All articles

KajoData / Career & business

Data analyst requirements: choose your learning order

I turn a list of tools into a learning plan with a specific task, an output and a way to check the result.

Kajo RudzińskiPublished Updated 14 min read

Three connected platforms carrying a table, a database and a chart as successive learning stages.

A data analyst requirements list can look like an inventory of the internet. Excel, SQL, Python, Power BI, Tableau, statistics, cloud platforms, communication. Open another vacancy and three more things appear. Try learning everything at once and you can finish with an impressive collection of half-completed courses without a single piece of work you can explain independently.

I started with Excel while working in customer service. It let me apply a new tool to problems I already understood. I still see that as a useful starting point, but I would not turn my own route into a compulsory sequence for everyone. Here is how I would choose the next skill based on the work I want to do and the evidence of what I can already deliver.

Choose the kind of work first

Before comparing courses, I choose a direction. Someone reporting sales, a product analyst and someone developing predictive models may share part of their toolkit, but their daily responsibilities can differ substantially. A job title alone does not tell me how much time I will spend writing SQL or discussing the definition of a metric. I need to understand the actual work.

Collect a few current vacancies that match the location, level and kind of employer you would realistically consider. Write down the tasks first, with the tools beside them. Separate essential requirements from desirable ones. This small sample can guide your learning decisions, but it is not a study of the entire employment market. You are comparing specific opportunities rather than discovering a universal analyst checklist.

For each task, describe something you would produce. Replace “SQL” with “prepare a weekly order summary and explain where the total comes from”. Replace “communication” with “agree whether a return reduces sales on the purchase date or the return date”. Those descriptions give you something to practise. Rating your own knowledge with five stars provides considerably less direction when you sit down to study.

A table makes a useful first assessment

From the beginning, I want to know what one row represents. An order, an order line, a customer and a store’s daily result are different levels of detail. Mix them up and a technically correct formula can still produce a misleading report. Before building a chart, I establish the key, the unit, the reporting period and the meaning of missing values. These skills remain useful when the software changes.

Take a synthetic example with three rows carrying order identifiers A, A and B. The first order contains two lines and the second contains one. We have three order lines and two orders. A report answering a question about the number of orders therefore needs more than a row count. A tiny dataset makes this distinction visible without hiding it behind an elaborate tool or an unfamiliar interface.

In Excel, I would practise a structured table, data types, filtering, simple summaries and matching information from another dataset. I also want checks: do store totals reconcile with the overall figure, are missing prices visible, and does a new row enter the summary? Start with a problem whose expected outcome you understand. If the answer changes, you should have a way to investigate rather than simply trusting the latest refresh.

SQL brings the work closer to the source

When data lives in a database, SQL lets me select relevant records, join tables and prepare aggregates. That does not mean every large table belongs in a relational database or every small dataset requires Excel. The choice also depends on the company’s systems, available access and how results will be handed over. I want to learn the environment in which the task will actually happen.

My starting set would include filtering, grouping and joins, followed by more complex queries, window functions and readable code organisation. I would keep checking row counts throughout. The PostgreSQL joins tutorial explains how joins pair matching rows. Those matches determine whether records multiply or disappear. Understanding that behaviour matters before adding another sophisticated feature to a query you cannot yet verify.

For our order example, imagine adding a store lookup table. First I would check whether each store code appears only once in that lookup. Two matching entries for one code can multiply order lines and inflate the total. Explaining this failure is stronger evidence of SQL understanding than naming a long list of keywords. There is more on choosing a learning sequence in my SQL learning plan.

Pick one BI tool for a specific purpose

Once I can prepare the data, I need to communicate a result. That is where Power BI, Tableau or the reporting environment already used by an employer comes in. I would begin with one. I want to understand its data model and filters, then build a report that somebody can read without me narrating every click. Learning two interfaces simultaneously can distract from those underlying questions.

The Microsoft PL-300 study guide includes data preparation, modelling, visualisation, and managing and securing Power BI. I use it as a map of one ecosystem. It is not a description of every employer’s requirements or evidence that you need to sit an exam immediately. It does show how much work comes before choosing the colours of a bar chart.

For the order exercise, a sales view by store and week is enough. Readers should see the period, unit and treatment of returns. I check what a store filter changes and whether the combined figure still reconciles with the source. A dashboard with attractive movement but an unexplained total can wait. First I need a result that can withstand a straightforward question about where its numbers came from.

Add Python when it has a job to do

Python can help with automation, data processing and tasks that are awkward in the current environment. It can also be a natural first language if you already have programming experience or are targeting a role that requires it from the outset. You do not have to wait until you have ceremonially finished all of Excel. That moment is unlikely to arrive, and it would be an unhelpful condition for progress.

Before adding a language, however, I ask which problem it will solve. Monthly files with the same structure may also make a good Power Query exercise. Microsoft describes Power Query as a data preparation and transformation engine with repeatable steps. I would compare the available options rather than deciding in advance that writing a script is the only professional response to a recurring task.

If I choose Python, I practise the complete process: read the data, inspect types, transform it, check the result and save something another person can use. I want to explain what happens with a missing value, a new column or a malformed input file. Copying code without understanding its inputs and outputs creates maintenance trouble. An impressive spreadsheet formula that nobody dares to touch has much the same problem.

A workbench connecting raw cubes, a table, a database and a chart

Connect statistics to the question

I would not postpone statistical thinking until every tool is mastered. Even a simple report requires a choice between a total, mean, median or proportion. I need to know what an observation is, what belongs in the denominator and whether the groups are comparable. At this stage, interpreting a simple result correctly gives me more value than mechanically applying a more complicated method whose assumptions I cannot explain.

Suppose, in a synthetic example, that one store handled two orders and another handled two hundred. Their average order values should not automatically receive equal weights in a company-wide result. Before choosing a formula, I decide whether I want to describe the typical order or the typical store. Those are different questions. Both can be useful, but changing between them without explanation can mislead the reader.

Start with the median and the distribution of values, then return to statistics as your questions develop. Avoid attaching a cause to a difference merely because it appears in a chart. Growth following a campaign does not establish that the campaign caused it. Being able to name this limitation is a practical communication skill, particularly when somebody wants to turn a descriptive report into an immediate decision.

Let one project connect the stages

Instead of creating a new project for every library, develop one small subject. A fictional shop’s orders, two related data sources and a clearly defined period are enough. Prepare the table first, move the relevant steps into SQL later, and finally build a simple report. Each stage addresses the same business question, making it easier to see what the new tool contributes and what you still need to understand.

Add a definition of the result. State whether amounts are before or after returns, which records you exclude and why. Describe missing information and reconciliation checks. Say explicitly when the dataset is synthetic. If it is public, identify the source and examine the conditions for reuse. Company data does not become suitable portfolio material merely because you deleted the column containing people’s names. Use a source you are authorised to share.

A good project explanation also includes decisions you declined to make. You might leave out a forecast because the history is insufficient. You might preserve a missing price as an exception instead of replacing it with zero. This gives the reader access to your reasoning. A reviewer should not have to guess whether an absent feature reflects a deliberate boundary or a tutorial abandoned before the final lesson.

Assess the skill without following along

After a lesson, close the material and recreate the solution with changed data. If you can only complete it with identical column names and the original file layout, you need more practice. That is an ordinary stage of learning. I simply want to distinguish recognising a familiar screen from solving a problem independently. The two can feel remarkably similar while watching somebody else perform every step in the correct order.

I use three questions for self-assessment. Can I start from an empty file? Can I check the answer by another route? Can I explain when this approach will produce an incorrect result? Then I add a more difficult variation. I do not expect every function to be memorised. Finding the relevant documentation is part of the job, provided I understand what I am looking for and how it applies to my input.

Ask another person to read the project description and identify the period, source and meaning of one figure without asking you. If they need basic clarification, you have a concrete improvement to make. This review does not replace technical checks, but it quickly reveals whether the result communicates clearly. The eventual report will also be used when you are unavailable, so the explanation needs to travel with the numbers.

Build the plan around available time

First choose one task for the next stage. Write down the expected output, how you will check it and why it matters to the role you selected. Only then search for a lesson. Starting from a course catalogue makes it easy to confuse the availability of teaching material with a personal priority. A fresh course will almost always look more appealing than the difficult unfinished section of your current project.

The plan should fit an ordinary week that includes work and rest. Two regular sessions may be a more sensible beginning than a promise of daily marathons. I would not attach a guaranteed number of weeks to finding a job. Progress depends on your starting point, time, feedback and target role. A calendar helps maintain direction; it cannot guarantee the outcome of a recruitment process or make every employer’s expectations identical.

At the end of a stage, review what you can now do independently, where you still need help and which problem appeared next. Then decide whether another technology is needed or whether another exercise in the existing one would help more. If joins are the obstacle, adding a beautiful dashboard is unlikely to fix them. Naming the actual difficulty makes the next learning decision much less mysterious and considerably easier to evaluate.

Change the sequence when your situation requires it

For somebody already working with spreadsheets, my default suggestion is data structure, SQL and one reporting tool, with Python added for a specific purpose. If you already program comfortably, start with the missing understanding of data and business questions instead. If an employer uses a particular platform, practical experience in that environment may be more valuable than studying a fashionable alternative late into the evening simply to keep up with online discussion.

Changing tools does not erase your ability to define a record, reconcile totals or ask useful questions. I carry those skills between environments. Syntax and interfaces change, but I still need to know what I counted. That is why I return to one project and ask whether its meaning survives the technology change. It offers a more honest assessment than counting the software logos that could fit on a CV.

For today, select one vacancy, one task and one small dataset. Write down what you intend to produce and how you will recognise a correct result. If you can do that, you already have the beginning of a useful plan. Add the remaining tools when there is work for them to do. The internet can wait with its next list of thirty indispensable competencies while you finish something you can actually explain.

Exercise: compare three starting points

Imagine three people. The first works in customer service and manually prepares a weekly case summary. The second writes scripts comfortably but has never agreed a result definition with its recipient. The third makes readable charts but receives prepared data and does not know what happened upstream. These are illustrative situations rather than descriptions of actual employees. Each person may need a different next step despite sharing an interest in data analysis.

For the first person, I would suggest structuring the source table and checking whether every case enters the report, followed by a repeatable import. For the second, I would set a short definition task: who will use the result, for which decision, over which period and with which limitations? For the third, I would suggest working from a raw file to an independent summary while checking record counts along the way. None automatically needs the same course package.

Describe your own starting point in four sentences. Which tasks do you already perform? Which tools and data are you authorised to use? Which result can you check independently? What prevents you from completing the next task? If the last answer is “everything”, narrow the task. Replace a complete reporting system with one question answered from one table. That helps distinguish a knowledge gap from a scope that is simply too broad to begin sensibly.

Set a completion condition. A case report might show day and category, reconcile to the source count and display missing categories separately. Prepare a refresh instruction and ask someone to read the result. If a changed input file leaves you unable to explain a different total, return to the data checks. That does not mean you must reconsider the entire career direction; it identifies something specific to practise in your current project.

What the plan cannot establish

A completed project demonstrates a portion of your skills. It does not establish that you can handle every company environment, a large model or responsibility for a production report. Describe that scope honestly in your portfolio. A beginner can do excellent work on a small dataset while still needing deployment support. Naming the boundary helps both sides agree an appropriate role level and understand which responsibilities would initially require help.

Similarly, unanswered applications do not establish that Python is the missing skill. Role fit, presentation of experience, location and other process conditions can matter. Look for specific feedback and compare the vacancy with the material you submitted. Adding another technology after every rejection can turn learning into a chaotic response to an outcome whose cause you have not identified. Use evidence to adjust the plan rather than treating every silence as a new syllabus requirement.

Key takeaways

Where should I start learning data analysis?

Start with a specific task and a well-structured table. If you already use Excel, practise defining records, checking data types and reconciling results there.

Should I learn SQL before Python?

The order depends on the role and your starting point. SQL is a useful next step for database work; add Python when a task or target role calls for it.

Is a certificate enough to get a job?

A certificate can structure learning, but it does not guarantee employment. Show an independent project, a clear result definition, checks and limitations as well.

AI was used in an automated way while working on this article. If you want to learn how to do it too, check out this project.

Kajo Rudziński

Newsletter KajoData

Let’s keep in touch.

I write about data analysis, learning and changing careers. My newsletter and the sign-up form are in Polish. If that works for you, I’d be glad to see you there.