All roadmaps

Data Analyst

From raw spreadsheets to decisions a business will actually make.

The complete analyst path for 2026: what analytics actually is, spreadsheets, SQL, Python, statistics you can defend, dashboards, and AI-assisted analysis you can verify - every topic explained, with free resources to go deeper. Fresher and experienced tracks.

Your progress0/139 topics · 0%

Progress is saved in this browser.

13 stages139 topics₹0 to follow

Filled nodes are the focus for this track; outlined ones can wait. Tap a topic to open it and mark it done, learning or skipped.

Data AnalystRelated roadmapsSQL roadmapPython roadmapExcel roadmapAI & Data Scientist roadmapPrompt Engineering roadmapAI Agents roadmapProve it as you goReading is half of it. This roadmap isbacked by a free, scored Data Analystcertification, a realistic job simulationand daily challenges - evidence arecruiter can verify, not a claim. Everytopic is checked against live jobpostings; open the market audit at thetop of the page.Get certified - freeJob simulationDaily challengesKnow the job before the toolsThe analyst job in 2026Which skills to learnSQL, Excel, BI& statsPython foranalystsAI literacy & workflowWhich role to prepare forAnalyst vs BIvs DSFresher vsswitcherSectors, cities & payWhat companies actuallyexpectHiring testsnowCerts &projectsAI on day oneWhat gets you hiredThe 90-day planMistakes toavoidYour proof stackThink like an analystBusiness questions & metricsTurning a vague ask into aquestionThe four kinds of questionsMetrics, KPIs anddefinitionsData grain and uniquenessThe data lifecycleStakeholders, requirementsand pushbackRequirements gathering & theanalytics specWorking in a team: tickets,Agile, reviewsLearn the business you analyseDomain primers: the metrics eachsector runs onLearn a domain in two weeksFinance & FP&A: P&L,variance, marginE-commerce & marketing:funnel, CAC, ROASOperations & supply chain:fill rate, lead timeBFSI: risk, collections,NPA, fraudHealthcare & pharma: claims,outcomesSaaS & product: MRR, churn,activationSpreadsheets, still the fastesttoolExcel & Google SheetsCore functions & referencesSUMIFS /COUNTIFSIF / IFS / AND/ ORDates & DATEDIFTRIM & textfunctionsLookups: XLOOKUP &INDEX-MATCHPivot tablesCharts & conditionalformattingPower Query: clean & combineAI inside spreadsheets:Copilot & GeminiSQL is non-negotiableSQL fundamentalsSELECT, WHERE, ORDER BY,LIMITJoins and keysGROUP BY, HAVING andaggregatesNULL handlingDates, strings and CASESQL for analysisSubqueries & CTEsWindow functionsValidating a query: counts &reconciliationCohorts, funnels andretentionEXPLAIN and queryperformanceWriting SQL with AI, safelySQL interview practice setsPython for analysisPython & pandasPython basics for datapandas: load, filter, group,mergeCleaning with pandasNotebooks & reproducibilityPulling data: files, APIs,databasesPlotting: matplotlib,seaborn, plotlyAutomating reports withscriptsAI-assisted coding:generate, review, testData quality: the real 70%Data quality & cleaningProfiling a new datasetMissing data strategiesDuplicates & outliersTypes, formats andstandardisationReconciling against thesource of truthDocumentation & datadictionariesData governance: definitions,owners, lineage, accessPrivacy: PII, masking andthe DPDP ActStatistics you can defendStatisticsDescriptive statisticsMean, median,modeRange,variance, SDDistributionshapePercentiles &outliersSampling & biasUncertainty & confidenceintervalsHypothesis tests & p-valuesCorrelation vs causationA/B tests & experimentdesignRegression basicsForecasting & time seriesbasicsVisualise it, then tell thestoryDashboards & storytellingChoosing the right chartKPI design & definitionsData storytelling &executive summariesPresenting and defendingyour numbersPower BI & Tableau, properlyChoose your tool: Power BI,Tableau, LookerConnect, clean and refresh:Power QueryData modelling: star schema& relationshipsMeasures & DAX basicsTime intelligence & DAXpatternsVisuals, slicers,drill-through & bookmarksPublish, scheduled refresh,RLS & sharingTableau: LODs, calculations& actionsPerformance & dashboardgovernanceAI in BI: Copilot &natural-language queriesThe AI-first analystAI-assisted analysisPrompting for analysis:context, schema, constraintsThe 2026 toolkit: ChatGPT,Claude, Copilot, GeminiVerify everything: thereview loopAI for cleaning &transformation codeDrafting insights andsummaries with AIAnalysing text with LLMs:classify, extract, tagData privacy with AI toolsSmall automations & agentswith LLM APIsLimits: hallucination,leakage, reproducibilityWhere the data livesModern data stack awarenessWarehouses: BigQuery,Snowflake, RedshiftPipelines you own: SQL,Python & schedulingELT and dbt basicsLakehouse & platforms:Databricks, Fabric, SynapseCloud analytics services &costGit & version control foranalystsWorking with data engineersMachine learning awarenessSupervised, unsupervised,reinforcementCommon algorithms atintuition levelEvaluating models honestlyWhen not to use MLGet hiredPortfolio & job searchPortfolio projects that getreadInternships & virtualinternshipsCase studies & take-hometestsSQL & analytics interviewprepResume & LinkedIn foranalyst rolesBehavioural & stakeholderstoriesCertifications: a plus, notthe filterProve it: certification &job simulationCommunity, competitions &applying smartlyPortfolio projects that get read1 · A Power BI sales dashboard on a real star schemaOlist's e-commerce tables modelled as facts and dimensionswith a date table, five DAX measures including last-year andshare-of-total, one page built for a named category head,published with scheduled refresh - and a two-minute recordedwalkthrough. A named tool, a named model and named measuresare what half of fresher postings ask for.2 · Retention teardown: who comes back, and what theyare worthMonthly cohorts from a public transactional dataset; end withthe one change you would test next and the number it wouldmove.3 · The data quality incident reportFind a dataset with real problems, quantify how much they movethe headline metric, fix what can be fixed and state whatcannot.4 · The weekly report that sends itselfA Python script that pulls from SQLite or a warehouse,computes the week's KPIs, writes an Excel pack and posts it ona schedule, with a log and a failure alert. Report automationis named in 37% of fresher and 48% of experienced postings.5 · A finance or operations deep-dive in the domain youwantVariance analysis of a listed company's quarterly P&L from itsown filings, or stock-outs and fill rate on the DataCosupply-chain dataset: one question, one reconciled number, onerecommendation.6 · Experienced: a metrics layer with a governance noteModel three source tables with dbt or plain SQL into onerevenue mart, define five metrics once with owners, and showthe before-and-after of two dashboards that used to disagree.Kaggle datasetsData Analyst job simulationWhat most learners get wrongCollecting courses instead of finishing one analysis end toend.Learning tools before learning to ask a precise businessquestion.Stopping SQL at joins; window functions and validation arewhere interviews are decided.Learning three BI tools superficially instead of one properly,with a model and measures.Ignoring the business: a sector is named in 80% of fresherpostings.A portfolio of clean tutorial data - real hiring testsjudgment under mess.Shipping a number without reconciling it against a source oftruth.Pasting AI output into a deliverable without checking thejoin, the grain or the maths.Chasing machine learning before statistics you can defend.Collecting certificates instead of an internship and twofinished projects.Certifications: proof, not the filterWhat postings actually sayA named certification appears in 0% of fresher and 2% ofexperienced Data Analyst postings; a degree, internships andprojects are what they name. A certificate is worth havingwhen it is verifiable proof of a skill the roadmap teaches -not as the door.OneRoadmap Data Analyst certification - free30 minutes, scored deterministically, and it feeds your TalentGraph so recruiters can verify it in one click.PL-300: Power BI Data Analyst AssociateThe one external certificate postings do name, for BI-heavyroles. Take it once you can model data - it tests DAX andmodelling, not chart-making.Google Data Analytics Professional CertificateFor a career switcher with no analytics job history; it forcesa portfolio out of you.Workplace proof: Tech Workplace and AI Work CompetencyCommunication is named in 79% of fresher postings and teamworkin 71%. OneRoadmap's two free workplace certifications turnthose into verified evidence too.Get certified - freeTech Workplace CompetencyPL-300 detailsKeep goingGo deeper on the tools this roadmap leans on, or step acrossto the AI and Data Scientist path once the analyst foundationsare in place.SQL roadmapPython roadmapAI & Data Scientist roadmapDaily challenges

Everything on this roadmap

Every stage and topic on the map, in reading order. Click any topic to open it.

The Data Analyst roadmap, topic by topic

Every topic on the map in reading order, with what to learn and the resources we recommend. Main topics are open; open any subtopic to read it. Select a topic name to open it in the topic panel, where you can track your progress.

  1. 1. Know the job before the tools

    The analyst job in 2026

    A data analyst turns business data into a decision somebody acts on: finds the right data, cleans it, questions it, and explains what it means in words a manager can use. In 2026 the typing is cheap - AI drafts the SQL, the pandas, the DAX and even the first paragraph - so what companies pay for is judgment: which question matters, whether the number can be trusted, and what to do next.

    This section answers the three questions every learner asks, in one sitting, each with the report behind it:

    Konsi skills seekhni hain? - which skills to learn, in what order, and the one most people skip (AI literacy).

    Kis role ke liye prepare karna hai? - analyst vs BI vs analytics engineer vs data scientist, the fresher and switcher paths, and where the jobs and pay are in India.

    Companies actually expect kya karti hain? - the four moments of a 2026 hiring process, the certifications and projects that count, and the AI workflow they assume on day one.

    The 2026 picture in five numbers. Analytical thinking is the core skill 7 in 10 employers want most, and AI & big data is the fastest-growing skill to 2030 (World Economic Forum, Future of Jobs 2025). 71% of leaders would rather hire a less experienced candidate with AI skills than a more experienced one without (Microsoft & LinkedIn Work Trend Index). Nearly 45% of data & analytics job postings now mention AI, against 4.2% of all postings (Indeed Hiring Lab, Jan 2026). Among data learners, critical thinking enrolments grew 168% in a year and data-quality skills 108% (Coursera Job Skills Report 2026). In India, fresher hiring grew 30% year on year in November 2025 and 18% in December (Naukri JobSpeak).

    A week in the job: clarify a vague request, pull data with SQL, check it against a known total, build the one chart that answers the question, write three sentences a manager can act on. Dashboards are a by-product; decisions are the product.

    Fresher: read all twelve sub-nodes before touching a tool - this is your map. Experienced: skim "AI literacy & workflow", "Hiring tests now" and "AI on day one" - they are what changed since you learned the job.

    Which skills to learn

    2026

    Four tools, one language, one habit - in that order. Companies hire for the first four, promote for the next, and in 2026 screen for the habit.

    The core four: SQL, Excel or Google Sheets, one BI tool (Power BI in India and the Gulf, Tableau or Power BI in the US), and practical statistics. Every job description names at least three of them.

    Then Python - pandas for cleaning and reshaping, not machine learning first.

    The habit: AI literacy. Using an assistant to draft, then checking the draft against the schema and the totals. LinkedIn ranks AI literacy the fastest-growing skill in the US for 2025 (Skills on the Rise); Coursera's 2026 data shows data learners racing to critical thinking (+168%) and data quality (+108%) - the checking skills (Job Skills Report 2026).

    Example: a fresher who knows SQL joins and window functions, pivots in Excel, builds one Power BI dashboard and can explain a p-value is interview-ready for most analyst roles. Adding pandas and a disciplined AI workflow is what moves the offer from ₹4 lakh to ₹7 lakh in the same city.

    Fresher: learn in this order; do not start with Python. Experienced: your gap is almost always statistics or the AI workflow, not another tool.

    SQL, Excel, one BI tool & statistics

    The four skills a 2026 analyst job description assumes, with the share of postings naming each in our October 2026 audit. Learn each to the level named here, not beyond, before moving on.

    SQL - to window functions. Named in 63% of fresher and 71% of experienced postings, required outright in 40% of fresher ones. SELECT, JOINs, GROUP BY, CTEs, then ROW_NUMBER, LAG and running totals. "Revenue per customer per month, and each customer's rank within the month" is one query, and the kind a timed test asks.

    Excel / Sheets - to pivots and Power Query. Named in 66% of fresher and 45% of experienced postings. SUMIFS, XLOOKUP, pivot tables, conditional formatting, cleaning with Power Query. Still the fastest tool for a one-off question.

    One BI tool - to a published dashboard. Power BI is named in 44% of fresher and 50% of experienced postings, Tableau in 30% and 37%. A data model, five measures, one dashboard with a slicer, published; the full topic sits in the visualise stage.

    Statistics - to the point of not fooling yourself. Named in 41% of fresher and 38% of experienced postings, required in 36% of fresher ones. Mean vs median, distributions, sampling, confidence intervals, what an A/B test can and cannot say. The interviewer asks "is this difference real?", not "define variance".

    Python, read carefully. Named in 49% of fresher postings, but required in 24% and preferred in 29%: learn it after the four above, in the next node.

    Example: an e-commerce "why did orders drop in March" case: SQL for daily orders by channel, Excel to pivot by week, a BI chart to show the one channel that fell, and statistics to say whether a 6% dip is noise. That whole cycle is the job.

    Python for analysts, not ML first

    Python is the fifth skill, not the first: it earns its place when the data no longer fits a spreadsheet or the cleaning has to be repeated every week.

    What to learn: pandas (read, filter, group, merge, reshape), basic plotting, reading from a database, and writing a script that re-runs a clean-up on next month's file.

    What to skip at first: scikit-learn, deep learning, "data science" courses. Analyst interviews ask you to clean and summarise, not to train a model.

    Where it pays: repeatable reports, joining more than three sources, text clean-up, and any dataset over a million rows.

    Example: a weekly sales file arrives with inconsistent dates and city spellings. A 40-line pandas script that standardises both and writes a clean table beats an hour of manual Excel every Monday - and it is a portfolio piece.

    Fresher: add Python after SQL and Excel are solid, roughly month three. Experienced: if you only use Excel today, pandas is the single highest-return skill on this map.

    AI literacy & the AI workflow

    AI

    The skill most learners skip, and the one hiring managers now screen for. 66% of leaders say they would not hire someone without AI skills, and 71% would rather take a less experienced candidate who has them than a more experienced one who does not (Microsoft & LinkedIn Work Trend Index, 31,000 people, 31 countries). The World Economic Forum names AI & big data the fastest-growing skill to 2030 (Future of Jobs 2025).

    AI literacy for an analyst is a four-step workflow, not a tool:

    Specify. Write the question precisely, with the table, the grain and the definition of the metric. A vague prompt gets a confident wrong query.

    Draft. Let ChatGPT, Claude, Gemini or Copilot write the SQL, the pandas, the DAX or the first paragraph.

    Check. Run it, reconcile the total against a number you already trust, look for the wrong join, the silent NULL, the double count.

    Own. Put your name on the number. The assistant is never the author.

    Example: "Monthly active users by plan" drafted by an assistant counted every login; the check against the finance dashboard was off by 30%, because the prompt never said "distinct users". Specify, draft, check, own - in that order, every time.

    Both tracks: every AI-tagged node on this map pairs a capability with the check that makes it safe.

    Which role to prepare for

    Aim at one title, in one sector, in one city - the preparation, the portfolio and the job search all get sharper. The three sub-nodes give you the map, the path and the market.

    The titles overlap but have different centres of gravity: data analyst (business questions, SQL + BI), BI analyst (reporting), analytics engineer (the transformation layer), data scientist (models and experiments).

    The path depends on where you start: a fresher builds credibility from zero; a switcher leads with the domain they already know.

    The market in India is wide and growing: fresher hiring rose 30% year on year in November 2025 (Naukri JobSpeak), and the WEF counts big data specialists among the fastest-growing jobs to 2030 (Future of Jobs 2025).

    Example: "Data analyst, fintech, Bengaluru" is a target. "Something in data" is not - it produces a portfolio nobody asked for and applications to roles that want a different person.

    Analyst vs BI vs analytics engineer vs data scientist

    Four titles, one spectrum from "answers questions" to "builds models". Knowing where each sits aims your learning and stops you applying for the wrong person's job.

    Data analyst - answers business questions with SQL, spreadsheets and BI; the broadest title and the most openings. This roadmap.

    BI analyst / BI developer - specialises in reporting: data models, measures, dashboards. Deepest in Power BI or Tableau.

    Analytics engineer - owns the transformation layer (dbt, warehouses) that analysts query. SQL-heavy, software habits, fewer stakeholder meetings.

    Data scientist - builds models and runs experiments; statistics and Python first, and most postings ask for a quantitative degree.

    Example: the same "churn" problem. The analyst finds which segment churns and why; the BI analyst builds the churn dashboard; the analytics engineer builds the clean customers table both use; the data scientist predicts who churns next.

    Fresher: start as an analyst - it has the most doors and teaches the judgment every other title needs. Experienced: moving to analytics engineering is the natural next step if you like SQL more than meetings.

    Fresher path vs switcher path

    Optional on the Experienced track

    Two starting points, one finish line: the same four interview moments. The difference is what you lead with.

    Fresher: credibility from scratch. One certification, SQL to window functions, one deep portfolio project on messy data, and the ability to explain both in two minutes. The portfolio replaces experience - one deep project beats five tutorials.

    Switcher: you already have a domain - finance, operations, marketing, healthcare, sales. Lead with it: analyse the data of the field you know, because domain sense cannot be learned in a week and is exactly what a hiring manager in that field is short of.

    Both: converge on SQL, one BI tool, statistics and the AI workflow.

    Example: an operations executive moving into analytics builds a project on warehouse dispatch data they understand, not on a Titanic dataset. The interviewer hears a colleague who now also writes SQL.

    The certification route works: Google reports that 75% of its certificate graduates saw a positive career outcome within six months (US graduate survey, 2024) - with a portfolio project beside it, not instead of one.

    Where the jobs are: sectors, cities and pay

    The India market for analysts in 2026, in figures you can plan around.

    Hiring is growing. White-collar hiring ended 2025 up 13% in December, with fresher hiring up 30% year on year in November and 18% in December (Naukri JobSpeak). Globally, the WEF expects 170 million jobs created and 92 million displaced by 2030 - a net 78 million - with big data specialists among the fastest-growing titles (Future of Jobs 2025).

    Sectors that hire analysts most: fintech and banking, e-commerce and quick commerce, consulting and the Big 4, SaaS and product companies, and the global capability centres (GCCs) of US and European firms.

    Pay, India (Glassdoor via Coursera's guide, Dec 2025): median total pay about ₹6.6 lakh; Hyderabad ₹7.8 lakh, Bengaluru ₹6.8 lakh, Gurugram ₹6.6 lakh, Mumbai ₹6.5 lakh, Chennai and Pune ₹5.5 lakh; senior data analyst median ₹15 lakh. Payscale puts 1-4 years at about ₹5.4 lakh (salary guide).

    What moves you up a band: SQL plus one BI tool plus Python, a domain, and the AI workflow - the same four things the rest of this section is about.

    Example: a fresher in Pune at ₹4.5 lakh who adds Power BI and a fintech portfolio project typically interviews a band higher within a year; the move to Bengaluru or Hyderabad adds another ₹1-2 lakh on the same skills.

    What companies actually expect

    Hiring

    Not a list of tools. In 2026 a company expects three things from an analyst it hires: that you can reason under messy data, that you have proof of it, and that you use AI the way a professional does - drafting with it and checking it.

    The tests: a timed SQL test, a case study on a messy dataset, a walkthrough of one project, and a story about a stakeholder who disagreed with you.

    The proof: one certification the recruiter recognises and one deep project, not five shallow ones.

    The AI expectation: nearly 45% of data & analytics postings now mention AI, against 4.2% of all postings (Indeed Hiring Lab, Jan 2026). They are not asking for a researcher; they are asking whether you can specify, draft, check and own.

    Example: a Big 4 analytics case round gives you a CSV of 50,000 invoices with duplicates and asks for "the top three risks". The grade is for how you found the duplicates and defined "risk", not for the chart.

    What hiring managers test now

    Screening rounds in 2026 look alike across companies. Build your preparation around these four moments, in this order.

    A timed SQL test - joins, aggregation, window functions, usually 45-60 minutes on a schema you have never seen. Example: "second-highest order value per customer, customers with no orders in 30 days".

    A take-home or live case on a messy dataset - duplicates, mixed date formats, a metric you must define yourself. The grade is for your checks and your definition, not the chart.

    A walkthrough of one portfolio project - "why this dataset, what did you check, what did you recommend, what would you do with more time".

    A stakeholder story - "tell me about a time someone disagreed with your number". They are listening for whether you reconciled it or defended it.

    Tools are rarely the filter; reasoning under mess and a clear explanation are. Analytical thinking is the single core skill 7 in 10 employers name (World Economic Forum, Future of Jobs 2025).

    Fresher: your portfolio replaces experience - make one project deep rather than five shallow. Experienced: expect the case to probe how you defined the metric and validated the result.

    Certifications and projects that count

    Recruiters scan for a degree and experience; hiring managers decide on one project. In our October 2026 audit, a bachelor's degree is named in 93% of fresher and 88% of experienced postings, a quantitative degree in 56% and 43%, internship experience in 43% of fresher postings, a portfolio or GitHub in 24% and 18%, a case study or assessment in the process in 19% and 16%, and a named certification in 0% and 2%.

    Certifications are proof, not the filter: OneRoadmap's free, scored Data Analyst certification, which turns into verified evidence on your profile here; Microsoft PL-300 for BI-heavy roles, the one external certificate postings name; the Google Data Analytics certificate for switchers with no analytics history.

    Projects that hiring managers believe: one messy public dataset in a domain you can talk about, cleaned with the checks shown, ending in a recommendation and a two-minute recorded walkthrough. Example: "Which Mumbai restaurant categories should a delivery app push in monsoon months?" on open order data - with a reconciled total, a definition of "push", and one chart.

    What does not count: a dashboard on a clean tutorial dataset, a certificate with no project, ten micro-courses and no finished analysis.

    Fresher: an internship (real, or the job simulation as a virtual one) and two projects first; the certification is the verifiable proof on top. Experienced: skip the beginner certificate; PL-300 or a published project in your domain is the signal.

    The AI workflow they expect on day one

    AI

    "AI-ready" in a 2026 job description does not mean machine learning. It means the hiring manager assumes you draft with an assistant and check what it gives you - and will notice in week one if you do neither.

    Day-one tools: Copilot in Excel and Power BI, Gemini in Google Sheets, ChatGPT or Claude for SQL and pandas drafts and for turning a table into a first paragraph.

    The expectation: specify the question with the table, grain and metric definition; let the assistant draft; reconcile the result against a number you already trust; then present it as yours.

    The figures behind it: nearly 45% of data & analytics postings mention AI (Indeed Hiring Lab, Jan 2026); 75% of knowledge workers already use AI at work (Microsoft & LinkedIn Work Trend Index); among data learners, data quality (+108%) and data cleansing (+103%) are the fastest-growing skills because checking AI output is now the job (Coursera Job Skills Report 2026).

    Example: asked for "weekly retention by cohort", you prompt with the exact table and the definition of "retained", run the draft, compare week 1 against last month's deck, find the assistant counted trial users, fix the filter, and send the number with one line on what you checked. That is the whole skill.

    What gets you hired

    Skills get you screened; proof gets you interviewed; the explanation gets you hired. The three sub-nodes are the plan, the traps, and the proof you can build here.

    A 90-day plan that ends in one certification, one deep project and twenty rehearsed answers.

    The mistakes that keep most learners at the same level for a year - every one of them avoidable.

    A proof stack on OneRoadmap: a scored certification, a job simulation, daily challenges and a readiness score employers can see.

    The honest context: 39% of today's skills change by 2030 and 70% of the skills in most jobs will have changed between 2015 and 2030 (WEF, LinkedIn). The analysts who stay hired are the ones who can learn the next tool in a month because the judgment underneath does not change.

    The 90-day plan

    Twelve weeks from zero to interview-ready, for somebody with two hours a day. A switcher compresses weeks 1-4 and spends the time on a domain project instead.

    Weeks 1-4: SQL and Excel. SQL to window functions (LeetCode Top SQL 50), Excel pivots, XLOOKUP, Power Query. Finish with the OneRoadmap Data Analyst certification.

    Weeks 5-8: one deep project. A messy public dataset in a domain you can talk about; clean it with the checks shown, define the metric, reconcile a total, end with a recommendation. Record a two-minute walkthrough.

    Weeks 9-10: one BI tool and statistics. Publish one Power BI or Tableau dashboard from the project; learn enough statistics to say whether a difference is real.

    Weeks 11-12: the AI workflow and interviews. Redo the project's SQL with an assistant and document every check you made; rehearse the four interview moments; apply to ten targeted roles, not a hundred generic ones.

    Example week 6: Monday find the duplicate orders, Tuesday decide what "active customer" means and write it down, Wednesday reconcile revenue to the published total, Thursday one chart, Friday the three-sentence recommendation.

    Mistakes that keep learners stuck

    Read first

    The pattern repeats in every cohort. Each of these has a node on this map marked "Missed" - do those first.

    Collecting courses without finishing one end-to-end analysis.

    Learning tools before learning to ask a precise question.

    Stopping SQL at joins - window functions are where the tests begin.

    A portfolio of clean tutorial datasets - nobody at work has clean data.

    Shipping a number without reconciling it against a total somebody already trusts.

    Pasting AI output into a deliverable unchecked - the fastest way to lose a manager's trust in 2026.

    Chasing machine learning before statistics.

    Never rehearsing the explanation - the interview is a conversation, not a test.

    Example: two candidates, same SQL score. One says "the revenue came to 4.2 crore"; the other says "4.2 crore, which matches finance's number within 0.3% - the gap is refunds I excluded". The second gets the offer.

    Your proof stack on OneRoadmap

    Reading is half of it. Everything this section asks you to prove can be proved here, and it becomes verified evidence on your profile that employers and the match score read.

    Data Analyst certification - free, scored, the role credential: SQL, statistics, data quality and written analyst cases in one assessment.

    Data Analyst job simulation - a realistic ticket on messy data, graded on the checks and the recommendation, not the chart.

    Daily challenges - short, scored, a streak that keeps SQL and statistics sharp between projects.

    Role Readiness - your score out of 100 for the role, built from the evidence above, with the gaps and what closes each.

    Prepare an application - when a job fits, Ori tailors the resume and answers from that evidence; you review and submit.

    Example: certification done, one simulation passed, a 20-day challenge streak and a readiness score of 72 - that is a profile a recruiter can trust before the first call, and the exact set this roadmap's stages are built around.

  2. 2. Think like an analyst

    Business questions & metrics

    Every analysis starts with a question a business actually has, translated into a metric with a precise definition and a known grain. Skipping this step is why so many dashboards exist and so few decisions change. This topic is tool-free on purpose: it is the part of the job that AI cannot do for you.

    Turning a vague ask into a question

    "Can you look at churn?" is not a question. "What share of customers who joined in Q1 made no purchase in the following 90 days, and how does that compare with Q4?" is. Practise rewriting requests until they name the population, the metric, the time window and the comparison. Then confirm the rewrite with the requester before you open a query editor - it saves days.

    Experienced: this is also how you push back on a bad request without saying no.

    The four kinds of questions

    Every request is one of four: what happened (descriptive), why did it happen (diagnostic), what will happen (predictive), or what should we do (prescriptive). Identifying which one you are being asked decides your entire approach - a descriptive question needs a clean aggregate, a diagnostic one needs segmentation and comparison, a predictive one needs a model and honesty about uncertainty. Most analyst work is the first two, done well.

    Metrics, KPIs and definitions

    A metric is a definition, not a number: numerator, denominator, time window, filters, and the exact table it comes from. "Active users" means nothing until you say active how, over what period, excluding whom. Write definitions down where the whole team can see them, and never let two dashboards report two different values for the same name.

    Data grain and uniqueness

    Missed

    Grain is what one row represents: one order, one order line, one customer per month. Almost every wrong total comes from joining or aggregating at the wrong grain - a join that fans out one order into three lines silently triples revenue. Before any analysis, state the grain of every table in one sentence and check it with a uniqueness query. This single habit prevents the most common mistake in interviews and on the job.

    The data lifecycle

    Collection, cleaning, exploration, analysis, visualisation, decision - and then measurement of whether the decision worked. Knowing where you are in the cycle keeps a project from becoming aimless chart-making, and tells you what "done" means at each step. Most learner projects skip the last two steps; most employers only care about them.

    Stakeholders, requirements and pushback

    Analysts work for people who are busy, sure of their hypothesis, and often wrong about what data exists. Learn to run a fifteen-minute requirements conversation, to write back what you heard, to say "the data cannot answer that, but it can answer this", and to deliver an inconvenient result without a fight. Interviews probe this directly with behavioural questions.

    Fresher: the job simulation on this roadmap puts you in this conversation with a realistic manager.

    Requirements gathering & the analytics spec

    HiringOptional on the Fresher track

    Requirements gathering is named in 12% of fresher and 18% of experienced postings in our October 2026 audit, the step that separates an analyst from a query writer. The fifteen-minute intake: who decides what, by when, with which number, and what they will do differently if the number is high or low. If no decision changes, the request is curiosity and can wait. The one-page analytics spec then fixes the question, the decision it serves, each metric with its definition and grain, the sources and their cut-off date, and the caveats you already know. Write it back to the requester before you open a query tool; most disagreements surface here, at the cost of an hour, not a week. Learned looks like a spec a colleague can execute without asking you, plus a change log when the question moves.

    Fresher: practise it on your portfolio projects, spec before analysis. Experienced: core; interviewers ask for a time a requirement changed under you.

    Working in a team: tickets, Agile, reviews

    Hiring

    Cross-functional teamwork is named in 71% of fresher and 72% of experienced postings in our October 2026 audit, Agile or Jira by name in 7% and 13%. Analytics work arrives as tickets: a user story ("as a sales lead I need churn by region to set quotas"), an owner, an estimate, a definition of done. Sprints of one or two weeks, with a short daily stand-up: yesterday, today, blockers. Say the blocker; nobody can remove one you kept quiet about. Asking for review is competence, not weakness: share the query, the reconciled totals and what you are unsure of, and review others the same way. Document decisions where the ticket lives, so "why did we exclude test accounts" outlives the people who decided it. Learned looks like a ticket taken from vague request to reviewed, documented delivery.

  3. 3. Learn the business you analyse

    Domain primers: the metrics each sector runs on

    Hiring

    A domain primer is the smallest amount of a business you need to know before your numbers mean anything to the people who run it. In our October 2026 audit a sector is named in 80% of fresher and 87% of experienced postings: finance and FP&A in 53% and 57%, operations and supply chain in 43% and 45%, banking, fintech and insurance in 38% and 42%, healthcare and pharma in 35% and 41%, marketing in 35% and 22%, SaaS and product in 6% and 12%.

    What a primer is: the five metrics the sector runs on, the two reports everyone reads, the vocabulary a practitioner uses without explaining it, and one public dataset you have rebuilt a report from. Learned looks like this: a finance manager says "margin slipped on mix" and you already know which table to open.

    Fresher: pick the sector of the jobs you are applying to and do one primer properly. Experienced: your current sector is your strongest card; add the adjacent one postings pair it with.

    Learn a domain in two weeks

    Two weeks is enough to stop sounding like an outsider, if you read the right things and finish with something built.

    Week one, read: one listed company's annual report (the management discussion, not the glossy front half), one industry report, and the metric definitions the sector publishes, so you can say what a number includes and excludes. Week two, do: interview one practitioner for thirty minutes about which report they open on Monday morning and what makes them distrust it, then rebuild one of those reports on public data and note where your version differs.

    In India start with IBEF for the industry picture and data.gov.in for datasets; globally, SEC EDGAR holds every listed company's filings. Learned looks like a one-page glossary in your own words and a rebuilt report you can defend.

    Finance & FP&A: P&L, variance, margin

    Finance and FP&A is the sector postings name most: 53% of fresher and 57% of experienced postings in our October 2026 audit. The vocabulary is small and precise. The P&L runs revenue, cost of goods sold, gross profit, operating expenses, EBITDA and net profit, in that order, for a period. Margin is one of those lines over revenue; gross and net margin answer different questions, and "margin fell" means nothing until you say which. Variance is actual against budget or last year, split into price, volume and mix, with a reason for every line past a threshold. Month-end close is the week the numbers lock; nothing reported before it is final.

    An FP&A analyst's dashboard shows revenue and margin against budget, the variance bridge, cash and headcount. Learned looks like rebuilding one quarter's variance bridge from a public annual report.

    E-commerce & marketing: funnel, CAC, ROAS

    Marketing and campaigns are named in 19% of fresher and 22% of experienced postings, e-commerce and retail in 13% and 14%, in our October 2026 audit; the sector most first dashboards are about. The funnel is visits, product views, add to cart, checkout, order, with a conversion rate at every step; the step that leaks is the finding. Order metrics: average order value, repeat rate and revenue share from returning customers. Spend metrics: customer acquisition cost (spend over new customers) and return on ad spend (revenue over spend), which flatter on a channel's own report because every channel claims every sale. Attribution is a model: last-click, first-click and data-driven disagree, so say which you used. The funnel usually lives in GA4. Learned looks like a funnel with a leak and CAC by channel on the Olist dataset.

    Operations & supply chain: fill rate, lead time

    Operations and supply chain is named in 43% of fresher and 45% of experienced postings in our October 2026 audit, and its numbers measure promises kept. Fill rate is the share of an order shipped from stock. OTIF (on time, in full) is the share of orders that arrived on the promised day and complete; the number customers judge you on. Lead time is order to delivery, measured as a distribution, not an average, because the tail breaks the plan. Inventory turnover is cost of goods sold over average inventory; low turnover is cash on shelves, and stock-outs are the opposite failure. Utilisation is capacity used over capacity available, for a warehouse, line or fleet. Learned looks like an OTIF dashboard by carrier on the DataCo dataset, with the late-order root cause as the headline.

    BFSI: risk, collections, NPA, fraud

    Banking, fintech and insurance is named in 38% of fresher and 42% of experienced postings in our October 2026 audit. The vocabulary is regulated, so take definitions from the RBI master circulars, not a blog. Delinquency buckets sort overdue loans by days past due; past ninety days an account is a non-performing asset. Roll rates are the share of each bucket that moves to the next in a month, the engine of every collections forecast. Collections are measured as recovery against what was due, by bucket and channel. Approval rate trades off against later defaults. In fraud, the fraud rate is tiny; the false positive rate is what the business feels, since every blocked genuine customer complains. Learned looks like a roll-rate matrix on Home Credit and a precision-recall view of the credit-card fraud dataset.

    Healthcare & pharma: claims, outcomes

    Healthcare and pharma is named in 35% of fresher and 41% of experienced postings in our October 2026 audit, the sector where a careless analyst does real harm. Claims: a bill from a provider to a payer; denial rate (claims refused, by reason code) and days to payment are the revenue-cycle metrics. Outcomes: length of stay, thirty-day readmission rate and mortality, always risk-adjusted, because a hospital that takes sicker patients looks worse raw. Pharma: adherence (the share of prescribed doses taken) and, in trials, enrolment, dropout and the primary endpoint, fixed before anyone sees the data. Compliance: patient data is PHI, so you work on de-identified extracts, never copy it to a laptop or a chat tool, and every access is logged. Learned looks like a readmission analysis on MIMIC with the de-identification steps written down.

    SaaS & product: MRR, churn, activation

    Optional on the Fresher track

    SaaS and product is named in 6% of fresher and 12% of experienced postings in our October 2026 audit, product analytics tools in 4% and 7%: small shares, but these product-company roles pay well. Revenue: monthly recurring revenue (ARR is the annual form), bridged each month into new, expansion, contraction and churned. Churn is customers or revenue lost over the opening balance; logo and revenue churn can move in opposite directions. Activation is the first moment a user gets value, defined per product, measured as a rate within a window. Retention cohorts follow each sign-up month over time. Net revenue retention above one hundred per cent means the base grows without new sales; CAC payback is the months to recover acquisition cost. Learned looks like a cohort chart and an MRR bridge that reconcile to each other.

    Fresher: vocabulary for now, unless you are applying to product companies. Experienced: core for product and SaaS roles.

  4. 4. Spreadsheets, still the fastest tool

    Excel & Google Sheets

    Spreadsheets are still the fastest way from a file a stakeholder emailed you to a first answer, and the tool every business person already trusts. Master the analytical core - lookups, pivots, Power Query - rather than every function. Then let AI write the awkward formulas while you check them.

    Fresher: do the functions box; interviews for entry roles still ask them.
    Experienced: skip straight to Power Query and the AI features.

    Core functions & references

    Optional on the Experienced track

    Absolute versus relative references, named ranges, and the everyday aggregates - SUM, AVERAGE, COUNT, MIN and MAX - that everything else builds on. Get comfortable reading a formula someone else wrote; you will inherit far more workbooks than you create.

    SUMIFS / COUNTIFS

    Optional on the Experienced track

    Conditional aggregation in a sheet: total sales for one region and one month, count of orders above a threshold. SUMIFS and COUNTIFS are the spreadsheet twins of SQL's GROUP BY with a WHERE, and the fastest way to answer a business question without a pivot.

    IF / IFS / AND / OR

    Optional on the Experienced track

    Business rules inside a sheet: flag late deliveries, bucket customers into tiers, mark exceptions. Nest sparingly - an IFS or a lookup table is usually clearer than five nested IFs, and clearer means checkable.

    Dates & DATEDIF

    Optional on the Experienced track

    Dates are numbers in disguise, which is why they break. Learn DATE, EOMONTH, DATEDIF and how to spot a date stored as text. Days between order and delivery, months since signup and fiscal periods are everyday analyst arithmetic.

    TRIM & text functions

    Optional on the Experienced track

    TRIM, SUBSTITUTE, UPPER/LOWER/PROPER, LEFT/RIGHT/MID and TEXTSPLIT standardise human-typed chaos into analysable columns. Half of "the numbers don't match" in a spreadsheet is a trailing space or a mixed-case category.

    Lookups: XLOOKUP & INDEX-MATCH

    XLOOKUP (and the legacy VLOOKUP you will meet in old workbooks) plus INDEX-MATCH are how spreadsheets join data across sheets. Learn exact match, what happens on no match, and why a lookup that returns the first duplicate silently corrupts a report. A guaranteed interview topic for entry-level roles.

    Pivot tables

    Group, aggregate and slice thousands of rows in seconds - the fastest route from raw data to a first answer. Learn rows versus columns versus values, how to add a calculated field, and how to spot when the source range silently excluded new rows.

    Charts & conditional formatting

    Optional on the Experienced track

    Quick trend, comparison and share charts directly from ranges and pivots - analysis at the speed of conversation - plus conditional formatting to make exceptions jump out of a table. Good enough for a meeting; not a substitute for a BI tool.

    Power Query: clean & combine

    Power Query (Excel and Power BI) records cleaning steps - remove blanks, split columns, unpivot, merge files - as a repeatable script, so next month's file cleans itself. It is the bridge between spreadsheet habits and proper data pipelines, and the most under-learned Excel feature in analyst interviews.

    AI inside spreadsheets: Copilot & Gemini

    AI

    Copilot in Excel and Gemini in Sheets write formulas, suggest pivots and explain existing sheets. Use them to draft, then verify: check a formula on three rows by hand, confirm a pivot's total against the raw column, and never accept a generated lookup without testing the no-match case. The productivity is real; so is the confident wrong answer.

  5. 5. SQL is non-negotiable

    SQL fundamentals

    SQL is the one skill every analyst interview tests and every analyst job uses daily. Pull your own data instead of waiting on someone else, and learn to reason about row counts before you run a query. This topic is the syntax; the next one is analysis.

    Experienced: the basics node is outlined for you - jump to NULLs and dates, which still trip people with years of experience.

    SELECT, WHERE, ORDER BY, LIMIT

    Optional on the Experienced track

    Reading exactly the rows you need: filtering with WHERE, sorting, limiting, expressions and aliases. Learn operator precedence with AND/OR and always parenthesise mixed conditions - a missing bracket is a classic silent bug.

    Joins and keys

    INNER, LEFT and anti-joins, and reasoning about row counts before you run the query. Identify the join key, check it is unique on at least one side, and predict whether rows will multiply or disappear. Most interview SQL is join logic in disguise, and most wrong revenue numbers are a fan-out.

    GROUP BY, HAVING and aggregates

    Turning transaction tables into the metrics a business runs on: COUNT, SUM, AVG, MIN, MAX, COUNT(DISTINCT), grouping by one or several columns, filtering groups with HAVING, and conditional aggregation with CASE inside SUM. Know why HAVING is not WHERE.

    NULL handling

    NULL is not zero, not empty and not equal to anything, including itself. It changes the result of comparisons, drops out of COUNT(column) but not COUNT(*), disappears from averages, and turns arithmetic into NULL. Learn IS NULL, COALESCE and NULLIF, and check every LEFT JOIN for the NULLs it creates. Still the most common source of a wrong answer.

    Dates, strings and CASE

    Truncating dates to month, extracting the year, computing intervals, casting text to dates, cleaning strings with TRIM and LOWER, and bucketing values with CASE. Dialects differ here more than anywhere - know your database's date functions rather than memorising one syntax.

    SQL for analysis

    Beyond syntax: the SQL that answers real questions and survives review. CTEs make multi-step logic readable, window functions rank and run totals, cohort queries measure retention, and validation queries prove the number before anyone sees it. This is the layer that separates juniors from mid-levels in screening rounds.

    Subqueries & CTEs

    WITH clauses that make multi-step logic readable, correlated subqueries, and knowing when each beats a join. Name every CTE for what its rows are (one row per customer per month), and you will catch grain mistakes as you write.

    Window functions

    ROW_NUMBER, RANK, LAG/LEAD, running totals and moving averages over PARTITION BY ... ORDER BY. They answer "first order per customer", "change from last month" and "top three per region" without self-joins - and they appear in almost every analyst SQL test.

    Validating a query: counts & reconciliation

    Missed

    Before a number leaves your hands: does the row count match the grain you expected, does the total reconcile with a known figure (finance's revenue, last month's report), does one spot-checked entity look right end to end, and did a filter or a NULL drop rows you meant to keep? Write these checks as queries and keep them next to the analysis. Most learners never do this; every good analyst does it every time.

    Cohorts, funnels and retention

    The three analyses product and growth teams ask for most. A cohort groups users by when they started; retention measures how many are still active n periods later; a funnel counts how many reach each step in order. All three are grain exercises - one row per user per period - built with date truncation, joins and window functions.

    EXPLAIN and query performance

    Optional on the Fresher track

    When a query takes minutes, read its plan. Learn to spot a full scan on a large table, a join without a usable key, and a filter applied after an expensive step. On warehouses, performance is also cost - know what partition pruning and column selection save.

    Experienced: expected of you; Fresher: awareness is enough for now.

    Writing SQL with AI, safely

    AI

    Give the assistant the schema (tables, keys, grain), the metric definition and the dialect, and ask for the query plus the assumptions it made. Then check the join keys, the grain of the result, the NULL handling and the date boundaries, and run a validation query against a known total. Treat generated SQL as a junior's draft: fast, useful, and yours to review.

    SQL interview practice sets

    Hiring

    Timed practice beats reading. Work the LeetCode SQL 50, then StrataScratch or DataLemur questions from real companies, and rehearse explaining your query out loud. Aim to reason about row counts before running - that is what interviewers listen for.

  6. 6. Python for analysis

    Python & pandas

    Python is how an analyst automates what spreadsheets cannot and unlocks the PyData stack: pandas for data, matplotlib and plotly for charts, notebooks for reproducible work, and small scripts that replace a monthly copy-paste ritual. In 2026 you will write much of it with an AI assistant - which makes reading and testing code the skill that matters.

    Fresher: learn just enough Python to be dangerous in a notebook, then live in pandas.
    Experienced: skip the basics; focus on cleaning, automation and AI-assisted coding.

    Python basics for data

    Optional on the Experienced track

    Variables, lists, dicts, loops, functions, files and the standard library - enough of the language to read and fix code, not to build software. R is the main alternative for statistics-heavy roles; pick one and learn it well.

    pandas: load, filter, group, merge

    DataFrames are spreadsheets with superpowers: read a CSV or a database table, filter rows, create columns, group and aggregate, merge tables and reshape wide to long. Learn the mental mapping from SQL - merge is join, groupby is GROUP BY - and you can move between the two without friction.

    Cleaning with pandas

    Detect and handle missing values, drop or flag duplicates, fix types (dates stored as text, numbers with commas), standardise categories and reshape. Write the cleaning as a function you can re-run, and print row counts before and after each step so nothing disappears silently.

    Notebooks & reproducibility

    Jupyter is the analyst's workbench: code, output and narrative in one document. Reproducibility means someone else can run it top to bottom and get the same result - fixed random seeds, pinned package versions, data loaded from a stated source, no hidden state from cells run out of order. Restart and run all before you share.

    Pulling data: files, APIs, databases

    Where analyst data actually comes from: CSV and Excel exports, REST APIs with pagination and rate limits, SQL databases through a connector, and - carefully and legally - web pages. Learn one clean pattern for each and the trade-offs: files are simple but stale, APIs are live but fragile, databases are authoritative but need access.

    Plotting: matplotlib, seaborn, plotly

    matplotlib for control, seaborn for statistical plots in one line, plotly for interactive charts you can hand to a stakeholder. Exploration plots exist to expose bad data before it fools you - a histogram and a scatter per key column is the cheapest data-quality check you will ever run.

    Automating reports with scripts

    Hiring

    Anything you do monthly by hand should become a script: pull, clean, compute, chart, export to Excel or a slide, send. Report automation is named in 37% of fresher and 48% of experienced postings in our October 2026 audit, so it is core for freshers now, not a later luxury. Parameterise by date, log what ran and how many rows it saw, and fail loudly when a source changes shape. Then schedule it: cron on Linux or a Mac, Task Scheduler on Windows, a GitHub Actions workflow on a schedule trigger, or Power Automate where the company lives in Microsoft 365. Every scheduled job needs a log and a failure alert - an email or chat message when it did not run or the row count fell - because a silent failure is worse than no automation.

    Example: the weekly KPI pack: a script queries the warehouse, writes an Excel workbook with openpyxl, posts it, and alerts you if the week's order count is zero.

    Fresher: core - project 4 in the box at the bottom of this map is exactly this. Experienced: this is where you get your Fridays back.

    AI-assisted coding: generate, review, test

    AI

    Copilot, ChatGPT and Claude write pandas faster than you can type. Use them for the first draft and for explaining unfamiliar code; then read every line, run it on a small sample with a known answer, and add an assertion for the row count and the total. The skill employers now pay for is reviewing generated code, not producing it.

  7. 7. Data quality: the real 70%

    Data quality & cleaning

    The unglamorous 70% of the job. Real data arrives with duplicates, impossible dates, silent NULLs, three spellings of the same city and a total that does not match finance. Cleaning is not a chore before analysis - it is analysis, and it is where judgment shows. Document every choice, because it changes your conclusions.

    Profiling a new dataset

    Before analysing, describe: row count, grain, column types, missing share per column, distinct values of every category, min and max of every number and date, and a few example rows. Ten minutes of profiling catches most surprises and gives you the questions to ask the data owner.

    Missing data strategies

    Detect NaNs and their disguises (empty strings, zeros, 9999, "N/A"), then decide: drop, fill, flag or model - and say which in the write-up. Missingness is often informative: customers without a phone number behave differently. Never fill silently with a mean and move on.

    Duplicates & outliers

    Deduplicate on the key that defines the grain, not on every column; then check what the duplicates were telling you. For outliers, use IQR or z-scores to find them and judgment to decide: is this point an error, or the most interesting row in the file? Removing outliers changes averages - state that you did.

    Types, formats and standardisation

    Cast text to dates and numbers, unify units and currencies, normalise categories (Delhi, delhi, New Delhi), trim whitespace, and reshape wide to long. A lookup table of canonical values that the whole team shares beats a hundred one-off replacements.

    Reconciling against the source of truth

    Missed

    Every headline number needs an anchor: revenue against finance, orders against the operational system, users against the product team's dashboard. Reconcile totals, explain differences to the rupee or the row, and write the reconciliation down. It is the difference between "my query says" and "the number is". Most learners never learn this because tutorials have no second source.

    Documentation & data dictionaries

    A data dictionary names each table's grain and each column's meaning, type, allowed values and source. A short README states where data came from, what was cleaned and what is still wrong. Interviewers read these to judge how you think; teammates read them to avoid repeating your work.

    Data governance: definitions, owners, lineage, access

    2026Optional on the Fresher track

    Data governance or lineage is named in 14% of fresher and 31% of experienced postings in our October 2026 audit, and nearly half of senior postings in the same audit; it climbs with seniority because it stops dashboards disagreeing. Definitions with owners: every metric has one written definition and one named person who may change it. Lineage: for any number you can trace the tables and transformations it came from, so "why did revenue change overnight" is answered in minutes. Master data: one customer table, one product table, and everything else joins to them. Access and retention: who may see which columns, how access is requested, and when data is deleted. Learned looks like a metric catalogue for one project, with owner, definition, source and lineage per metric, and a dbt model contract on the table it feeds.

    Fresher: know the vocabulary and write definitions for your own projects. Experienced: core; this is the material of the senior interview.

    Privacy: PII, masking and the DPDP Act

    2026

    Names, phones, emails, Aadhaar numbers and locations are personal data. Know what you may collect and keep, mask or hash it before analysis, aggregate before sharing, and never paste raw customer rows into an AI tool. India's Digital Personal Data Protection Act (2023) and the GDPR make this a legal duty, and hiring managers increasingly ask about it.

  8. 8. Statistics you can defend

    Statistics

    Say "this difference is real" and defend it. Analysts need working statistics, not proofs: describe a distribution honestly, know how sampling misleads, put uncertainty on an estimate, read a test without being fooled by noise, and design a fair experiment. Skip this stage and every dashboard you build will eventually lie to someone.

    Descriptive statistics

    The numbers that summarise a column: centre, spread, shape and position. Always look at the distribution before quoting the average - a mean with a few whales in it describes nobody.

    Mean, median, mode

    Mean for symmetric data, median when a few large values distort the picture (income, order value, response time), mode for categories. Report the one that describes a typical case, and say why.

    Range, variance, SD

    A metric's average means little until you know its spread. Range, interquartile range, variance and standard deviation tell you whether the average is representative and whether a change is large relative to normal noise.

    Distribution shape

    Skewed, bimodal, heavy-tailed or roughly normal - the shape decides which summary and which test are honest. Plot a histogram before you compute anything; normality is the silent assumption behind many tests.

    Percentiles & outliers

    P50, P90 and P99 describe experience better than averages for anything with a long tail - delivery times, page loads, ticket resolution. Percentiles also define outliers (the IQR rule) and service targets.

    Sampling & bias

    A survey of customers who answered is not a sample of customers. Learn selection bias, survivorship bias and non-response, why sample size shrinks noise but not bias, and how to state who your data actually represents. Most wrong business conclusions are sampling problems, not maths problems.

    Uncertainty & confidence intervals

    An estimate from a sample is a range, not a point. Confidence intervals put honest bounds on a conversion rate or an average, and stop a 2% "improvement" measured on 50 users from becoming a headline. Report intervals whenever the sample is small or the decision is expensive.

    Hypothesis tests & p-values

    Null hypotheses, p-values and the difference between statistically significant and practically important. Know what a p-value is not (the probability the result is due to chance), why peeking at results inflates false positives, and how to explain a test to a manager in one sentence.

    Correlation vs causation

    Measuring relationships, spotting confounders, and the honesty to say "correlated" when you cannot say "caused". Ice cream and drownings both rise in summer. Before claiming a driver, ask what else changed and whether an experiment could settle it.

    A/B tests & experiment design

    Randomisation, a pre-registered metric, a sample size you computed before starting, a fixed duration, and one comparison - that is a fair test. Learn the common failures (peeking, multiple metrics, novelty effects, segment fishing) because you will be asked to read a test someone else ran badly.

    Regression basics

    Linear regression as the workhorse of "what drives this metric?" - coefficients as effect sizes, fit as how much is explained, residuals as a sanity check. Logistic regression for yes/no outcomes. Enough to read a data scientist's output and challenge it.

    Forecasting & time series basics

    Optional on the Fresher track

    Trend, seasonality and noise; moving averages and exponential smoothing; why a forecast needs an interval and a horizon; and how to judge one against a naive baseline. Demand, revenue and capacity questions all arrive as forecasts.

    Experienced: expected in most senior analyst roles. Fresher: know the vocabulary.

  9. 9. Visualise it, then tell the story

    Dashboards & storytelling

    Ship the one chart that answers the question, and write the paragraph that tells a manager what to do about it. This topic is the craft that does not depend on the tool: the right chart for the question, KPIs defined once so two people never report two numbers, a story that leads with the answer, and a number defended in front of someone who disagrees. The tool itself - Power Query, the model, DAX, publishing - lives in the next topic, Power BI & Tableau, properly. Postings in our October 2026 audit name dashboards and reporting in 84% of fresher and 85% of experienced cases, insights and recommendations in 70% and 77%, and presenting to leadership in 65% and 69%: the chart is expected, the sentence after it is what gets you hired.

    Example: "Orders fell 6% in March; the drop is one channel, paid social, and it is outside its weekly noise band only from the 18th. Recommend pausing the new creative for two weeks."

    Choosing the right chart

    Bar for comparison, line for trend, scatter for relationship, histogram for distribution, heatmap for two-way patterns, funnel for stages, and (rarely) pie for a share of a whole - matching the chart to the question, and knowing when a table beats them all. Start every chart with a title that states the finding.

    KPI design & definitions

    Define metrics precisely - numerator, denominator, time window, filters - so two people never report two numbers for the same KPI. Put the definition on the dashboard, version it, and pair every KPI with the decision it informs. A KPI nobody acts on is a chart.

    Data storytelling & executive summaries

    Overview first, drill-down second; every view answers a question someone actually asks. In writing: lead with the answer, then the evidence, then the caveat, then the recommendation - in that order, in under 150 words. Executives read the first sentence; make it the finding.

    Presenting and defending your numbers

    Hiring

    Every interview and every review has the moment when someone asks "are you sure?". Be able to say where the data came from, what the grain is, what you excluded and why, what the number was last period, and what would change your conclusion. Rehearse a two-minute walkthrough of your best project until it is boring.

    Power BI & Tableau, properly

    Hiring

    One BI tool learned to the point of a published, refreshing, correctly totalling report is now a core skill for freshers, not an extra. In our October 2026 audit of 7,431 postings, Power BI is named in 44% of fresher and 50% of experienced postings, Tableau in 30% and 37%, and dashboards and reporting as a deliverable in 84% and 85%; put together, roughly half of fresher postings and two thirds of experienced ones name at least one BI tool.

    In build order: connect and clean with Power Query, model a star schema, write DAX measures, design visuals and slicers for one named reader, publish with scheduled refresh and row-level security, translate the same skills into Tableau, then tune performance and govern the workspace.

    Learned looks like: real e-commerce tables modelled with a date table, five measures including last year and share of total, one page for a named category head, published and refreshing, explained in a two-minute recorded walkthrough.

    Fresher: every node except the two marked optional - this is the project half of fresher postings describe. Experienced: fill in time intelligence, performance and governance - the gap between "makes dashboards" and "owns the reporting layer".

    Choose your tool: Power BI, Tableau, Looker

    Choose by the jobs you want, not by a ranking. In our October 2026 audit of 7,431 postings, Power BI is named in 44% of fresher and 50% of experienced postings, Tableau in 30% and 37%, Looker or Looker Studio in 6% and 11%. In India the gap is wider: 42% and 56% for Power BI, 27% and 41% for Tableau. So for a fresher in India the default is Power BI, learned to a published report with a model and measures; Tableau first only if the roles you are targeting name it; Looker Studio as a free practice ground and for Google-heavy companies. Learn one deeply, then translate: a star schema, filter context, a measure defined once and the right chart for the question are the same skill in every tool, only the menus differ. Three tools learned superficially do not transfer.

    Fresher: Power BI, then the Tableau node for translation. Experienced: the tool your stack runs, plus enough of the other to pass a screen.

    Connect, clean and refresh: Power Query

    Get Data is where a report starts: a database, a folder of monthly CSVs, an API. Power Query records every cleaning step - change types, remove blanks, unpivot, merge, append - as a repeatable query, so next month's file cleans itself on refresh. Learn query folding (which steps the source database runs for you), parameters for the file path and date range, and the idea behind incremental refresh: load only the new partition each night, not the whole history. Power Query by name appears in 4% of fresher and 8% of experienced postings in our October 2026 audit, but it is the cleaning engine inside the Power BI that 44% and 50% name, and the same engine as the Excel Power Query node earlier on this map. Tableau Prep is the Tableau equivalent.

    Example: twelve regional files land in one folder each month; one folder query appends them, fixes the three columns the regions name differently, and refreshes without you.

    Data modelling: star schema & relationships

    Facts (orders, events) in the middle, dimensions (customers, products, dates) around them, one-to-many relationships filtering from dimension to fact, and a date table. Get this wrong and every total in the report double counts or a slicer filters nothing. Postings name data modelling or a star schema in 12% of fresher and 28% of experienced cases in our October 2026 audit, but a fresher needs it anyway: every dashboard bug is a model bug, the Power BI named in 44% of fresher postings cannot total correctly without one, and PL-300 tests modelling more than charts. Learn the grain of each fact, surrogate keys, why a many-to-many relationship is a warning sign, and when to flatten a snowflake into one dimension.

    Example: orders, order items and products: model items as the fact at line grain, and revenue by category works without a single workaround.

    Fresher: core - build the model before the first visual. Experienced: core - most of the dashboard bugs you inherit are here.

    Measures & DAX basics

    Measures compute at query time in the report's filter context: total sales, sales last year, share of total, a running total. Learn CALCULATE, what a filter context is, how a slicer and a row of a matrix change it, and why a calculated column is usually the wrong answer. DAX by name appears in 3% of fresher and 7% of experienced postings in our October 2026 audit, but Power BI appears in 44% and 50%, and no Power BI report totals correctly without measures - which is why PL-300 and interview whiteboards ask for them. Learned looks like five measures you can explain, including one that uses CALCULATE with a filter. Tableau's LOD expressions and Looker's measures cover the same ground.

    Example: share of total is DIVIDE([Sales], CALCULATE([Sales], ALL(Product[Category]))) - say in words why ALL is there.

    Fresher: core - the first five measures. Experienced: core, then time intelligence and patterns in the next node.

    Time intelligence & DAX patterns

    Optional on the Fresher track

    Almost every business question has a time comparison in it: this month against last, year to date against the same period last year, a rolling twelve months. In DAX that needs a proper date table - one row per day, marked as the date table, related to every fact - and then TOTALYTD, SAMEPERIODLASTYEAR and DATEADD do the heavy lifting. The pitfalls are all filter context: a measure that works in a month visual breaks at year level, a slicer on the fact's own date column instead of the date table, a running total that ignores the current filter. DAX Patterns catalogues the standard solutions; copy the pattern, then explain in words why it works. Build order: date table, year to date, same period last year, growth percentage, rolling average.

    Fresher: optional - a last-year and a year-to-date measure are enough for a first portfolio. Experienced: core - finance and operations reports are mostly time comparisons, and interviewers ask you to write one on a whiteboard.

    Visuals, slicers, drill-through & bookmarks

    The chart rules from the storytelling topic apply inside the tool, with the tool's own habits added: a card for the headline KPI, a bar for comparison, a line for trend, a matrix when the reader needs the numbers, and raw rows nowhere near the first page. Slicers filter the page; sync them across pages so a region picked once stays picked. Drill-through takes a reader from a summary bar to the detail page for that one product, filter carried along. Bookmarks capture a state - a filter set, a hidden visual - for a guided walk through the story, and tooltips answer the next question without a click. Design one page for one named stakeholder: what she decides on Monday goes top left, in no more than six visuals.

    Example: a category head sees revenue against last year, the three categories falling, and drills through to the SKUs behind one of them.

    Publish, scheduled refresh, RLS & sharing

    A report on your laptop is a draft. Publish it to a workspace in the Power BI service, set scheduled refresh (through a gateway if the source sits on a company server), and share it as an app so readers get a clean view rather than the editing surface. Row-level security restricts each manager to their own region with one DAX filter on the dimension table, tested with View As before anyone sees it. This is the mechanics behind report automation, which postings name in 37% of fresher and 48% of experienced cases in our October 2026 audit: a report that refreshes itself at six in the morning replaces a person pasting numbers into slides. Learned looks like a published report, a clean refresh history, two tested RLS roles, and a reader who has never seen the editor.

    Example: one sales report, twelve regional managers, each seeing only their own rows.

    Tableau: LODs, calculations & actions

    Tableau is named in 30% of fresher and 37% of experienced postings in our October 2026 audit, so once Power BI is solid, learn enough Tableau to translate. The concepts carry: a data model, measures, filters, one page per reader. What is new is the vocabulary. Calculated fields do what measures do; table calculations (running total, percent of total, rank) run over the marks on the view; level-of-detail expressions - FIXED, INCLUDE, EXCLUDE - compute at a grain different from the view, which is how orders per customer lands on a chart of regions. Dashboard actions make a click on one sheet filter or highlight another. Publish two dashboards to Tableau Public, a portfolio a recruiter can open without an account. Sit the Desktop Specialist exam only if the roles you want name Tableau.

    Example: the same e-commerce dashboard rebuilt in Tableau, with one FIXED LOD for a customer's lifetime orders.

    Performance & dashboard governance

    Optional on the Fresher track

    A dashboard nobody waits for gets used; one that takes twenty seconds dies. Performance in Power BI is mostly the model: import only the columns you use, prefer a star schema to one wide table, keep high-cardinality text out of the fact, and write measures rather than calculated columns. Performance Analyzer shows which visual fires which query and how long each takes, and fewer visuals per page cuts queries more than any DAX trick. Governance is the other half: one certified semantic model that several reports share instead of each report carrying its own copy of the data, named owners, and a development, test and production path with deployment pipelines so a change is checked before readers see it. Learned looks like a page under three seconds and a model other analysts build on.

    Fresher: optional - know what Performance Analyzer shows and why a shared dataset beats copies. Experienced: core - owning the reporting layer is the step from analyst to senior, and the metrics-layer node in the leadership stage builds on it.

    AI in BI: Copilot & natural-language queries

    AI

    Copilot in Power BI, Tableau Pulse and similar features let users ask questions in plain language and get charts, DAX and summaries. They are only as good as the model and the metric definitions underneath - which makes your modelling work more important, not less. Learn what they do well (drafting, explaining) and where they mislead (ambiguous metrics, wrong grain).

  10. 10. The AI-first analyst

    AI-assisted analysis

    AI

    The 2026 analyst works with an assistant open: it drafts SQL and pandas, explains unfamiliar code, summarises a table, classifies free text and writes the first version of the insight. The analyst's job is to give it the right context, verify everything it returns, and take responsibility for the result. This stage is the difference between an analyst AI replaces and one AI multiplies.

    Prompting for analysis: context, schema, constraints

    A good analysis prompt states the goal, the tables with their keys and grain, the metric definition, the dialect or library, the output format, and what to do when unsure. Ask for assumptions to be listed. Iterate: paste the error or the wrong total back and ask for the fix. Vague prompts get confident, wrong answers.

    The 2026 toolkit: ChatGPT, Claude, Copilot, Gemini

    ChatGPT's data analysis mode runs Python on an uploaded file; Claude reasons well over long schemas and documents; GitHub Copilot lives in your editor and notebook; Copilot in Excel and Power BI and Gemini in Sheets sit inside the tools you already use. Learn one general assistant well and the in-tool features for your stack, and know what each is allowed to see.

    Verify everything: the review loop

    Missed

    Generated SQL, code and text are drafts. For every AI output: check the join keys and grain, run it on a sample with a known answer, reconcile the total, read the maths in any summary, and confirm every quoted number appears in the data. Keep the checks as code next to the analysis. Most people skip this; it is the single habit that lets you use AI aggressively without getting burned.

    AI for cleaning & transformation code

    Describe the mess - three date formats, categories with typos, a wide table that should be long - and let the assistant write the pandas, Power Query or SQL. Then test on the ugliest rows you can find, print row counts before and after, and read the regular expressions it produced. Cleaning code is where generated code most often looks right and is wrong.

    Drafting insights and summaries with AI

    Give the model the final table and the audience, ask for a three-sentence summary that leads with the finding, and then edit: remove any claim the data does not support, add the caveat, name the action. AI is good at fluent prose and bad at knowing what matters to this manager - that part stays yours.

    Analysing text with LLMs: classify, extract, tag

    Support tickets, reviews, survey comments and call notes were unanalysable at scale until recently. Now an LLM can classify sentiment, extract the product mentioned, tag the reason for churn, and return structured rows you can count. Design the label set first, sample and hand-check a hundred results for accuracy, and report that accuracy with the numbers.

    Data privacy with AI tools

    Never paste raw customer rows, credentials or unreleased financials into a consumer AI tool. Know your company's approved tools and their data-retention terms, aggregate or mask before you share, and prefer enterprise deployments that do not train on your data. A privacy incident ends careers faster than a wrong number.

    Small automations & agents with LLM APIs

    Optional on the Fresher track

    With a few dozen lines of Python and an LLM API you can classify incoming tickets nightly, draft a weekly metrics note, or answer common data questions from a schema. Start with one narrow task, log every input and output, keep a human in the loop for anything that leaves the team, and measure accuracy before you trust it.

    Experienced: this is fast becoming the analyst's edge. Fresher: awareness now, build later.

    Limits: hallucination, leakage, reproducibility

    Models invent plausible numbers, functions and citations; they may have seen the answer to a public dataset; and the same prompt can return different code tomorrow. Treat outputs as untrusted, pin versions and seeds where you can, save the exact prompt and result with the analysis, and never report a figure you cannot reproduce from the data yourself.

  11. 11. Where the data lives

    Modern data stack awareness

    Optional on the Fresher track

    Analysts increasingly query a cloud warehouse or lakehouse fed by pipelines, modelled with dbt, everything in Git. In our October 2026 audit a cloud platform is named in 14% of fresher and 28% of experienced postings, ETL or data pipelines in 16% and 29%, a data warehouse by name in 6% and 13%. You need not build the platform, but you must speak it: where a table comes from, why it refreshed late, what a query costs, and how to ask an engineer for a change. The topic also covers what analysts do build: the small pipelines you own (a SQL model, a Python extract, a schedule, a log, an alert) and the lakehouse platforms (Databricks, Fabric, Synapse) that postings now name alongside the warehouses. Learned looks like whiteboarding your company's data flow from source to dashboard.

    Experienced: core - most mid-level roles expect it. Fresher: know the words; Git is the one to actually learn.

    Warehouses: BigQuery, Snowflake, Redshift

    Optional on the Fresher track

    Columnar, cloud-scale databases built for analytics. Learn what makes them different from a transactional database (columns, partitions, separation of storage and compute), why SELECT * costs money, and how to use partition filters. Most analyst SQL in 2026 runs on one of these.

    Pipelines you own: SQL, Python & scheduling

    HiringOptional on the Fresher track

    ETL or data pipelines are named in 16% of fresher and 29% of experienced postings in our October 2026 audit. Most of that is not the platform team's work but the small pipelines analysts own. The shape: a SQL model that builds a clean table, a Python extract from a file or an API, a nightly schedule, a log of rows in and rows out, and an alert when the count is zero or the job fails. Without the log and alert it is a script, not a pipeline. Scheduling starts with cron or Windows Task Scheduler; Airflow and Prefect at awareness level: read a DAG, know why a task retries. dbt is the step up: models, tests and documentation in one repository. Learned looks like a daily report that ran unattended for a month, failures logged.

    Fresher: one scheduled script with a log is enough. Experienced: core; be ready to describe one pipeline you own end to end.

    ELT and dbt basics

    Optional on the Fresher track

    Extract-load-transform: raw data lands in the warehouse first and is transformed there with SQL models. dbt organises those models, tests them and documents them. Understanding it tells you which table to trust, and writing a small dbt model is the natural next step for an analyst who owns metric definitions.

    Lakehouse & platforms: Databricks, Fabric, Synapse

    2026Optional on the Fresher track

    A lakehouse keeps data as open files in cloud storage and puts a warehouse-style SQL engine, a catalogue and transactions on top, so one copy serves analysts, engineers and models. In our October 2026 audit any cloud platform is named in 14% of fresher and 28% of experienced postings: Databricks in 5% and 9%, Azure in 5% and 9%, Snowflake in 6% and 12%, BigQuery in 2% and 7%, Microsoft Fabric in 2% and 4%. Databricks is the lakehouse by name, with a free edition for SQL and notebooks. Microsoft Fabric bundles a lakehouse, pipelines and Power BI in one workspace; Synapse is its predecessor, still common in enterprises. Snowflake and BigQuery are warehouses with free sandboxes. Learned looks like one dataset queried in Databricks Free Edition and again in a BigQuery sandbox, with notes on what differed.

    Fresher: know what each name is; one free sandbox is plenty. Experienced: core where the posting names the platform; the SQL transfers, the catalogue and permissions do not.

    Cloud analytics services & cost

    Optional on the Fresher track

    Cloud of any kind is named in 14% of fresher and 28% of experienced postings in our October 2026 audit, Azure by name in 5% and 9%. An analyst meets a short list of services, not the whole cloud. Azure: Synapse or Fabric for the warehouse, Power BI on top. AWS: S3 for files, Redshift or Athena for SQL. GCP: BigQuery for SQL, Cloud Storage for files, Looker Studio on top. IAM and access: you get a role, not a password; request access properly, naming the dataset, columns, purpose and duration. Query cost: a scheduled query scanning a terabyte a day is a monthly bill; partition filters, selecting only needed columns and knowing the price per terabyte earn analysts the freedom to run what they like. Learned looks like naming the three most expensive queries on your platform's cost dashboard.

    Fresher: one free sandbox and the vocabulary. Experienced: core where a cloud is named in the posting.

    Git & version control for analysts

    Commit your SQL, notebooks and dbt models; branch for a change; open a pull request so someone reviews the metric definition before it ships. Git is how analysis stops being a file called final_v3_REAL.xlsx, and it is expected in most teams now.

    Working with data engineers

    Optional on the Fresher track

    Write a good ticket: the table, the column, the grain, the example rows that look wrong, the business impact and a deadline. Learn the pipeline's refresh schedule and failure alerts. Engineers help most the people who bring specifics.

    Machine learning awareness

    Optional on the Fresher track

    Speak the language of the team next door. An analyst should recognise which business problems map to which model types, read an evaluation honestly, and know when a simpler method beats a model. You are not training models here - you are learning enough to ask the embarrassing question.

    Fresher: outlined on purpose - finish statistics first.

    Supervised, unsupervised, reinforcement

    Optional on the Fresher track

    Supervised learning predicts a label from examples (churn, fraud, price); unsupervised finds structure without labels (segments, anomalies); reinforcement learns from rewards (rare in analytics). Which business problems map to which, and where the analyst hands off to a data scientist.

    Common algorithms at intuition level

    Optional on the Fresher track

    Linear and logistic regression, decision trees and random forests, k-nearest neighbours, naive Bayes and k-means - the recognisable cast of everyday ML. Know what each is for and what it assumes; leave the maths to the people who tune them.

    Evaluating models honestly

    Optional on the Fresher track

    Train/test splits, why accuracy lies on imbalanced data, precision versus recall, and leakage - when a feature secretly contains the answer. Enough to ask "how was this validated, and on what period?" before a model's output reaches a dashboard.

    When not to use ML

    Missed

    Most business questions are answered by a good GROUP BY, a chart and a conversation. A model costs time, needs maintenance, and hides its reasoning. Reach for it when the pattern is genuinely complex, the data is plentiful, and the decision repeats. Saying "we do not need a model for this" is a senior skill.

    Big data & Spark, concept level

    Not on the Fresher track · Optional on the Experienced track

    When data outgrows one machine: distributed storage and processing with Spark, and the Hadoop lineage it replaced. Warehouses hide most of this from analysts now; know the concepts, not the cluster tuning.

    Deep learning (optional)

    Not on the Fresher track · Optional on the Experienced track

    Neural networks, CNNs and transformers with PyTorch or TensorFlow - optional territory for analysts, useful vocabulary in AI-heavy teams. The LLMs on this roadmap are deep learning; you use them without training them.

  12. 12. Lead the analytics

    Senior analyst craft

    HiringNot on the Fresher track

    After about two years the job changes shape: less of the week is writing queries and more is deciding which queries are worth writing, whose numbers to trust, and who else should be doing the work. The market is mostly here. Of the experienced postings in our October 2026 audit, 45% ask for two to three years, 44% for four to six, and 18% for seven or more. Stakeholder management is named in 71% of experienced postings, leading or owning a workstream in 35%, mentoring or leading analysts in 23%, and business acumen in 11%. The craft has six parts: intake and prioritisation, owning the metric definitions, reviewing other people's work, mentoring, running a workstream, and measuring your own impact. Learned looks like a team that still ships correct numbers the week you are on leave.

    Experienced: this whole topic is for you. Fresher: skip it until the first promotion conversation.

    Analyst to senior to lead: what changes

    Not on the Fresher track

    Postings describe three bands; in our October 2026 audit the experienced market splits 45% at two to three years, 41% at four to six and 14% at seven or more. Analyst, to two years: correct SQL, one BI tool, clean deliverables to someone else's spec. Senior, two to six: postings add data modelling (named in 28% of experienced postings against 12% of fresher), pipelines you own (37% against 16%) and governance (36% against 14%); you write the spec and own the definitions. Lead, six and up: mentoring or leading analysts, named in 23% of experienced postings, plus the team's plan, hiring and the business relationship. What to build: at two years, one end-to-end project you can defend; at five, a metrics layer a team uses; at eight, analysts you grew now running their own.

    Intake & prioritisation: saying no with data

    Not on the Fresher track

    A senior analyst's scarcest resource is the team's week, and every request arrives marked urgent. The request queue: one board or form every ask goes through; nothing that bypasses it gets done. The one-line intake form: what decision this informs, by when, and what happens if it is not done. Most requests fail the first field. Impact versus effort: score each request on the money or risk behind the decision and the days it costs, work the high-impact, low-effort corner first, and show the grid to whoever asks why theirs waits. The weekly triage: thirty minutes with the requesters, grid on screen, trade-offs made in the open. Declining well: say no to the request, not the person; offer the cheaper answer, an existing dashboard or a date. Learned looks like a queue stakeholders trust more than their own escalations.

    Own the metrics layer & semantic model

    Not on the Fresher track

    KPIs and metric definitions are named in 49% of experienced postings in our October 2026 audit and data modelling in 28%; the metrics layer is where the two meet. One definition per metric, written once with its grain, filters and owner, computed in one place every dashboard reads instead of re-derived in each report. The semantic model is that place: a Power BI dataset, dbt metrics or the dbt Semantic Layer, exposing revenue, active customers and churn as named measures. Certified datasets mark the blessed tables, so a new analyst never builds on a scratch copy. The before-and-after: finance's and sales' dashboards showed different revenue because one counted cancelled orders; after the metrics layer there is one revenue measure and the argument is over. Learned looks like two dashboards that used to disagree now reading the same measure.

    Reviewing SQL, models and dashboards

    Not on the Fresher track

    Review is how a team ships correct numbers faster than one person can check. The checklist, for every query, model and dashboard before a stakeholder sees it: grain (one row is one what; does the output respect it), joins (keys unique on the one side, no fan-out), NULLs (their source, and how each aggregate treats them), reconciliation (the total ties to a source of truth, visibly), definitions (each metric matches the catalogue), and naming (a reader knows what a column holds without opening the code). How to comment: Google's engineering review practices are the pattern. Comment on the work, not the person; mark what blocks and what is optional; explain why; approve when it is better than before, not perfect. Learned looks like a review you gave that caught a fan-out before finance did, and one you took without defending.

    Mentoring & leading a workstream

    Not on the Fresher track

    Mentoring or leading analysts is named in 23% of experienced postings in our October 2026 audit, leading or owning a workstream in 35%. Neither is management; both are the senior job. Pairing: an hour a week beside a junior on their query, and let them drive. Delegating with a spec: hand over an analysis as a one-page spec (question, decision, definitions, sources, deadline) and review the result against the spec, not against your own approach. The weekly analytics review: thirty minutes of work in progress, one question each; the standard is set by what gets praised. Hiring loops: write the take-home, score it against a rubric agreed before reading any answer, and debrief in writing. Learned looks like a mentee presenting to leadership without you in the room, and a workstream that ran on your spec in your absence.

    Measuring the impact of your analytics

    Not on the Fresher track

    Insights and recommendations are named in 77% of experienced postings in our October 2026 audit, and a senior is judged on whether those recommendations changed anything. Keep an impact log from day one. Decisions influenced: the analysis, the decision that followed, who made it and the date, a sentence each. Money moved: revenue gained, cost avoided or loss prevented, with the stakeholder's estimate and your own, and the method written down. Time saved: hours per week before and after an automation, times the people who used to run it. Adoption: weekly active viewers per dashboard, from the BI tool's usage metrics; an unopened dashboard is zero impact however correct. The resume bullet is the log compressed: action, metric moved, size of the move, period. Learned looks like an impact log you could hand to a hiring manager tomorrow.

  13. 13. Get hired

    Portfolio & job search

    Hiring

    Turn analyses into interviews. Two or three end-to-end projects on messy real data beat ten tutorial notebooks; a rehearsed walkthrough beats a long resume; verified proof beats a claim. This stage is about presenting what you can do to the people who decide.

    Fresher: your portfolio is your experience. Experienced: lead with impact and domain, and refresh the SQL test muscle.

    Portfolio projects that get read

    Pick messy, public, real data: one business question, cleaned and validated, ending in a recommendation and the number it would move, published with a README on the cleaning decisions and a two-minute recorded walkthrough. The six projects in the box at the bottom of this map are built for this, and each gets read for one reason. The Power BI sales dashboard on a real star schema names a tool, a model and five measures, which is what half of fresher postings ask for. The retention teardown ends with the one change you would test next and the number it moves. The data quality incident report quantifies how much the problems shift the headline metric. The weekly report that sends itself shows a log and a failure alert, not just a script. The finance or operations deep-dive sits in the domain the posting names, with one reconciled number. The experienced metrics layer with a governance note shows two dashboards that used to disagree and now do not.

    Fresher: projects 1, 2 and 4, then one domain deep-dive. Experienced: projects 5 and 6, in your own domain.

    Internships & virtual internships

    2026Not on the Experienced track

    Fresher postings name internship experience more often than anything except a degree: 43% of fresher postings in our October 2026 audit mention it, against 24% for a portfolio and about 0% for a named certification. And 92 of the 299 fresher postings analysed were internship roles themselves, so the internship is both the entry ticket and the first job. Routes in: the AICTE internship portal and Internshala for listed openings, Unstop for competitions that turn into offers, your campus careers cell, and a self-made one - a free analysis for a local business (a clinic's no-shows, a shop's slow stock) with a written recommendation, which counts because you can describe it in an interview. The OneRoadmap Data Analyst job simulation is built as a virtual internship: a realistic ticket on messy data, graded on the checks and the recommendation, with a result a recruiter can verify.

    Fresher: one real or virtual internship before you apply in volume. Experienced: skip - your work history replaces this.

    Case studies & take-home tests

    You get a dataset and a vague question and 48 hours. Winners restate the question precisely, profile and validate the data first, answer with two or three charts and a recommendation, list what they would do with more time, and keep the whole thing short. Losers build twelve charts and never say what to do.

    SQL & analytics interview prep

    Expect a live SQL round (joins, aggregation, window functions, a NULL trap), a metrics question ("how would you measure engagement?"), a probability or statistics question, and "tell me about a time your analysis was wrong". Practise all four out loud; the reasoning you narrate is what is scored.

    Resume & LinkedIn for analyst roles

    Quantified bullets (what you analysed, what changed, by how much), tools named where they were used, a link to the portfolio, and nothing that a screening question would expose as inflated. Match the language of the job description without keyword stuffing; applicant tracking systems rank on it. One page for freshers.

    Behavioural & stakeholder stories

    Prepare five short stories in situation-action-result form: a stakeholder who disagreed, a number you got wrong and fixed, a vague request you sharpened, a deadline you protected, a finding nobody wanted to hear. Interviewers are checking judgment and communication - the parts of the job tools cannot do.

    Certifications: a plus, not the filter

    Read the postings, not the course adverts. In our October 2026 audit a named certification (Google, PL-300, Tableau, a cloud badge) appears in 0% of fresher and 2% of experienced postings; the word certification in any form appears in 18% and 18%. What postings name is a bachelor's degree, internships and projects. So a certificate is verifiable proof of a skill you already have, not the door. The ones that earn their time: the OneRoadmap Data Analyst certification, free and scored, becomes verified evidence on your profile; PL-300 is the one external certificate postings do name, for BI-heavy roles, once you can model data; the Google Data Analytics certificate suits switchers with no analytics history; Tableau Desktop Specialist only if your roles name Tableau. Add workplace proof: communication is named in 79% of fresher and 78% of experienced postings, cross-functional teamwork in 71% and 72%, and OneRoadmap's two free workplace certifications turn those into verified evidence too.

    Fresher: the OneRoadmap certification and the two workplace certificates; PL-300 after your first published model. Experienced: PL-300 if your roles are BI-heavy; otherwise your published work is the certificate.

    Prove it: certification & job simulation

    Fresher postings name internships (43%) and a portfolio (24%) far more than a named certificate (0%) in our October 2026 audit, so prove the things they name. The OneRoadmap Data Analyst job simulation is built as a virtual internship: a realistic ticket on messy data, a manager with a hypothesis, and a grade on the checks and the recommendation rather than the chart, with a result a recruiter can verify in one click. The free Data Analyst certification is the proof of the skill itself - SQL, statistics, data quality and written analyst cases, scored and verifiable. Both feed the Talent Graph that ranks you for employers on this platform. Add the two workplace certifications, Tech Workplace Competency and AI Work Competency: communication is named in 79% of fresher and 78% of experienced postings, teamwork in 71% and 72%, and without evidence those are words on a resume. Claims are cheap; this is not.

    Fresher: simulation first, then the certification, then daily challenges to keep the streak. Experienced: the certification and one simulation are enough; your work history carries the rest.

    Switching in: leverage your domain

    Not on the Fresher track

    If you come from finance, operations, marketing, sales or healthcare, you already have what freshers lack: you know which questions matter and what the numbers should look like. Build your portfolio on your own field's data, apply to analyst roles in that field first, and frame your experience as analysis you were already doing without the title.

    Community, competitions & applying smartly

    Optional on the Experienced track

    Learning in public compounds: share a project write-up, enter a Kaggle competition for the feedback, and join analytics communities where hiring managers actually post. Apply in focused batches to roles that match your portfolio, and follow up with a specific comment about the company's data - it works far more often than volume.

Prove it as you go

Reading is half of it. This roadmap is backed by a free, scored Data Analyst certification, a realistic job simulation and daily challenges - evidence a recruiter can verify, not a claim. Every topic is checked against live job postings; open the market audit at the top of the page.

Portfolio projects that get read

1 · A Power BI sales dashboard on a real star schema

Olist's e-commerce tables modelled as facts and dimensions with a date table, five DAX measures including last-year and share-of-total, one page built for a named category head, published with scheduled refresh - and a two-minute recorded walkthrough. A named tool, a named model and named measures are what half of fresher postings ask for.

2 · Retention teardown: who comes back, and what they are worth

Monthly cohorts from a public transactional dataset; end with the one change you would test next and the number it would move.

3 · The data quality incident report

Find a dataset with real problems, quantify how much they move the headline metric, fix what can be fixed and state what cannot.

4 · The weekly report that sends itself

A Python script that pulls from SQLite or a warehouse, computes the week's KPIs, writes an Excel pack and posts it on a schedule, with a log and a failure alert. Report automation is named in 37% of fresher and 48% of experienced postings.

5 · A finance or operations deep-dive in the domain you want

Variance analysis of a listed company's quarterly P&L from its own filings, or stock-outs and fill rate on the DataCo supply-chain dataset: one question, one reconciled number, one recommendation.

6 · Experienced: a metrics layer with a governance note

Model three source tables with dbt or plain SQL into one revenue mart, define five metrics once with owners, and show the before-and-after of two dashboards that used to disagree.

What most learners get wrong

Collecting courses instead of finishing one analysis end to end.

Learning tools before learning to ask a precise business question.

Stopping SQL at joins; window functions and validation are where interviews are decided.

Learning three BI tools superficially instead of one properly, with a model and measures.

Ignoring the business: a sector is named in 80% of fresher postings.

A portfolio of clean tutorial data - real hiring tests judgment under mess.

Shipping a number without reconciling it against a source of truth.

Pasting AI output into a deliverable without checking the join, the grain or the maths.

Chasing machine learning before statistics you can defend.

Collecting certificates instead of an internship and two finished projects.

Certifications: proof, not the filter

What postings actually say

A named certification appears in 0% of fresher and 2% of experienced Data Analyst postings; a degree, internships and projects are what they name. A certificate is worth having when it is verifiable proof of a skill the roadmap teaches - not as the door.

OneRoadmap Data Analyst certification - free

30 minutes, scored deterministically, and it feeds your Talent Graph so recruiters can verify it in one click.

PL-300: Power BI Data Analyst Associate

The one external certificate postings do name, for BI-heavy roles. Take it once you can model data - it tests DAX and modelling, not chart-making.

Google Data Analytics Professional Certificate

For a career switcher with no analytics job history; it forces a portfolio out of you.

Workplace proof: Tech Workplace and AI Work Competency

Communication is named in 79% of fresher postings and teamwork in 71%. OneRoadmap's two free workplace certifications turn those into verified evidence too.

Keep going

Go deeper on the tools this roadmap leans on, or step across to the AI and Data Scientist path once the analyst foundations are in place.

Frequently asked questions

Is data analysis a good career choice in 2026?

Yes. Every company produces data and needs someone to turn it into decisions, so analyst roles exist in every industry - finance, e-commerce, healthcare, SaaS, consulting, government. AI has not removed the job; it has raised the bar. Writing a query is cheap now, so the value has moved to defining the right metric, checking that a number is true and explaining what to do about it. Analysts who work that way are in more demand, not less, and the role leads naturally into analytics engineering, product analytics or data science.

How was this roadmap checked against real job postings?

Every analyst posting on the OneRoadmap job board - company career sites, applicant-tracking boards, and the LinkedIn and Naukri sweeps that feed it, 7,431 postings from 3,792 companies at the last run - is read by a script that bands it fresher or experienced and counts which tools, methods, workplace skills, domains and requirements the description names. The roadmap was then changed where the postings disagreed with it: Power BI and Tableau became a full topic at core for freshers, domain primers and data governance were added, and certifications were reframed as proof rather than the filter. Open the market audit at the top of this page for the figures, the companies and the date of the last run.

Are data analysts well paid?

Reasonably, and pay rises fast with SQL depth, BI modelling and domain knowledge. In India, freshers at product and consulting firms typically start around ₹4-8 LPA (service companies pay less), analysts with three to five years earn roughly ₹10-20 LPA, and senior or lead analysts more. In the US the common range is about $65k-$110k. Treat these as ranges to verify on current job boards for your city, not promises - the fastest way to move up the range is a portfolio that shows judgment, not a list of tools.

What qualifications do I need to become a data analyst?

Most postings name a bachelor's degree - 93% of fresher postings in our audit, and 56% a quantitative one (statistics, mathematics, economics, computer science). Beyond that, employers hire on demonstrated skills: SQL you can defend in a live test, one BI tool learned properly, working statistics, and a portfolio of two or three projects on messy real data with a written recommendation. A named certification appears in 0% of fresher postings, so treat certificates as verifiable proof of a skill - the free OneRoadmap Data Analyst certification, Microsoft's PL-300 for Power BI roles, Google's certificate for switchers - not as the door itself. Internships and projects are what postings actually name.

Can I teach myself data analysis, and how long does it take?

Yes - most working analysts did. Following this roadmap for about two hours a day, a fresher can be interview-ready in four to six months; a career switcher who already knows a domain is usually faster. What slows people down is collecting courses instead of finishing one analysis end to end. Pick one dataset, ask one business question, clean, validate, answer and present it, then repeat. Three of those beat any number of tutorials.

Do I need to code? Is Python required?

SQL is non-negotiable - it is named in 63% of fresher and 71% of experienced postings and every analyst interview tests it. Python is named in 49% of fresher postings, but read the wording: it is required in 24% and preferred in 29%. So learn SQL and Excel first, then Python, and AI assistants have made Python far cheaper to learn because they write the first draft while you learn to read and check it. Some companies still hire on Excel plus a BI tool alone, but that is a shrinking door.

Is data analysis an IT job?

Not in the traditional sense. Analysts sit between the business and the data: they usually report into an analytics, product, finance or operations team rather than IT, and their output is a decision, not a system. The work is technical - SQL, Python, BI tools, statistics - but it is judged by whether the business acted on the answer, which is why communication is on this roadmap next to the tools, and why there is now a stage on the business domains postings name most.

How is a data analyst different from a data engineer, a business analyst or a data scientist?

A data engineer builds and runs the pipelines and warehouses the analyst queries. A business analyst focuses on requirements and process, often with lighter data skills. A data scientist builds models and runs experiments. A data analyst answers business questions from existing data with SQL, spreadsheets, statistics and BI, and increasingly with AI assistants - and hands off to the others where the problem needs a pipeline or a model.

Will AI replace data analysts?

AI replaces the typing, not the judgment. Assistants already draft SQL, pandas code, DAX and summaries, so the parts of the job that were mostly mechanics are worth less. The parts that remain - specifying the question, knowing the grain of the data, catching a wrong join or a silent NULL, reconciling a number against a source of truth, and defending the result to someone who disagrees - are worth more. Postings agree: AI is mentioned in a third of fresher and nearly half of experienced postings, and the share rose month on month during the audit. This roadmap treats AI as a tool you must use well, with a verification step on every AI-tagged topic.

Which BI tool should I learn: Power BI, Tableau or Looker Studio?

Power BI is named in 44% of fresher and 50% of experienced postings in our audit; Tableau in 30% and 37%; Looker in 6% and 11%. So learn Power BI properly - a real data model, DAX measures, a published dashboard - and learn enough Tableau to translate: level-of-detail expressions, calculated fields and dashboard actions. Looker Studio is free and good for practice and for Google-heavy companies. The skills that transfer - a proper data model, precise metric definitions, the right chart for the question - matter more than the tool.

How do I get hired as a fresher with no experience?

Make projects and an internship do the work experience would. Internships are named in 43% of fresher postings and a portfolio in 24%; a named certificate appears in 0%. So: two or three projects on messy public data, each ending in a recommendation and a two-minute recorded walkthrough, one named BI tool with a published dashboard, SQL fluent with timed practice, and a real or virtual internship - the OneRoadmap job simulation is built as one, with a verifiable result. Apply in focused batches to roles that match your projects.