Why this course exists
Every week someone slides a dashboard across the table and asks you to decide. Data literacy is not learning to code - it is knowing what that dashboard can and cannot tell you, which question to ask next, and when to push back. Over 8 sessions you will learn the language of data, the machinery behind it, the metrics that run internet, SaaS and e-commerce businesses, and just enough machine learning to be dangerous in the right way. Each session uses a real company story and a hands-on interactive exercise - no code anywhere.
What data actually is 5 min live
"Data" covers your ERP tables, last quarter's NPS free-text, and the security camera feed in the warehouse. Those are wildly different materials, and they cost wildly different amounts to turn into decisions. Two carves matter: how structured it is, and what kind of measurement it holds.
LiveStructured, semi-structured, unstructured3 min▶
Structure decides cost and speed. When you ask "can we analyze what customers complain about?", you are asking someone to turn unstructured text into structured categories first - that is real work, and knowing it buys you realistic timelines.
- Structured - already rows and columns. Orders, payments, inventory. An analyst answers questions about it in hours.
- Semi-structured - has labels but not a fixed table: app events, JSON payloads, emails. Days, not hours.
- Unstructured - text, audio, images. Most of what a company stores, least of what it uses. Session 7 shows the basic trick (sentiment analysis); the fancy tricks are the AI literacy course.
Moneyball started as a structure advantage: the 2002 Oakland A's could not outspend the Yankees, so they out-analyzed them - decades of structured play-by-play stats revealed that on-base percentage, not batting average, predicted runs. The insight was sitting in tables everyone already had. Same data, better question, playoff run on one-third the payroll.
Self-studyThe four measurement levels, and the mistakes they prevent2 min read▶
Nominal, ordinal, interval, ratio - a 1946 psychology paper (Stevens) that still decides which math is legal on which column.
| Level | Example | Legal math | Classic mistake |
|---|---|---|---|
| Nominal | region, payment method | counts, mode | "average region" - meaningless |
| Ordinal | NPS bucket, support tier | median, ranking | averaging 1-5 ratings as if the gaps were equal |
| Interval | dates, temperature | differences | "March is twice February" - no true zero |
| Ratio | revenue, units, duration | everything incl. ratios | rare - this one is safe |
The exec takeaway: when a slide averages an ordinal scale ("satisfaction is 3.7"), treat it as a rough signal, not arithmetic truth - and feel free to say so.
Data → information → insight → action 5 min live
Data has zero value until it changes a decision. The ladder is the whole business case for analytics in four rungs - and the reason most dashboards disappoint: they stop at rung two.
LiveClimbing the ladder (and where climbs stall)3 min▶
- Data → information: aggregation. Counting, averaging, grouping. Machines do this well - it is the cheap rung.
- Information → insight: comparison and surprise. An insight is information that changes what you believe: a segment that behaves differently, a trend that broke, a cause behind a number. This rung needs a human question.
- Insight → action: a decision with an owner and a date. The most skipped rung in corporate life - meetings end at "interesting" instead of "therefore".
- The loop-back: every action creates new data. Session 8 turns this into a management routine.
Self-studyWhen data climbs too far: the Target lesson2 min read▶
The ladder has a guardrail. In 2012, Target's statisticians built a pregnancy-prediction score from shopping patterns (unscented lotion, supplements, cotton balls) and mailed baby coupons - famously reaching a teenager before her father knew. The analytics worked perfectly; the action was the failure. Data literacy includes rung-4 judgment: just because data supports an action does not mean the action is wise, legal, or kind. Ask "how would this feel published on the front page?" before shipping the campaign.
The analytics project lifecycle 5 min live
Every analytics project - a churn study, a pricing review, an ML model - walks the same seven steps. Industry calls the formal version CRISP-DM; you just need the plain-language loop, because the step that fails is almost never the math.
LiveWhere to spend YOUR attention as a leader3 min▶
- Step 1 is yours. "Look into churn" produces a wandering month. "Which customer segments churned most last quarter, and what did they have in common?" produces an answer. Sharp questions are the cheapest analytics investment that exists.
- Step 3 explains the timeline. Cleaning and joining data is 60-80% of real project time - the famous, unglamorous truth. When an analyst says "two weeks", this is where it goes.
- Step 6 is where ROI lives or dies. An insight without an owner and a date is a screensaver. End every readout by assigning both.
- Step 7 closes the loop: the action you took becomes the next experiment. That habit - act, measure, adjust - IS a data culture; everything else is decoration.
The A's again, one level up: Moneyball was not one insight but a lifecycle on repeat - question (undervalued skills?), data (play-by-play history), analysis (OBP wins games), action (sign the walkers nobody wanted), measure (wins per dollar), new question. Twenty years later every major sports franchise runs this loop as a department.
Self-studyThe cast: who does what in a data team2 min read▶
| Role | Owns | Ask them for |
|---|---|---|
| Data engineer (DE) | pipelines, the plumbing (Session 2) | "can we get this data flowing reliably?" |
| Data analyst (DA) | steps 2-5: queries, EDA, dashboards | "what happened, where, to whom?" |
| Data scientist (DS) | step 4 heavy lifting: models, forecasts (Sessions 6-7) | "what will happen? what drives it?" |
| BI / analytics engineer | metrics layer, self-serve dashboards (Sessions 4-5) | "one agreed number everyone trusts" |
Small companies collapse all four into one heroic human. Knowing the hats exist keeps your requests fair - and tells you who is missing when everything feels slow.
Inventory your own data ★ 6 min · everyone lists
Two minutes solo: list 5 data assets your business touches every week (order table, support tickets, reviews, web traffic, payroll - anything).
Tag each one: structured, semi-structured, or unstructured. Then tag the measurement level of its key column (nominal / ordinal / interval / ratio).
Compare with a neighbor: which of your ten combined assets is the most valuable AND least used? (Almost every pair finds one - usually unstructured review or support text.)
Keep the list. Session 2 asks where each asset physically lives; Session 7 shows what basic ML squeezes out of the text ones.
Climb the ladder yourself ★ 8 min · everyone sorts
Six statements from one e-commerce company. Place each on its rung. The distinctions feel obvious until statement four.
Autopsy one real decision ★ 8 min · everyone maps
Pick a decision your company made in the last quarter that involved data (a price change, a campaign, a hire, a kill).
Walk it through the 7 steps: what was the question? what data? who cleaned it? what analysis? what was the insight sentence? who owned the action? and the honest one - did anyone measure the impact afterward?
Mark where the walk broke down. In most rooms, 70%+ break at step 1 (no sharp question) or step 7 (no follow-up measurement).
Write one sentence you will say at the next such meeting to fix your broken step. That sentence is today's real deliverable.
Amazon institutionalized step 7 as the Weekly Business Review: every important action becomes a metric someone re-reads next week, out loud, against expectations. Session 8 builds you a lightweight version.
This week ◐ 25 min total
- Run the ladder test in one real meeting: for each chart shown, silently ask "what does this change?". Count how many survive. Report your number next session - the room average is usually under half.
- Finish your data inventory from Apply-along 1: 10 assets, tagged by structure and measurement level.
- Read the self-study cards above: measurement levels, the Target lesson, and the data-team roles.
- Optional deep end: Provost & Fawcett, Data Science for Business, chapter 1 - the best 30 pages ever written on data thinking for non-engineers.
Three questions before you go 🎯 ◐ 90 seconds
1 · "Customer satisfaction averages 3.7 on our 1-5 survey." What should a data-literate leader flag?
Ordinal scales rank, they do not measure distance. The gap between "4" and "5" may be a chasm while "2" to "3" is a shrug. Medians and distributions tell the honest story.
2 · "Cart abandonment was 71% last month" is on which rung of the ladder?
Aggregation makes information. It becomes insight only when a comparison adds a "because" or a "which means" - like the shipping-cost finding.
3 · Your analyst says the churn study needs three weeks and you are surprised. Where is most of that time statistically going?
The unglamorous truth of the lifecycle: joining, deduplicating and fixing data dwarfs the analysis itself. Fund the plumbing (Session 2) and the three weeks becomes one.
Concepts covered & their origins
This course teaches the working 80% of each idea and credits the canon - the books and papers stay the deep end for anyone who wants more.