[Beginner] — the analyst discipline, how it differs from data science and data engineering, and the one thing that defines the job.

The kid version first

A company collects mountains of data — every order, click, signup, support ticket. On its own it’s just numbers. Data analysis is the work of answering questions with that data so someone can make a better decision: “Why did sales drop last month?” “Which customers are about to churn?” “Is the new feature working?”

The analyst’s job isn’t to build the warehouse (that’s engineering) or to train a predictive model (that’s science) — it’s to turn data into a decision, usually with SQL and a chart, fast enough to matter.

The three data roles, distinguished

This is the most common confusion, so it’s worth being precise. The three data disciplines form a stack, each with a different deliverable:

DeliverablePrimary toolsThe question
Data engineerA reliable pipeline / warehouseSQL, Python, Spark, dbt, Kafka”How do we get the data here, clean and fresh?”
Data analystA decision or a recommendationSQL, BI tools, spreadsheets, stats”What does the data say we should do?”
Data scientistA model or a rigorous statistical answerPython/R, stats, ML”Can we predict / prove this?”
  • Data engineering builds the warehouse the analyst queries → the plumbing
  • Data analysis (this track) turns that data into answers → the last mile, and the most common data job
  • Data science brings statistical rigour and predictive modelling → the deep end

The lines blur — a “data analyst” at a small company does all three; “analytics engineer” is a hybrid of analyst and engineer who models data in dbt; and the analyst reaches into the data-scientist toolkit for A/B tests and hypothesis testing. But the deliverable distinguishes them: the analyst’s output is a decision, communicated to someone who acts on it.

The analytics maturity spectrum

Analysis comes in four levels of ambition, and knowing which one a question needs keeps you from over- or under-engineering the answer:

   DESCRIPTIVE   "what happened?"           ← reports, dashboards. Most analysis
       ↓
   DIAGNOSTIC    "why did it happen?"        ← drill-down, root-cause. The hard, valuable part
       ↓
   PREDICTIVE    "what will happen?"         ← forecasting, models → data science
       ↓
   PRESCRIPTIVE  "what should we do?"        ← optimisation, recommendation

Most day-to-day analysis is descriptive and diagnostic — what happened and why — and diagnostic is where analysts earn their keep, because “sales dropped” is easy and “sales dropped because the checkout broke for mobile users in Germany” is the insight someone can act on → diagnostic analysis. Predictive and prescriptive shade into data science.

The analysis workflow

Every good analysis follows the same arc, and the beginner mistake is starting in the middle (at the data) instead of the start (the question):

1. QUESTION    what decision does this inform? Be specific
2. DATA        find it, get it, understand it, TRUST it
3. ANALYSE     explore, slice, compute, test
4. COMMUNICATE the finding — as a decision, not a data dump
5. DECISION    someone acts. THIS is the point

Steps 1 and 4 are where analyses fail, and they’re the non-technical ones. A technically perfect analysis of the wrong question is useless; a correct finding communicated as a wall of numbers gets ignored. The analyst’s real skill is bracketing the technical middle with a sharp question and a clear “so what” → communicating.

The one idea that defines the job

The deliverable is a decision, not a number.

An analysis that produces a correct number nobody acts on has failed. This reframes everything:

  • Start from the decision, not the data — “what would we do differently depending on the answer?” If nothing, don’t do the analysis
  • Precision matters only up to the decision — you don’t need the churn rate to four decimal places if the action is the same at 8% and 8.4%
  • Communication is half the job, not an afterthought — the same lesson the data-scientist track opens with
  • Actionable beats interesting — a small finding that changes a decision beats a fascinating one that changes nothing

This is why the best analysts are as much business people as technical ones: they understand what the organisation is trying to do, so they ask the questions worth answering.

What this track covers

The analyst’s actual toolkit, and where each fits:

It borrows the statistical rigour (hypothesis testing, A/B tests, causal inference) from the data-scientist track rather than duplicating it, and consumes the warehouse the data-engineering track builds.

Key insight

Data analysis is the discipline of turning data into a decision — distinct from engineering (which builds the pipeline) and science (which builds models) by its deliverable, which is an answer someone acts on. Most of the work is descriptive and diagnostic (“what happened, and why”), the primary tool is SQL, and the two places analyses actually fail are the non-technical bookends: asking a question tied to a real decision, and communicating the finding as a “so what” rather than a number. Internalise “the deliverable is a decision, not a number” and the rest of the craft organises itself around it.

Source: [reference] — Sep 2026.

↑ Back to Data Analysis