Skills for Data Analysts

SQL is the non-negotiable floor for every data analyst role. But "knows SQL" is not a skill — hiring managers mean specifically: window functions (ROW_NUMBER, RANK, LAG, LEAD, SUM OVER), CTEs for readable multi-step queries, and the ability to explain why a query is slow and how to fix it. Surface-level SELECT/JOIN knowledge will not clear a technical screen at any company that takes analytics seriously.

Despite what data science marketing says, most analysts spend 20-40% of their time in spreadsheets. The hiring bar is not basic cell editing — it is knowing the difference between VLOOKUP and INDEX/MATCH (and why the latter is usually better), building pivot tables stakeholders can filter without breaking, and designing a spreadsheet model someone else can audit.

Building a chart that accurately represents data is not the skill — the skill is building a dashboard that a non-technical stakeholder can use correctly without training and cannot easily misinterpret. Hiring managers evaluate chart choice, axis labeling, filter design, and whether you understand the difference between a chart for exploration and a chart for decisions.

The ability to design and interpret an A/B test — including statistical significance, p-values, confidence intervals, and common misinterpretations (peeking, multiple comparisons) — is tested in analytics interviews across consumer tech, e-commerce, and SaaS. Getting this wrong in practice costs companies millions in bad product decisions.

Python has crossed from "nice-to-have" to table stakes at most mid-size and large tech companies. The specific expectation: using pandas for data manipulation (groupby, merge, pivot), numpy for numerical operations, and at least one visualization library. Not machine learning — pure data wrangling and EDA.

The most common failure mode for data analysts is producing technically correct analyses that no one uses. The skill is framing the output for the audience: what decision does this inform, what is the recommendation, and what is the one number a VP will remember in a meeting. This is consistently ranked as the hardest thing to hire for and the easiest thing to demonstrate in an interview.

Real-world data is dirty. Analysts who cannot handle missing values intelligently (not just dropping nulls), deduplicate records across sources, or join tables with inconsistent keys produce analyses that are technically wrong. This is a hiring filter at companies with real data — not a nicety.

dbt is the fastest-growing tool in the modern data stack and is now explicitly required in an increasing share of analytics job postings. It brings software engineering practices (version control, testing, documentation) to SQL transformations. Analytics engineers who know dbt can build reliable data models that scale; analysts who only know ad-hoc SQL cannot.

Relevant at companies with a strong quantitative culture (tech analytics, financial analytics, research-heavy orgs). R remains the standard for statistical modeling and publication-quality figures in academic-adjacent analytics work. For pure product analytics at tech companies, Python covers the same ground.

The ability to present an analysis to a room of non-technical stakeholders — verbally, not just in a slide — and field questions without getting defensive is one of the highest-value skills for senior analyst progression. Most analytics candidates underinvest in this relative to technical skills.

Analytics teams that version their SQL and Python in Git catch errors faster and collaborate better. Not a universal requirement — some companies still email CSVs — but increasingly expected at companies with a modern data stack. Shows engineering maturity.

Product analytics roles specifically look for candidates who speak the language of product metrics without needing a dictionary. Knowing what DAU/MAU ratio signals, how to calculate LTV correctly, and what a healthy retention curve looks like for a SaaS vs. e-commerce product is prerequisite knowledge at consumer tech companies.

Analysts who use LLMs effectively (Claude, ChatGPT, Copilot) to accelerate SQL writing, interpret visualizations, draft stakeholder narratives, and explore datasets faster are measurably more productive. The skill is knowing when AI output is reliable and when to check it — not using it blindly.

The majority of modern analytics work runs on a cloud data warehouse, not on-premise databases. Snowflake, BigQuery, and Redshift have specific syntax and performance characteristics that differ from standard SQL. Companies increasingly list specific warehouse experience as a requirement.

Senior analysts at data-mature companies are increasingly asked to evaluate or interpret ML model outputs — feature importance, model drift, precision-recall tradeoffs — even when a data scientist owns the model. Not building ML from scratch, but literacy in what the outputs mean and how to validate them.

Written by the creator of pandas. More reference than tutorial — you will return to it repeatedly. Covers the full pandas API with examples on realistic datasets. The best single investment for Python data analysis depth.

Best for: Python for data analysis; must-have tier.

Freely available, structured SQL curriculum that covers window functions and advanced queries in a real query editor. Better than LeetCode for analytics-specific SQL because the problems are framed as business questions, not abstract algorithms.

Best for: SQL depth; must-have tier.

The official dbt learning path, freely accessible. Covers dbt fundamentals, testing, documentation, and deployment. Taught through hands-on projects, not lectures. The fastest path to a credible dbt portfolio project.

Best for: dbt analytics engineering; nice-to-have tier.

The most useful book on data visualization for non-statisticians. Teaches which chart to use when, how to reduce chart junk, and how to direct the reader's attention. Before-and-after examples are directly actionable.

Best for: Data visualization and stakeholder communication; must-have tier.

Unlike LeetCode, StrataScratch problems are framed as business questions with real company contexts (Airbnb, Lyft, Microsoft). Window function and cohort analysis problems map directly to what analytics interview screens test.

Best for: SQL and Python interview prep; must-have tier.

BigQuery's sandbox gives you 1TB of free monthly query processing with access to public datasets. The best way to learn cloud data warehouse SQL patterns without a paid subscription. Use it with public datasets like Wikipedia or GitHub activity.

Best for: Cloud data warehouse skills; emerging tier.

Covers product metrics, growth frameworks, and experimentation design with examples from real consumer tech companies. Not a teaching resource — context-building for analysts who want to understand how PMs and executives think about data.

Best for: Product metrics fluency; nice-to-have tier.

The best free statistics curriculum for non-statisticians. Goes from basic probability through hypothesis testing and confidence intervals. The explanations are visual and intuitive — better than most textbooks for building genuine understanding vs. procedural knowledge.

Best for: Statistics fundamentals; must-have tier.