SQL Is Still the Most Useful Skill for Early-Career Analysts

Students preparing for data roles tend to over-invest in modelling and under-invest in querying. In the first two years of an analytics career, the ratio of time spent is closer to nine to one in the other direction.

What the job actually is

Someone asks a question. Usually a vague one — "why did returns go up last month?" Your job is to convert it into something a database can answer, get the data, check whether it is trustworthy, and come back with an answer the person can act on.

Almost all of that is SQL, plus judgement. Very little of it is modelling.

The specific things worth being good at

Beyond basic selects and joins:

Window functions. Running totals, rank within group, period-over-period comparison. A large fraction of business questions are window functions in disguise, and students who know them save hours.

Careful aggregation. Knowing what a join does to your row count before you aggregate. The most common analytical error I see is a sum inflated by a fan-out join, reported confidently.

Date handling. Almost every question has a time dimension. Time zones, partial months, fiscal versus calendar — dull, and the source of most incorrect numbers.

Reading a query plan. Not to optimise everything, just to understand why a query takes eleven minutes.

Practise against messy data

Clean tutorial datasets teach syntax, not the job. Real tables have duplicate rows, nulls that mean three different things, and a status column with values nobody documented.

Find something messy — a public dataset, your college's own records with permission, an export from a club you run. The useful skill is noticing that a number is wrong before you present it.

The habit that separates good analysts

Before running a query, write down what you expect the answer to look like. Then compare.

When the result matches, you have confirmation. When it does not, you have either learned something real or found a bug in your query — and you will investigate, instead of pasting a wrong number into a deck.

Analysts who skip this step are the ones who present confidently incorrect numbers, which is the fastest way to lose credibility in a team.

What to do about the modelling course

Do it, after you can comfortably answer a business question end to end in SQL. Understanding validation, overfitting and honest evaluation matters. But a model built on a query you did not verify is worse than useless — it is convincingly wrong.

A reasonable first three months

• Weeks 1–3: selects, joins, group by, against a messy dataset
• Weeks 4–6: window functions and date handling
• Weeks 7–9: one end-to-end analysis of a real question, written up in one page
• Weeks 10–12: repeat with a different dataset, and make the write-up shorter

That one-page write-up is what you show in an interview. It demonstrates more than a certificate does, because it shows judgement rather than attendance.