By Victor Aston

There’s been a lot of noise lately about AI replacing data analysts. I get it, the pace of AI development is staggering. Every other day there’s a new model that promises to analyze faster, predict better, or generate cleaner visualizations than anything before it. But let’s pump the brakes for a second.

Can AI write SQL queries? Sure. Can it clean data? Absolutely. Can it build dashboards that look half-decent? You bet. But ask it to figure out what question to ask in the first place, and that’s where things get shaky.

Here’s the thing: data analysis isn’t just about crunching numbers and throwing charts on PowerPoint slides or visualisation tools. It’s about translating messy, often vague business needs into clear, focused, and measurable questions, then digging into the data to find answers that actually make sense in the real world.

And that real world is chaotic.

After years of working with various business stakeholders, marketers, engineers, and customer service teams, one thing is clear: most stakeholders aren’t always sure what they want. They come to meetings with gut feelings, not hypotheses. It’s up to the data analyst, the human one, to listen closely, read between the lines, and guide the conversation toward clarity. No AI, no matter how smart, can replicate the experience of sitting in a product meeting, knowing the history of a feature roll-out, the business goals behind it, and sensing when a stakeholder is asking the wrong question.

Could AI eventually connect to a company’s entire data stack, ingest its KPIs, understand context, and chat with users like a data-savvy colleague? Maybe. It’s not impossible. But today? It’s mostly reactive. It works with what it’s given. And when it’s given bad input, it churns out confident nonsense.

We’ve all been there: asking an AI to generate a chart or summarize a trend, only to spend 45 minutes trying to phrase the request just right. It’s like explaining your Spotify Wrapped to a robot that’s never listened to music. Even voice-based AI isn’t immune, if you slip up and say “churn” when you meant “conversion,” guess what? The machine’s halfway through building the wrong dashboard before you realize.

That missing bit, the intuition, the business context, the emotional intelligence, that’s the analyst’s secret sauce. It’s what lets us know when something feels off, even when the data looks fine. It’s what lets us turn numbers into stories that people trust. And that’s a feeling no AI can replicate. Have you ever asked an AI to generate an SQL query based on your database tables and columns? Sure, it runs fine, syntax is perfect, no errors. But when you look at the result, you just know it’s wrong. Not technically, but contextually. It doesn’t actually answer the business question you’re trying to solve. And so, you go back and forth, teaching the AI, rephrasing your intent, and slowly guiding it toward what you actually meant in the first place. That back-and-forth is where the human analyst shines not just asking questions, but refining them, explaining why an answer is wrong even when the code is correct. In that moment, you’re not just using AI, you’re managing it.

And here’s another thing: AI, at least right now, isn’t designed to disagree with you. You could be confidently wrong, misreading a metric or asking a misguided question, and AI will politely nod along. It doesn’t push back. It doesn’t challenge assumptions. It agrees. Which might feel nice, but it’s not how you build trust or transparency in your data process. Ask yourself this: would you confidently base a key business decision on an AI-generated insight, based solely on one prompt? Without second-guessing it, or validating it with someone who knows the business inside and out? Probably not. And that’s the point. Data analysts don’t just answer questions, we question the answers, too.

Now, I’m not saying AI has no place in analytics. On the contrary, I love it. It’s brilliant for automating the heavy lifting, query generation, summarizing trends, even building out first-draft visualizations. In fact, it might handle 30–40% of the data analyst’s day-to-day work. But the remaining 60%? That’s where the real value lies, in interpretation, in advising teams, in aligning insights with strategy, mission, and reality.

Will companies need fewer analysts as AI evolves? Possibly. But what they’ll need more than ever are better analysts, ones who understand not just tools and queries, but products, customers, and outcomes. Analysts who can think, not just analyze.

So if you’re a data analyst reading this with a few years under your belt, my advice is simple: evolve. Learn data engineering. Understand how models work. Stay close to AI, not to compete with it, but to work with it. The best analysts of the future will be the ones who can train, tune, and translate alongside the machines.

Because at the end of the day, AI may be able to analyze data. But it still needs humans to ask the right questions and tell the stories that matter.