I Was Wrong About AI Replacing Statisticians

In August 2025, I published a blog titled “3 Reasons AI Won’t Replace Statisticians Anytime Soon.” A month later, I followed it with “Why Statisticians Will Survive the AI Job Takeover.”

I was pretty confident when I wrote those articles.

I was also wrong.

I started questioning my position only a few months later, but I decided not to remove those posts. I wanted to let them sit there and see how they aged.

They didn’t age very well.

What Changed My Mind

I’ve spent more than 20 years working with data and statistics. Over that time, I’ve used tools such as Excel, SQL, R, JMP, Minitab, Python, and business intelligence systems.

For most of my career, the software was just a tool.

You still had to know what analysis to perform. You had to organize the data, write the code or build the model, interpret the output, create the charts, and then explain what everything meant.

That required experience.

When generative AI first arrived, I viewed it the same way: another tool that could make a statistician more productive.

I underestimated it.

The biggest change isn’t that AI can calculate a regression, run an ANOVA, build a predictive model, or write Python code. We’ve had software capable of sophisticated statistical calculations for decades.

What impresses me is that AI is increasingly able to participate in the entire analytical process.

Give it a dataset and start asking questions.

What’s happening in these data?

Are there trends?

Are these groups statistically different?

What variables are driving the result?

Are there outliers?

What analysis should I perform?

Can you build a model?

Can you explain the results without using statistical jargon?

Can you create a chart that an executive can understand?

Then ask another question.

And another.

And another.

You don’t necessarily need to know the programming language. You don’t even need to know the name of every statistical method that might answer your question.

You increasingly just need to know what you want to learn.

That is a much bigger change than I anticipated.

The Barrier to Data Analysis Is Collapsing

This is the part I got wrong in my earlier articles.

I assumed that advanced statistical knowledge would continue to provide a substantial barrier between statisticians and everyone else.

AI is rapidly lowering that barrier.

A business manager shouldn’t have to learn R to investigate a sales problem. A scientist shouldn’t necessarily have to become an expert programmer to explore experimental data. A small-business owner shouldn’t need to hire someone simply because they don’t know how to write SQL.

They can increasingly upload their data and have a conversation with it.

That’s remarkable.

It is also disruptive for people like me who have made a career out of being the person who knows how to get answers from complicated data.

But AI Still Needs Something From Us

I’m not ready to make the opposite mistake and claim that human expertise no longer matters.

AI can analyze the data you give it.

That doesn’t mean the data are good.

Anyone who has worked with real business data knows how messy it can be. Product descriptions don’t match. Units are inconsistent. Fields are missing. Definitions change. Two databases supposedly measuring the same thing produce different numbers.

Sometimes the hardest part of an analysis has nothing to do with statistics.

It’s figuring out what the data actually mean.

There is also another problem: AI doesn’t automatically know which question matters most to your business.

A human still needs to provide context, challenge questionable results, recognize when something doesn’t make sense, and decide what to do with the answer.

In other words, prompting AI isn’t simply typing:

“Analyze this spreadsheet.”

The quality of the analysis still depends heavily on the questions being asked and the information being provided.

But that doesn’t change my larger conclusion.

The amount of technical work AI can perform continues to grow, and I no longer believe it is reasonable to assume that statisticians and data analysts will be protected simply because their work is complicated.

I Have to Admit: I’m Impressed

There is something humbling about watching technology learn to perform tasks that took me years to master.

But there is also something exciting about it.

Imagine giving advanced analytical capabilities to millions of people who previously couldn’t use them.

A person with a question and a spreadsheet can potentially perform analyses that once required specialized software, programming knowledge, and an experienced analyst.

That’s an enormous increase in capability.

And despite what I predicted in 2025, I now believe we’re only beginning to understand what that means.

So What Happens to Statisticians?

I don’t know.

And after getting my earlier prediction wrong, I’m going to be a little more careful about predicting the future.

I don’t believe statisticians disappear tomorrow.

But I do believe the profession is going to change dramatically.

The value may shift away from simply knowing how to perform an analysis and toward knowing what should be analyzed, whether the data can be trusted, what questions should be asked, and whether the answer makes sense.

The statistician of the future may spend much less time writing code and much more time directing, validating, and interpreting analyses performed with AI.

And some work that once required a statistician simply won’t anymore.

That’s difficult for someone in my profession to say.

But pretending otherwise won’t stop it from happening.

Leaving My Old Predictions Up

I considered deleting my earlier articles.

I’ve decided not to.

They’re staying.

They represent what I genuinely believed at the time, and I think there is value in being willing to admit when new evidence changes your mind.

Statistics itself teaches us to do that.

We form a hypothesis. We collect evidence. We analyze the results.

And sometimes the evidence tells us our original hypothesis was wrong.

When that happens, the answer isn’t to hide the data.

It’s to change your conclusion.

So that’s what I’m doing.

I was wrong about how quickly AI would become capable of performing sophisticated data analysis.

I underestimated it.

And as a statistician, I have to admit:

The data changed my mind.

— Brian Anderson, Statistician and Founder, Topline Statistics LLC