The first few minutes of my chat with Som didn’t exactly go smoothly. His webcam froze.

“I think something is wrong with my webcam,” he said. I joked back that sometimes tech saves you four clicks but adds eight more. He laughed: “Right, yeah, we are apparently on the cusp of replacing humans with AI and we can’t get phones and cameras to work. Two thumbs up.”

That exchange set the tone. This wasn’t going to be a conversation about hype. It was going to be about what really matters when it comes to technology and growth.

From Tools to Fundamentals

I asked Som where he’s been spending his time lately, and he didn’t start with the latest product release. He went straight to the basics.

“At the end of the day, my passion is essentially deriving value from data,” he said. “And it doesn’t matter what data—it could be structured, unstructured, documents, industrial measurements, or sales and ERP. My passion is that, and whatever tools are necessary to do that. I’ve gathered a whole bag of them.”

He explained that while most people measure expertise by how many tools they’ve used, he focuses on the fundamentals. Tableau and Power BI? Those are just implementations. “Which graph to put together—I have studied the theory behind that. Where 3D is necessary, where it’s not, where it detracts from the answer. When to use two axes versus three versus four.”

It was a good reminder: the power isn’t in the tool itself, but in knowing how to apply the right method.

Not Everything Is AI

Som shared a story from consulting that stuck with me. A company came to him insisting they needed AI for process monitoring.

“They said, ‘Hey, we need AI to do this.’ So I said, ‘Well, you don’t need AI. You need statistical process control. Here’s what I call VECO rules, and you apply those rules and you have 90% of your answer.’”

I told him I’ve run into the same thing. People are almost disappointed when I say it’s just math. They want the AI answer because it sounds more impressive. But as Som reminded me, five minus two is still three—you don’t need a machine to tell you that.

Data + Intuition

What impressed me most was how much Som values intuition in the process.

“For a data person, it’s very important that they work with a domain expert,” he said. “Sometimes that’s lost on people. What is the application domain? Is this bioinformatics? Is this a business decision? An industrial automation decision?”

He wasn’t dismissing data—far from it. His point was that the best results happen when you combine the math with the instincts of people who know the space. “I do believe strongly in human intuition,” he said. “Your brain is the best pattern recognizer I know of yet.”

That hit home for me. Technology can validate or challenge intuition, but it shouldn’t ignore it.

When Models Fall Short

Som also talked about the limits of AI. He gave the example of asking a model about a niche hobby. The results sounded confident but were completely wrong.

“If you ask about a niche topic, the newer information is sparse, and there’s a lot of old information available. So it weights it that way. It gives you the wrong answer with high confidence,” he explained.

He compared it to interrogating a child prodigy. “It’s like a four-year-old who is a genius with a wild imagination. They know a lot of stuff, but they’ll make up stories. Imagine that they knew the entire internet—but they had that same wild imagination.”

That description is going to stick with me. It’s a reminder to use these tools, but to keep questioning them.

Where the Real Opportunities Are

When I asked Som where the biggest opportunities are right now, he didn’t throw out buzzwords.

“Of course there’s a lot of noise around large language models,” he said. “But think a step back. Where do enterprises have their data? They still have it in databases. Payment data, claims data, EHR records—they’re structured.”

His point was that the real wins come from combining structured and unstructured data. “Can you bring structured data together with language models that look at contracts or documents? That’s where you can build something better.”

Back to Basics

Our conversation started with a broken webcam and ended with a discussion about vectorization and feedback loops. In between, one theme kept surfacing: fundamentals matter.

AI isn’t magic. Tools come and go. The companies that thrive are the ones that understand data basics, respect intuition, and apply technology pragmatically.

As Som said, quoting a familiar line: “All models are wrong. Some are useful.”