Experiment DesignTutorialAnalyse an A/B test in Python: rates, z-test and confidence interval
Compare two conversion rates the careful way: the difference, a p-value, a confidence interval, and how many visitors you needed in the first place.
Author
Grokem Editorial
Editorial desk · Grokem
The Grokem editorial desk. Every piece is reviewed and approved by an editor before it is published, and shows the date it was published and last updated.
Published
Everything by Grokem
Experiment DesignTutorial
Hypothesis TestingExplainerHypothesis testing and p-values without the folklore
What a p-value is, what it is not, why five percent of tests with no real effect still come out 'significant', and what to report instead of a verdict.
Python & SQLTutorialClean a messy CSV with pandas: text, numbers, dates and duplicates
Work through the five problems almost every export has: stray spaces and case, numbers stored as text, bad dates, missing values and repeated rows.
Data PlatformsComparisonData warehouse, data lake or lakehouse: how to choose
Three ways to store analytical data, compared on the things that decide a real project: what is stored, who queries it, how it is governed and what it costs to run.
Interview PreparationHow-toSix SQL interview questions, worked through
Second-highest salary, top per group, rows with no match, comparing to a manager, month over month change and filtering groups, each with a tested solution and the reasoning to say out loud.
UTM ParametersBest practicesUTM parameters: naming conventions that survive a year of campaigns
Five tracking parameters, one rulebook and a small script that enforces it, so 'Email', 'email' and 'e-mail' stop appearing as three different channels.
Tag ManagementExplainerThe data layer: how a website hands information to its tags
A data layer is a plain JavaScript array where the page states what is happening, so tags read agreed values instead of scraping the page and failing at the next redesign.
GA4 Tracking AuditTroubleshootingGA4 events are missing or wrong: a systematic checklist
Work from the browser to the report in order: does the tag fire, does the event arrive, is it filtered or delayed, and is it configured so the report can show it?
Data QualityHow-toFind and remove duplicate rows in SQL
Count the copies, number them with ROW_NUMBER, delete all but one per group, and add a unique index so they cannot come back.
Data ModelingTutorialDesign a star schema for sales reporting
Turn a flat sales export into one fact table and three dimensions, decide the grain, and answer questions with a single join pattern.
Supervised LearningExplainerPrecision, recall and F1: judging a classifier beyond accuracy
Why a model that is 62% accurate by never predicting anything is useless, and how the confusion matrix, precision, recall and F1 tell you what a classifier really does.
WebhooksIntegrationReceive a webhook safely: verify the signature, check the age, ignore repeats
A webhook endpoint is a public door. Build one that checks who sent the request, rejects old replays and survives the same event arriving twice.
Version ControlHow-toUndo the last Git commit: reset, revert and restore
Three ways to take back work in Git, and the one question that picks between them: has the commit already been shared?
AttributionExplainerAttribution models explained: who gets the credit for a sale?
First click, last click, linear, time decay and position-based, applied to one customer journey so you can see exactly how much each model changes the answer.
ETL / ELTExplainerETL versus ELT: what changed and when each one fits
Both move data from where it is made to where it is analysed. The difference is where the transformation happens, and that decides cost, flexibility and who can change it.
IndexingExplainerDatabase indexes explained: why a lookup stops scanning the whole table
What an index is, what it costs, how to read the difference between a scan and a search, and the four reasons a query ignores the index you made.
Python & SQLTutorialBuild a small analytics pipeline with Python and SQLite
Load a CSV export into SQLite, reject the rows that do not fit, and answer a business question with SQL. Nothing but Python's standard library.