness of the data was key to customer satisfaction.  For AI to offer valuable outcomes, consistency is critical. Duplicate identities across systems,  it could even be different PIN codes, can undermine trust, he said.

At this point, Lakshminarayan Swaminathan, Head of Product Management and Design at Myntra highlighted the unevenness of systems across the enterprises. “Each system has very different levels of maturity, even as we try to prepare for readiness. I think it will vary because there is a kind of prioritisation that we need to do. Each of our data systems is at a different level of maturity. We can’t make everything AI-ready at once, so we have to prioritise where the impact will be highest.’‘

According to him, there is a lot of optimism when looking at a surface from a height of 30,000 feet but as soon as one starts climbing down to the ground, things start getting very messy.

“What will it actually mean? There is some level of anxiety about where this is eventually going to lead to. No one is able to visualise exactly what it is. In some sense, the way we are at least trying to see it is that usable data is not very contextual. I don’t think each of our data systems are at the same level of maturity.’‘

Challenge of AI powering legacy organisations  

Joining the conversation was Suresh Vijayraghavan, CTO of The Hindu Group, with his narration of the early years of the 150-year-old publishing group’s digital transformation and content digitisation journey ushered in by computerisation in 1992.

According to him, it was in 2003 The Hindu Group started digitisation, although pages were already being done on computers way back in 1992 itself. Still, for some years, data remained manual and today the data of the organisation has evolved over multiple systems.

“Data is gold, but there’s also glass inside. You must isolate the glass before you can use the gold,” cautioned Mr. Vijayraghavan.

Also, he said, the content that the group had in 1878 was the content that was generated yesterday, but the whole structure was different.  “The machines that generated data 25 years back and the machines that generate data today, they have different data sets altogether, different capabilities.  So, when they want to use this huge amount of data they have to ensure that this is correct.’‘

For the media, like in the case of manufacturing or insurance, the challenge was integrating data produced by very different systems across decades and through multiple generations as in the case of The Hindu Group.

Is there a visible tension between speed and rigour?

For fast-moving digital sectors, speed is the essence. Experiments are launched quickly, sometimes with little structure, but in regulated and asset-heavy industries, sloppy experimentation can damage trust or invite regulators, commented moderator, Mr. Nagaraj leading the panel to another round on the tension between speed and rigour of deploying AI.

Dipesh Shah, Executive President & CTO, Havells India, argued for discipline and said, process won’t really slow you down — it actually speeds you up in the long run. That’s why I’ll always side with rigour over rushing.

“If I have to take a side, I would take the side on rigour because of a lot of downstream implications. You don’t want to go fast. Avoid temptation and greed. You need to build the framework around governance: confidentiality, privacy, the Digital Personal Data Protection Act, etc.

Go down and take a rigorous path process, actually the process doesn’t slow you down. Eventually, it speeds you up, for the time being, it slows you down, which is fine, Mr. Shah said, cautioning, “We need a culture of no greed, no temptation. Just because the data is there doesn’t mean we should use it without guardrails.’‘

The panellists in general agreed that a balanced approach was needed and there was also a need to move quickly where stakes were low, yet enforcing strong frameworks where stakes were high.

Dismantling data silos

In the breaking silos and building abstractions round, much of the conversations were circled around data silos. They (silos) weren’t just the product of organisational turf wars, several argued, but said they were often historical accidents.

“Silos aren’t just turf wars. They’re historical accidents — old mainframes, mid-generation databases, and today’s clickstreams. The key is to abstract the data model,’‘ as Mr. Ramarao of ACKO puts it crisply, appending that abstraction, understanding what data means, regardless of where it sits, was what allowed enterprises to unify decades of fragmented information.

No single cloud world anymore

As the panel moved on to talking about hybrid cloud and cost factor, the moderator stated that cloud was no silver bullet, observing, few companies were purely cloud-native, but most live with some hybrid or multi-cloud configuration.

Sandhya Kapoor, Senior VP, Head, Central Platform Organisation at Flipkart, was of the opinion that: “Cloud choices are often forced by regulations, by availability of certain models, or simply by cost. No one is living in a single-cloud world anymore,’‘ she observed.

Again, cost was the sharpest constraint for adoption, for many. “Moving a terabyte of data from one cloud to another can blow up your budget before you even run a query,’ added another panelist.

A pragmatic approach should be maintained, to keep some workloads on-premise, burst to the cloud when scale is needed, and weigh every move against cost and compliance, the participants concluded.

Ownership breeds fear of blame; championship creates pride

Moving on to discussing culture and data strategy, the moderator said, technology solved only half the problem, while culture solved the rest.

Sudhir Kumar, MD, Petronet MHB Ltd rephrased it: “Call someone a data owner and they get cautious. Call them a data champion, and they take pride in solving the problem.”

It’s not just semantics. Ownership language breeds fear of blame; championship language creates pride and accountability. “But then as a custodian, there   are responsibilities. Sometimes the data stops flowing. They become stale. Some calls go for a toss, and the downstream pipeline, downstream models, downstream reports, everything goes for a toss. And at that time, all the blame games started,’‘ Mr. Kumar reacted candidly.

From curiosity to ROI

Almost every enterprise has been investing in AI preparedness and in building required skill sets in the last some years. The challenge now is around showing results, remarked  moderator, Mr. Nagaraj.

Mr. Shah chipped in saying: there has been a lot of  focus, energy and attention currently on getting some ROI out as already a lot of money has been sunk into AI investments, or so-called AI investment people thinking that they’re investing in air but they really have no clue. “But they want ROI. So, give me a company in the last one year that has not done at least upskilling their employees on some AI tool. Now, it has to pay off, which means people have to experiment,’‘ he commented.

Dr. Pavankumar Gurazada, Associate Director-AI/Data Science at Great Learning, described the gap he has been witnessing in markets. “No enterprise has come to us with squeaky-clean data and said, just train our people on generative AI and they’ll be off to the races. That has never happened. So, in one sense, to have that expectation itself is sort of incorrect.’‘

Vinoth Vijayan, Head of Hybrid Cloud & Data Services, IBM Consulting, India and South Asia shared his perspective saying:  “Aligning your platform capabilities with business alignment is critical — otherwise you’ll spend heavily and never reach production.”

Siddhesh Naik, Country Leader, Data and AI, India and South Asia, IBM, , IBM’s Data & AI leader in India, emphasised on the pressure from the boardrooms stating, “Boards are asking for outcomes, not pilots. Every company has upskilled for AI, now the pressure is to deliver ROI.”

Responding to a moderator’s query on what are the strategies that companies are deploying to unify their siloed data sources, Mr. Naik said, “It starts in a modular approach. For example, it starts from a data-pipeline observability. I need to figure out what’s failing and how do I get to the right level and go fix it.’ Data pipeline observability is the ability to deeply monitor and understand the real-time health, performance, and behaviour of data pipelines. The second approach according to him was to create an AI data foundation, a lake house and also build the piece of end-to-end data governance.

Trust, Security and Resilience

As AI moves into critical processes, the risks continue to rise. Regulators demand explainability. Customers demand privacy. CIOs and CISOs are under pressure to secure the foundations, the panel conceded.

According to Mr. Naik, a fundamental truth that always remained was: “The customer is ultimately the owner of the data. Consent can be withdrawn — if you can’t pull it out on demand, your system is broken.”

The message that was loud and clear was that trust and resilience are as important as accuracy or efficiency.

Commenting on building on centralised foundations, Pranjal Singh, Staff Data Scientist at B2B e-commerce platform, Udaan said. “We moved from the ML era to the GA era by first centralising data. Without that foundation, nothing scales.” A genetic algorithm (GA) is an optimisation and search technique that mimics natural selection to find optimal or near-optimal solutions to complex problems. The journey so far has been about building foundations and then trying to build innovation around it.

A simple message was the takeaway for all: no amount of model sophistication can compensate for weak data plumbing.

What’s the ground reality around AI readiness?

As the interactive session neared a closure, the discussion quickly circled back to the key theme: Data Without Chaos. Every participant recognised that AI was inevitable and also agreed that the challenges were only real. The message for business leaders was both simple and profound. They cannot leapfrog into AI maturity by buying a model or signing a cloud contract. Instead, they must begin by tending to their data pile: cleaning it, governing it, abstracting it, managing it and securing it.  They must shape an AI culture and data strategy that values rigour over hype, clarity over greed and temptation.

“AI adoption was not about chasing magic. It’s about building trust — trust in your data, trust in your systems, trust in the outcomes,’‘ sums up moderator Mr. Nagaraj.

Ms. Kapoor reiterated, “The businesses are keen on AI adoption, as they especially hear about the magic of generative AI. They are keen to see how it can benefit. So, there are proposals and ways of doing things discussed.’‘

However, she added, there are healthy conversations around AI adoptions, and sometimes conflicts too, when technical stakeholders and product teams put their heads together and start looking under the hood as part of preparedness for deployments.