What does a data analyst actually do? The job behind the title

“Data analyst” is one of the most searched job titles of the decade and one of the least understood. The course adverts show someone building a glowing dashboard. The reality is closer to a person spending Tuesday morning working out why last week’s revenue number in the finance report does not match the one in the sales report, and discovering that someone changed the definition of “customer” in April.

That is not a complaint. Working out why the numbers disagree is the job, and it is interesting if you like puzzles. But the gap between the marketing and the work explains why so many people finish a certificate, apply to two hundred roles, and hear nothing. This guide describes what analysts actually do all day, which tools you genuinely need, what kind of portfolio gets a first interview, and an honest picture of the junior market in 2026.

A realistic week

A data analyst sits between the people who have questions and the systems that hold the answers. In a typical week at a mid-size company, the work looks something like this.

Monday is requests. A marketing manager wants to know whether the spring campaign brought in customers who stuck around. A product lead wants the sign-up funnel broken down by device. Finance wants the monthly report refreshed. Each request arrives vaguely worded, and the first real skill is turning “did the campaign work” into a question a database can answer: which customers, acquired between which dates, measured how, compared with what.

Tuesday and Wednesday are mostly SQL and cleaning. Pulling the data, discovering that the campaign tag was applied inconsistently, joining it to the orders table, noticing that refunds are recorded in a different table with a different customer ID format. Most analysts will tell you that half their working time goes into getting data into a state where it can be trusted. This is not a failure of the job; it is the job.

Thursday is building and checking. A chart, a table, a dashboard tile, a short written summary. Then sanity-checking it: does the total match the number finance already reports? If not, why not? A number that contradicts what everyone believes had better be right.

Friday is explaining. Presenting to the marketing manager, who wants a yes or no and is getting “the campaign brought in customers with slightly lower retention than average, but the sample is small and the attribution is fuzzy.” Learning to say that clearly, and to say what you would need to be more confident, is what separates analysts who get promoted from analysts who get ignored.

The tools, in order of importance

There is a lot of noise about tools. Here is the honest ranking for most analyst roles.

SQL comes first and it is not close. Nearly every analyst job involves querying a relational database or a cloud warehouse (Postgres, MySQL, SQL Server, BigQuery, Snowflake, Redshift; the dialects differ slightly but the skill transfers). You need joins, aggregation, window functions, common table expressions and the judgment to know when a query is returning nonsense. Interviews for analyst roles almost always include a live or take-home SQL exercise.

Excel or Google Sheets comes second, and people underrate it because it is not fashionable. Stakeholders live in spreadsheets. You will be handed messy spreadsheets and expected to produce clean ones. Pivot tables, lookups, and the discipline to build a sheet someone else can follow are daily tools. Our guide to learning Excel for work covers the functions that matter.

A BI tool comes third: Tableau, Power BI, Looker, or one of the newer alternatives. Companies usually have one and expect you to learn it on the job, so knowing any of them well enough to build a clean dashboard is sufficient. Power BI is more common in corporate and finance-heavy environments; Tableau in marketing and consulting; Looker in tech companies on Google’s stack.

Python or R comes fourth, and for many analyst roles it is optional. It matters when the data is too big or too messy for a spreadsheet, when you need to automate a recurring pull, or when you are doing anything statistical. Pandas and a plotting library cover most analyst use. If you learn Python, learn it for data cleaning and automation, not for machine learning; that is a different job.

Statistics underpins everything and is rarely tested directly. You need to understand distributions, why averages mislead, what a confidence interval is, why a correlation in a dashboard is not a cause, and how to read an A/B test. Not a degree’s worth, but a good textbook’s worth.

Then there is the AI question. Large language models now write competent SQL from a plain-English prompt, and every analyst uses them. This has not removed the job; it has shifted it. The value is no longer in typing the query. It is in knowing what question to ask, whether the tables the model picked are the right ones, and whether the answer is plausible. An analyst who cannot read SQL cannot check the SQL the model wrote, which is why the skill still matters.

Portfolio projects that get interviews

Hiring managers for junior analyst roles see the same three projects over and over: the Titanic survival dataset, a Netflix titles dashboard, and a COVID case tracker. They prove you completed a course. They do not prove you can do the job, because the data was clean, the question was given to you, and there was no stakeholder.

Projects that stand out share a few features:

• You chose the question, and it is a business question. “Which neighborhoods in my city have the biggest gap between short-term rental supply and hotel capacity” is a question. “Exploratory analysis of Airbnb data” is not.
• The data was messy and you show how you cleaned it. A short section on what was wrong (duplicates, inconsistent dates, missing values) and what you decided to do about it is more impressive than the final chart.
• There is a written narrative. Two to four paragraphs, the kind you would send to a manager: here is what I found, here is how confident I am, here is what I would look at next. Most junior portfolios have charts and no words.
• You state the limitations. A sentence like “this dataset only covers listings active in March, so seasonal patterns are not captured” is exactly what a senior analyst would write.
• The code and data are reachable. A GitHub repository with a clear README, a SQL file or notebook, and a link to the dashboard if there is one. Reviewers spend two minutes. Make those two minutes easy.

Good sources of messy, real data: city open data portals (permits, 311 calls, transit ridership), government labor and census statistics, sports results, and your own life (bank exports, a spreadsheet of every job application you sent and what happened). Two or three strong projects beat ten weak ones.

The junior market, honestly

The entry-level analyst market in 2026 is crowded. A large number of people completed analytics certificates and bootcamps between 2020 and 2024, and the roles did not multiply to match. Generic junior analyst postings routinely attract hundreds of applications within days. A certificate alone, including the well-known ones from Google and Microsoft, does not get you an interview; it gets you past a keyword filter and into a pile with everyone else who has the same certificate.

What does work, in rough order of how often we see it:

The internal move. By far the most common way people become analysts is from inside a company, in a role that touched data: operations, customer support, finance, marketing, sales ops. They start doing the reporting nobody else wants, become the person who “knows the numbers,” and move sideways into an analyst title. If you are employed anywhere, this is the highest-probability route. Volunteer for the report. Fix the spreadsheet. Ask the analytics team if you can shadow.

The adjacent title. Reporting analyst, operations analyst, business analyst, financial analyst, marketing analyst, revenue operations, sales operations. These roles are often less contested than “data analyst,” involve the same skills, and are a natural stepping stone. Do not screen them out because the title is not the one you wanted.

Domain first, data second. An analyst with a background in healthcare, logistics, retail buying or insurance claims is worth more to an employer in that industry than a generalist with a slightly better portfolio, because they know which numbers matter. If you have a previous career, aim your applications at that industry and say so in the first line.

Smaller companies. A 60-person company hiring its first or second analyst cares less about a pedigree and more about whether you can be trusted to build the sales dashboard without supervision. These jobs are found through direct applications, local networks and founders posting on LinkedIn, not through large job boards.

For a realistic view of how applicant filtering works before a human sees your application, read our piece on AI screening and how to apply.

What junior analysts get wrong in the first year

The technical mistakes are recoverable. The ones that hurt careers are about judgment.

Presenting a number without checking it against one the business already trusts. Answering the question that was asked instead of the one the person needed answered. Building a dashboard with fourteen filters that nobody uses. Not writing down where the data came from, so that six weeks later nobody, including you, can reproduce the result.

The habit that fixes most of these: before sending anything, write one sentence starting “So what this means is…” If you cannot finish the sentence, you are not done.

Where the role leads

A data analyst title is a starting point, not a destination, and the ladders diverge after two or three years.

Senior analyst and analytics lead. The straightforward path: bigger questions, more autonomy, mentoring juniors, owning a domain (marketing analytics, product analytics, finance analytics).

Analytics engineer. The fastest-growing adjacent role over the past few years. Analytics engineers build and maintain the data models that analysts query, usually with tools like dbt on a cloud warehouse, and they need solid SQL, some software engineering discipline (version control, testing) and an understanding of how the business defines its metrics. Analysts who like the cleaning-and-modeling half of the work more than the presenting half often move here, and it pays better.

Product analyst and product management. Analysts embedded with product teams get close to decisions about what to build. Some move into product management, where the analytical instinct is valued and the data skills become a differentiator.

Data science. Less of a default step than it was five years ago. Data science roles increasingly expect a quantitative degree and strong programming, and many companies have merged the “run experiments and build dashboards” part of data science back into analytics. It is still a route, but it needs deliberate study in statistics and machine learning.

Pay in words: junior analyst roles in the US and UK typically start around the median wage for all workers, sometimes a little below in smaller cities; within three to five years, a senior analyst or analytics engineer in a tech-heavy company earns well above it. Finance, tech and consulting pay most; nonprofits and the public sector pay least but often offer better hours and more interesting problems than the pay suggests.

If you are starting from zero, the realistic plan is this: three months of SQL and spreadsheets until you can answer a business question end to end, one messy real-world project written up properly, and applications aimed at adjacent titles and industries you already know. Then, in whatever role you land, become the person who checks the numbers before the meeting. That person gets asked to do the next analysis.