The fastest-moving frontier in office work
Analysis sits near the top of real-world AI usage for a simple reason: its raw materials are already digital, its outputs are structured, and a wrong first draft is cheap to catch. Writing a query, cleaning a table, summarizing a dashboard and drafting a plausible explanation of a trend are all tasks current systems perform quickly. Anthropic's Economic Index finds analytical work among the most common uses of AI assistants, mostly in collaboration with a person rather than instead of one.
The catch is that plausible is not the same as true. A generated explanation of why revenue dipped will always sound confident. Whether it survives contact with how the data was actually collected is a question the model cannot answer, because the answer lives in your organization, not in the table.
From chart producer to decision owner
When chart production is nearly free, an analyst paid to produce charts is competing with a prompt. The durable role moves up one level on each side of the pipeline. Upstream: metric and experiment design, deciding what is worth measuring, how it can be gamed and what would count as evidence. Downstream: decision ownership, turning a finding into a recommendation someone acts on, and standing behind it when it is challenged.
Data literacy about generation also compounds: knowing which pipeline produced a table, what silently changed last quarter and which fields are unreliable is organizational knowledge that no assistant has. Analysts who hold it become the verification layer for everyone else's AI-generated numbers, an increasingly explicit part of the job.
A realistic 90-day plan
First month: adopt AI assistance for query drafting, but validate every generated query against a case where you already know the answer before trusting it on unknowns. Keep the failure examples; they teach you the tool's blind spots faster than documentation does.
Second month: write definitions for the five metrics your team argues about most, including edge cases and known gaming patterns. Publishing them makes you the reference point. Third month: take one dashboard nobody acts on and either connect it to a real decision or retire it. The story of that change is a better promotion case than any number of new charts.
