AI at Scale: Why Trust, Not the Model, Decides Who Wins

Tactical insights from Grant Thornton, Ligentia, and Swire Coca-Cola

Every AI conversation eventually turns into a model conversation: which one is fastest, which one hallucinates least, which one your competitors are already using. 

Three recent episodes of The Data and AI Chief, hosted by ThoughtSpot's Chief Data and AI Strategy Officer Cindi Howson, make a different case entirely about what it takes to scale AI in the enterprise. 

Across conversations with leaders at a global supply chain provider, a top global advisory firm, and one of North America's largest Coca-Cola bottlers, the model was rarely the subject. 

The subject was trust: how you earn it, how you prove it, and how close you're willing to let AI get to a real decision once you have it.

Howson's guests this season, Alpa Lally, Chief Product Officer at Grant Thornton; Boris Rabkin, CIO at Ligentia; and Bharathi Rajan, VP of Data, Digital, and AI at Swire Coca-Cola, run AI programs in three industries with almost nothing in common on the surface. But the questions Howson asked each of them kept circling the same territory. What happens when you stop treating AI as a reporting tool and start treating it as a partner in making decisions? And what has to be true of your data before you're allowed to make that leap?

Reimagine the process instead of automating the mess

Howson has a habit of pushing past the productivity story in these conversations, and her exchange with Alpa Lally at Grant Thornton was a clear example. Asked how the firm approached its AI transformation, Lally didn't start with tools. She started with a warning about the most common mistake she sees other organizations make.

At Grant Thornton, avoiding that trap meant asking a harder question before building anything: if we were designing this process today, with AI available, would we build it this way? That question led the firm to rethink audit workflows, including compliance and risk assessments that once took days or weeks. The goal was to make relevant insights available before an auditor even walked into a client meeting. Lally called it leapfrogging, not upgrading.

Boris Rabkin described almost the same fork in the road at Ligentia, just from the infrastructure side. When he joined the company, he found what he called "a very well-built reporting factory," a lot of effort, a lot of reports, and still no real speed, because the reports were answering yesterday's questions instead of today's. His solution wasn't a better dashboard. It was refusing to treat internal tooling and the customer-facing product as two separate data problems.

That decision produced Ligentia Insights, an analytics experience embedded directly inside the company's own operating platform rather than shipped as a separate dashboard. The goal, in Rabkin's words, was making sure "people are where decisions happen," not looking back at a report after the moment to act on it has already passed.

Governance as the price of entry, not a tax on speed

If there's a single myth this season's guests dismantled in unison, it's the idea that governance slows AI down. Bharathi Rajan built her entire program at Swire Coca Cola on the opposite premise.

Rajan didn't inherit that discipline. She built it, largely through education and relationship building rather than mandate, and she was candid about the early resistance she encountered before teams saw governance actually accelerate their own work rather than gate it.

Lally's own research at Grant Thornton put numbers behind that same instinct. The firm's AI impact survey found that 78% of companies lack confidence they could pass an AI governance audit, what Lally calls the "AI proof gap." But closing that gap doesn't slow organizations down. It's what lets them move faster: companies that fully integrate AI are nearly four times more likely to report revenue growth than those still piloting.

"The goal can't just be deploy AI," Lally said. "The goal is to deploy AI responsibly so that people can have confidence in the outcomes that it produces."

Rabkin runs the same calculation globally, operating across the EU AI Act and a comparatively lighter US regulatory posture, and refuses to build two separate standards."We design often to a higher standard from the start, making sure that we build explainable, governed, and scalable solutions across different markets," he said. "It's a little bit more effort upfront... but it does avoid the rework that comes in later. And more importantly, it builds trust."

Put AI where the decision actually happens

Howson pressed Rabkin directly on a question dividing a lot of organizations right now: who should own AI? His answer was immediate. "AI should be owned where value is created," Rabkin said. "It shouldn't be a centralized function inside a lab. If it's not close to the product and the people that are using it, AI won't create the value."

Rajan's version of that same idea was less about org charts and more about where human attention goes once routine decisions get automated. She's careful to frame AI as a redistribution of effort rather than a replacement for it.

"It's an enablement factor for us internally," Rajan said, "helping the enterprise really grow, create operational efficiencies, but also have people do more impactful work. We all want to do something that's more impactful."

Lally described a similar shift in her own role. Three years ago, she said, much of her time went into gathering information and building presentations for others to make decisions. Now, her focus has shifted toward interpreting that information and helping drive those decisions.

"The work is actually less about producing information and more interpreting it," Lally said, "connecting the dots, driving the decisions. And that's ultimately where the greatest value has always been."

The pattern across three industries

A clear pattern had emerged across all three conversations: Proximity to the decision, backed by governance strong enough to earn trust, is what actually determines whether an AI investment produces value.

Rabkin summed up the stakes plainly, describing what happens when a shipment gets stuck, and nobody trusts the number in front of them fast enough to act on it.

"Speed is one thing," he said. "But confidence in the data is something that really drives decisions."

That's the thread running through every conversation this season. Rajan built trust through education, one team at a time, long before anyone asked her to prove ROI. Rabkin built it by holding customer-facing and internal tools to the identical standard, so there was never a second-class version of the data. Lally built it by showing that governance and growth aren't opposing forces. They move together.

None of that shows up on a model leaderboard. All of it shows up in whether the person closest to a decision can act on it with confidence, in the moment that matters. 

Catch the full conversations with Alpa Lally, Boris Rabkin and Bharathi Rajan on The Data and AI Chief, hosted by Cindi Howson, wherever you get your podcasts. Catch up on past episodes of the show at The Data and AI Chief podcast.