AI is not replacing analysts. It is eating the dull half of the job and moving the value to judgement. Here is the safe place to start using it this week.

AI will not replace your analysts, but it will change the job

The headline says the machines are coming for the analysts. The reality, in the teams we actually sit in, is duller and a good deal more useful. AI is not replacing the analyst. It is eating the dull half of the work, the documentation, the first-draft code, the test scaffolding, the patient explaining of some inherited mess nobody wrote down, and in doing so it is pushing the value of the role somewhere more interesting. Towards judgement. Towards definitions. Towards asking the right question in the first place.

We have written more on this via our AI work, and What AI can’t do for a mid-market business, and where it goes wrong takes a closer look at a related part of the picture.

That is a promotion, not a redundancy, but only for the people who treat it as one.

Two ways to get this wrong

The first is to ban it, formally or by sheer disapproval. The team falls behind without anyone noticing the moment it happens, and the best people, the curious ones, leave for somewhere that lets them work the modern way.

The second is to trust it blindly. An AI tool will write you a measure, summarise a dataset, or answer a business question with perfect fluency and no idea whether it is right. Feed it loose definitions and it will invent its own and state them with conviction. A good many organisations are about to learn, in front of their own boards, exactly how solid their data governance was, because a confident machine is now repeating their messiest assumptions back to them in a tidy sentence.

The safe path runs between the two, and it is not complicated.

Where to start, this week, safely

Point AI at the work that is high on tedium and low on risk, the tasks where a human checks the output anyway and a mistake costs little. Documenting an existing model. Drafting DAX or SQL an analyst will review before it goes anywhere near production. Generating test cases. Explaining what an inherited report is actually doing. Real time saved, with a safety net built in.

Keep people firmly on the judgement, the part where being confidently wrong is expensive. What should we measure. What does this number mean. Should we trust this result, or does it merely look tidy. That is not drudgery to automate away. It is the job, and it grows more valuable as the drudgery disappears.

So, something concrete to try. Take your messiest, least documented report, the one only one person understands, and ask an AI tool to document it and explain what it does. Then have the person who knows it check the result. You will learn two things fast: how much time this genuinely saves on the tedious work, and exactly where it confidently gets things wrong. Both are worth more than any opinion piece on whether AI is coming for anyone's livelihood.

This is the instinct behind how we build on Microsoft Fabric and how we think about analytics capability: let the tools take the toil, keep people on the judgement, and never let a fluent answer pass for a correct one. The analyst's job is not vanishing. It is being upgraded, and the people who see that early are the ones worth hiring.

If any of this sounds familiar, talk to us about your data.

Related reading

• Are your people happy? The first question in any Data or AI project
• Decision intelligence versus the dashboard: what the deal is really about
• What Decision Intelligence Actually Means