A written version of the interactive roadmap above — every station, what you'll learn, and a small thing to build — laid out for reading, reference and search.
Data Foundations Start here
F1. Think in Questions
Beginner · 40 min
The hardest part of analysis was never the tool — it was asking the right question. Learn to turn a vague business ask ("why are sales down?") into a precise, answerable question, to distinguish the question someone asked from the one they actually need answered, and to scope an analysis so it produces a decision, not just a chart. This is the judgment AI can’t supply, and it stays valuable no matter how good the tooling gets.
Skills: From vague ask to precise question · The question behind the question · Scoping for a decision · Defining success upfront
Build it: A stakeholder asks "is the new feature working?" Write the three sharper questions you’d actually answer, and what data each needs.
✓ Checkpoint: Take a vague request like “how are we doing?” and rewrite it as a question whose answer would change somebody's decision. If no decision moves, the question is not scoped yet.
F2. SQL, the Core Skill
Beginner · 55 min
SQL is the language of data, and even in an AI era where it can write the query, you must read it fluently — to check it, fix it, and trust it. Learn SELECT, WHERE, GROUP BY, and the JOINs that combine tables, then window functions for running totals and rankings. AI writes SQL well; the analyst who can’t verify it ships wrong numbers confidently.
Skills: SELECT / WHERE / GROUP BY · JOINs across tables · Window functions · Reading and verifying AI-written SQL
Build it: AI hands you a JOIN that silently double-counts rows. Explain how a fan-out JOIN inflates a SUM, and how you’d catch it.
✓ Checkpoint: Explain what a window function does that GROUP BY cannot, and describe how you would check AI-written SQL you did not author.
F3. Beyond Spreadsheets
Beginner · 35 min
Most analysts start in spreadsheets, and they remain a genuinely useful tool for quick, exploratory work. Learn what they’re great at (fast ad-hoc analysis, pivot tables, one-off models) and where they break down (reproducibility, scale, versioning, shared truth) — so you know when to reach for SQL, a database, or a real pipeline instead. Knowing the limits is as valuable as knowing the tool.
Skills: Pivot tables and lookups · When spreadsheets shine · Where they break (scale, reproducibility) · When to graduate to SQL
Build it: A critical monthly report lives in one analyst’s spreadsheet and keeps breaking. List what makes it fragile and what you’d move it to.
✓ Checkpoint: Name the point where a spreadsheet stops being the right tool, using a property of the work rather than a row count.
F4. Data Cleaning & Trust
Beginner · 50 min
Real data is messy — missing values, duplicates, inconsistent formats, silent errors — and an analysis on untrusted data is worse than none, because it looks authoritative while being wrong. Learn to profile a dataset, handle missing and duplicate data deliberately, and, most importantly, judge whether the data can be trusted for the question at hand. That trust judgment is core analyst work AI can’t do for you.
Skills: Profiling a dataset · Missing data and duplicates · Inconsistent formats and silent errors · Judging data trust
Build it: A column of "revenue" has nulls, negatives, and two currencies mixed in. Describe how you’d investigate before reporting a single total.
✓ Checkpoint: Describe a data error that passes every schema check and still makes the analysis wrong — the silent kind you have to go looking for.
F5. Statistics that Matter
Intermediate · 55 min
You don’t need a statistics degree, but you do need the handful of ideas that stop you drawing wrong conclusions. Learn distributions and why the mean lies when data is skewed (use the median), variance and what "significant" actually means, correlation versus causation, and sampling bias. These are the judgment guards that separate a real analyst from someone who reads charts.
Skills: Mean vs median on skewed data · Variance and significance · Correlation ≠ causation · Sampling bias
Build it: Average revenue per user "jumped" this week. Explain how one whale customer could cause that and why the median tells a truer story.
✓ Checkpoint: Explain when a mean actively misleads and a median does not, and give a real pair of variables that correlate with no causal link between them.
F6. The Analysis Workflow
Beginner · 35 min
An analysis is a repeatable process, not a scramble: clarify the question, get and clean the data, explore, analyze, and communicate a decision. Learn to work reproducibly (so someone can rerun and trust your result), document your assumptions, and treat the deliverable as a recommendation, not just a number. A disciplined workflow is what makes your work trusted and reusable.
Skills: The end-to-end workflow · Reproducibility · Documenting assumptions · Deliver a recommendation, not a number
Build it: Your one-off analysis becomes the number leadership tracks weekly. What would you change to make it reproducible and trustworthy?
✓ Checkpoint: Explain why an analysis nobody can reproduce is worth less than a rougher one they can, and name the assumptions you would write down.
Analysis & Communication Build the craft
T1. Exploratory Analysis
Intermediate · 50 min
Before you answer, you explore — to understand the shape of the data and to let it surprise you. Learn to systematically look at distributions, relationships, and outliers, to form and test hypotheses against the data, and to notice the anomaly that changes the whole story. Exploration is where the real insight and the good next question come from.
Skills: Distributions and outliers · Relationships between variables · Hypothesis forming and testing · Following the anomaly
Build it: Signups look flat overall, but you suspect a segment is churning. Describe how you’d slice the data to find it.
✓ Checkpoint: Explain what you do when you find an outlier — and why “remove it” and “keep it” are both sometimes the answer.
T2. Visualization Done Right
Intermediate · 50 min
A chart is an argument, and the wrong chart lies without meaning to. Learn to choose the right visualization for the question (comparison, trend, distribution, composition), to design for clarity over decoration, and to avoid the classic distortions — truncated axes, misleading scales, chartjunk. AI can generate a hundred charts; choosing the one that tells the truth clearly is your job.
Skills: Chart type for the question · Clarity over decoration · Avoiding misleading axes/scales · Designing for the audience
Build it: A dashboard uses a truncated y-axis that makes a 2% change look like a cliff. Explain the distortion and how you’d fix it honestly.
✓ Checkpoint: Justify a chart type from the question being asked rather than from habit, and name a way an axis can mislead without a single false number.
T3. Dashboards & BI
Intermediate · 50 min
Dashboards turn a one-off analysis into a self-serve product, and building good ones is a real skill: pick the few metrics that matter, design for the decision the viewer needs to make, and avoid the sprawl of vanity charts nobody acts on. Learn the BI tools (Looker, Power BI, Tableau, and their kind) and, more importantly, the judgment of what belongs on a dashboard and what doesn’t.
Skills: BI tools (Looker, Power BI, Tableau) · Metrics that drive a decision · Avoiding dashboard sprawl · Self-serve design
Build it: An exec dashboard has 40 charts and no one uses it. Which handful would you keep, and what decision does each one serve?
✓ Checkpoint: Explain what makes a dashboard metric worth showing, and why most dashboards keep growing when they should be shrinking.
T4. Python for Analysis
Intermediate · 55 min
When SQL and spreadsheets run out, Python (with pandas) takes over: complex transformations, joining many sources, automation, and analysis you can version and rerun. Learn enough pandas to load, clean, and reshape data, and to automate the reports you’d otherwise rebuild by hand. Use AI to help you write it — and understand it well enough to debug when it’s wrong.
Skills: pandas: load, clean, reshape · Joining multiple sources · Automating repeated analysis · Reading and debugging AI-written Python
Build it: You rebuild the same three-source report every Monday by hand. Sketch how you’d automate it, and where you’d still check the output.
✓ Checkpoint: Explain when pandas beats SQL for a task and when it is the wrong reach, and describe how you would verify AI-written Python before trusting its output.
T5. Experiments & A/B Testing
Advanced · 55 min
The most valuable analysis often answers "did this change cause that outcome?" — and that needs experiments, not just dashboards. Learn A/B test design, why you need a control group, what statistical significance and power really mean, and the traps: peeking early, p-hacking, and confounds. Reading an experiment correctly is high-value judgment work that AI can set up but not own.
Skills: A/B test design · Control groups and randomization · Significance and power · Peeking, p-hacking, confounds
Build it: A PM says the test "won" after checking results daily and stopping when it hit significance. Explain why that result may be false.
✓ Checkpoint: Explain what goes wrong when you check an experiment early and stop as soon as it looks significant.
T6. Metrics & Definitions
Advanced · 45 min
The most political, most valuable question in analytics is "what does this metric actually mean?" Is "active user" daily or monthly? Does "revenue" include refunds? Learn to define metrics rigorously, to spot when two teams use the same word for different numbers, and to own the definitions — because whoever owns what the metrics mean owns the conversation. This is durable, human, judgment-heavy work.
Skills: Defining a metric rigorously · Same word, different numbers · Owning definitions · North-star vs vanity metrics
Build it: Two teams report different "revenue" for the same month and both are "right." Explain how, and how you’d resolve it.
✓ Checkpoint: Take a metric everyone at a company uses and explain how two teams could compute it differently — and who should own resolving that.
Advanced & AI-Era Own the future
P1. Analytics Engineering
Advanced · 55 min
Analytics engineering is where many analysts are heading, and it’s one of the strongest AI-era moves: modeling raw data into clean, tested, reusable tables that everyone trusts. Learn the modern stack (dbt-style transformation, version control, testing your data like software) and the mindset shift from answering one-off questions to building the trusted data foundation the whole company runs on.
Skills: Data modeling (dbt-style) · Version control for analytics · Testing your data · From queries to a data foundation
Build it: Every team computes "active users" differently. Describe the modeled, tested table you’d build so there’s one trusted answer.
✓ Checkpoint: Explain what testing your data means when there is no code to unit test, and what breaks without it.
P2. The Semantic Layer
Advanced · 45 min
A semantic layer is the single place where metric definitions live, so "revenue" means one thing everywhere — in dashboards, in queries, and increasingly in what AI tools answer. Learn what it is and why it matters more in the AI era: when people (and AIs) ask questions in natural language, the semantic layer is what makes the answer correct and consistent. Owning it is owning the truth.
Skills: Centralized metric definitions · Consistency across tools · Why AI makes it matter more · Governing the single source of truth
Build it: Your company is adding a natural-language "ask your data" AI tool. Explain why a semantic layer decides whether its answers are trustworthy.
✓ Checkpoint: Explain why a semantic layer matters more once AI is writing the queries, not less.
P3. Causal Thinking
Advanced · 50 min
Beyond "what happened" is "why, and what would happen if we changed it" — the causal questions leadership actually cares about. Learn the difference between correlation and causation in practice, the basics of causal inference when you can’t run an experiment, and how to reason about confounders. This is deep judgment work; AI can compute the numbers but cannot supply the causal reasoning about your specific business.
Skills: Correlation vs causation in practice · Causal inference without experiments · Confounders · Reasoning about "what if"
Build it: Users who use feature X retain better. Explain why you can’t conclude X causes retention, and what would let you.
✓ Checkpoint: Describe a confounder that would make an obvious causal story wrong, and how you would probe for it without running an experiment.
P4. Storytelling with Data
Intermediate · 45 min
An analysis nobody acts on is wasted, and the gap between a correct answer and a decision is communication. Learn to build a narrative — lead with the "so what", tailor depth to the audience, connect the number to the decision it should drive, and be honest about uncertainty. Turning analysis into action is where analysts become influential, and it’s squarely human.
Skills: Lead with the "so what" · Tailoring to the audience · Number → decision · Communicating uncertainty honestly
Build it: You have a correct, nuanced 12-slide analysis and a 5-minute exec meeting. What’s the one slide and the one sentence?
✓ Checkpoint: State the “so what” of a recent analysis in one sentence, before any numbers — and say how you would communicate uncertainty without hedging it into meaninglessness.
P5. The AI-Era Analyst
Advanced · 50 min
Text-to-SQL and auto-dashboards automate the mechanical half of analysis — so lean hard into the half that becomes more valuable. Learn to use AI as a fast query-writer and first-draft tool while owning what it can’t do: asking the right question, judging whether data can be trusted, catching the confidently-wrong AI answer, and interpreting results in real business context. The analyst who owns judgment thrives; the one who only wrote SQL is exposed.
Skills: AI as a fast query-writer · Validating AI answers · Judgment over mechanics · Interpreting in business context
Build it: An AI tool confidently answers "churn is up 3%" but pulled the wrong date range. Explain how you’d catch it and why that catch is your value.
✓ Checkpoint: Explain what an AI cannot supply when it writes you a correct query, and why the interpretation is where the job now lives.
P6. The Analyst Career
Advanced · 40 min
Data analysis branches into several futures: the analytics engineer who owns the data foundation, the data scientist who goes deep on modeling, the analytics leader who drives strategy, and the domain-expert analyst whose business judgment is irreplaceable. Learn where the role is going in the AI era, how to build a portfolio of real analyses that show judgment, and how to grow from pulling numbers to owning the questions and decisions.
Skills: Analytics engineer / data scientist / leader paths · Portfolio of judgment · From numbers to decisions · Growing your scope
Build it: Write the six-month plan that moves you from "runs the reports people ask for" to "owns the metrics and the questions worth asking".
✓ Checkpoint: Describe what the analyst, analytics engineer and data scientist tracks actually differ on — and which one your last six months of work resembles.