A data analyst turns data a business already has into answers it can act on: building reports and dashboards, checking whether a number is real or an artefact, and explaining what changed and why. They work mostly in SQL, spreadsheets and a BI tool, and their output is a decision, not a pipeline.
A data analyst turns data a business already holds into answers it can act on. In practice that means building the reports and dashboards people actually read, checking whether a number is real or an artefact of how it was counted, and explaining what changed and why — usually in SQL, a spreadsheet and a BI tool.
The output of the job is a decision, not a dataset. An analyst who produces a beautiful dashboard nobody uses has not done the work; one who tells you that your margin fell because of one customer's return rate has.
Four recurring activities. Pulling data out of the systems that hold it, usually with SQL. Cleaning it, which takes far longer than anyone budgets and is where most errors are caught. Analysing — comparing periods, segmenting, testing whether a difference is meaningful or noise. And communicating: a chart, a short written conclusion, a recommendation someone can act on this week. The last one is what separates a good analyst from a report generator.
An analyst answers questions with data that exists. A data engineer builds the plumbing that makes that data arrive reliably. A data scientist builds models that predict or classify rather than describe. In a company under fifty people these are usually the same person wearing three hats, and the honest sequence is analyst first — there is no point predicting next quarter if nobody trusts last quarter's number.
When decisions are being delayed or argued about because nobody agrees on the number, and when the person currently producing reports is doing it as an evening job on top of another role. Below that threshold a well-built dashboard and a clear definition of each metric usually solves it. Above it, the cost of not having one shows up as decisions made on the loudest opinion in the room.
Usually not to start. An analyst can work directly against your operational systems for a first set of questions, and doing so surfaces exactly which data problems are worth solving. A warehouse earns its cost when several systems must be combined routinely and the manual joining becomes the bottleneck — the threshold set out in do you need a data warehouse.
Fewer than the job adverts imply. The core is stable and has been for years.
It replaces parts of the task and none of the responsibility. Models are now good at writing queries and drafting explanations.
Depends on whether the questions are recurring or one-off.
Want this applied in your business? See how we take it to production:
We build this AI in production — at fixed prices, with one named expert. Start with a free consultation.
Book a free consultation →