A request that once took an analyst an afternoon (pull the data, build the chart, write a few lines on what changed) can now be drafted by AI in the time it takes to describe it. Copilot in Power BI drafts report pages and calculations, assistants write first-pass commentary on variances, and leaders increasingly put their questions to an assistant instead of sending them to the analytics team.
It is tempting to conclude that analysts matter less. The opposite is true, but the job changes. Producing a first draft, the part of the work that AI does well, was never where most of the value lay. The parts it cannot do, such as knowing what a number means, judging whether an answer is right and connecting it to a decision, become the job. Leaders who treat AI only as a way to shrink the team tend to discover that the drafts got faster and the decisions did not get better.
What AI takes off the desk
First drafts go first: standard charts, routine queries, commentary that restates what a chart already shows, and the documentation nobody had time to write. So does much of the monthly production of report packs, and the simple questions that used to arrive by email and wait in a queue.
Drafts still need checking, and that is the point. An analyst who spent most of the week building now spends more of it reviewing, correcting and deciding what deserves attention. A draft that looks finished is easy to accept; checking a plausible answer takes more discipline than producing one, and that discipline is now the core of the job. The time freed is only valuable if it is redeployed deliberately; otherwise it fills up with more of the same requests, delivered faster.
From building reports to curating definitions
When AI answers questions from the semantic model, the layer that holds the measures and business rules behind every report, the definitions in that model become the product. Someone has to decide what counts as an active customer, which date a booking follows and why one margin is certified while the others are not, and then write it down in names and descriptions an assistant can use.
That someone is the analyst. Curating definitions means agreeing the meaning of each measure with finance, sales and operations, keeping descriptions and synonyms current, retiring duplicates and noticing when the business changes faster than the model does. It is quieter work than building dashboards, and it decides whether every AI answer in the organization is right. It often starts with a glossary: the terms leaders actually use, each with an owner, a definition and the measure that implements it.
From answering questions to testing answers
As leaders ask assistants directly, the analyst moves from answering each question to making sure the assistant answers well. That means keeping a set of real questions with known answers, running it whenever the model or the assistant changes, and investigating every failure.
Investigation is a skill in its own right. A wrong answer can come from the data, from a missing or ambiguous definition, from a relationship that allows two paths, or from the instructions the assistant was given. Finding which, and fixing it at the source instead of patching the symptom, is the analyst’s new quality control. Over time, the pattern of failures shows where the model is weakest and which questions leaders keep asking that nothing yet answers.
From charts to decisions
The most valuable work moves closer to the decision. Before any analysis, the analyst helps leaders frame the choice: what is being decided, which options are realistic and what evidence would change the outcome. Afterwards, the analyst explains the trade-offs, such as price against volume, service against inventory or growth against margin, and is candid about what the data cannot say.
AI is good at describing what happened. It is far less reliable at judging what matters, what is noise and what a leader should do next week. That judgment, grounded in knowledge of the business, is what leaders value most and what the analyst is best placed to give. A simple test of the new role: after a meeting, can the analyst say which decision was taken and what evidence moved it?
The skills to build
The profile of a strong analyst shifts accordingly. These are the skills worth investing in:
- Business and financial literacy: knowing how the organization makes money and how each measure connects to the financial statements.
- Modelling and definitions: designing semantic models that are simple, well described and safe for AI to read.
- Evaluation: writing test questions, checking answers against certified numbers and tracing errors to their cause.
- Writing for machines and people: clear descriptions, precise instructions for assistants and short, plain commentary for leaders.
- Facilitation: running the conversations in which functions agree on a definition or a decision.
Technical depth still matters, but it is no longer enough on its own. The analysts who thrive combine it with the judgment and communication that AI cannot supply. When hiring, look for curiosity about the business as much as for technical skill; the tools can be taught faster than the instinct.
How teams reorganize
Teams that adapt well tend to make the same moves. They separate run from change: keeping reports available, fixing incidents and handling routine requests becomes a service with clear service levels, run by a platform team or a managed analytics partner. They embed analysts with business domains as decision partners rather than report builders. And they keep a central group that owns the semantic models, the definitions and the standards that every assistant depends on.
Run work handled as a service tends to become more reliable and more visible, and it frees analysts for the work above. Measure analysts on decisions supported and answers trusted, not on reports shipped. Career paths have to follow: if promotion still rewards the number of dashboards built, the best analysts will keep building dashboards.
The bottom line
AI does not remove the need for analysts. It removes the excuse for spending their time on first drafts. Organizations that move their analysts toward definitions, testing and decisions, and rebuild their teams around that work, will get more from AI and more from their people. Those that simply ask the same team to produce more, faster, will get more reports and no better decisions. The shift will not happen on its own; it needs leaders who ask their analysts for judgment, not just for output.