For most of data analysis's history, it was a specialist skill. You needed to know SQL to query a database, Python or R to run statistical models, and enough statistics to know what the output meant. Most people in organisations — marketers, managers, operations staff — couldn't do any of it without help from a data team.

AI is changing that. Not by eliminating the need for expertise, but by removing the technical barriers that kept non-experts out.

What AI has actually changed

Natural language queries. The most significant shift is the ability to ask questions in plain English and get data back. "What were our top five product categories by revenue last quarter?" used to require writing a SQL query. Now it can be answered by typing that sentence into an AI-powered tool. The translation from question to code to answer happens invisibly.

Automatic summarisation. When you give a modern AI model a dataset and ask it to summarise what it contains, it can identify the main patterns, flag anomalies, note what data is missing, and describe the distribution of key columns — tasks that used to require hours of exploratory analysis. The AI handles the "what does this look like" question so the analyst can focus on "what does this mean?"

Generated insights. Beyond describing data, AI can propose possible interpretations. "Sales are down 18% in the northern region over the last three months — this aligns with the period when the two largest accounts shifted to a competitor." A human still needs to verify and contextualise this, but the AI shortens the path from data to hypothesis.

Code generation. For users who want to go further, AI can write the code. Ask for a Python script to clean a dataset, an SQL query to join two tables, or a formula to calculate the rolling average — and get working code you can run and modify. This doesn't replace the need to understand what the code does, but it dramatically lowers the learning curve.

What hasn't changed

AI has lowered the technical floor but not the analytical ceiling. The hard parts of data analysis were never mostly technical — they were always about thinking clearly about the right questions, understanding the context of the data, and communicating findings to people who will act on them. AI doesn't do any of that automatically.

Asking the wrong question. AI will answer the question you ask, including a poorly framed one. "What's the average customer lifetime value?" might look like a useful query, but if you haven't defined lifetime value consistently, the number you get back is meaningless. Garbage in, garbage out doesn't change just because there's AI in the middle.

Domain knowledge. Understanding that a 2% churn rate is unusually high for an enterprise SaaS company but normal for a consumer app requires knowing the industry. AI can identify that churn went up 2%, but it can't tell you whether that's alarming or expected without context you supply.

Scepticism and verification. AI models can hallucinate — they can produce plausible-sounding results that are wrong. Every AI output in a consequential analysis needs to be verified against the underlying data. The standard for accepting an AI-generated insight should be the same as the standard for any analytical claim: show me the data.

The skill that matters now

The most valuable data skill in the AI era isn't coding or statistics — it's the ability to ask good questions and critically evaluate answers. "Is this what I actually wanted to know?" "Does this number make sense given what I know about the business?" "What would change my mind about this conclusion?"

AI lowers the cost of exploration, which means more time available for the interpretation and communication that creates real value. The people who benefit most are those who use AI to do more analysis, not those who use it to do less thinking.