Data Analytics cookbook

12 recipes · 87 steps · about 69 minutes of reading

Give your KPI one company-wide definition
Marketing says revenue is up. Finance says it's down. Everyone's looking at the same word and a different number.
Beginner to intermediateRead 6 minDoing: a few hours of analysis, plus however long both teams take to actually reply and sign off6 steps
Choose a chart type that won't mislead your reader
You built the chart in two minutes. Now three people are reading three different stories from it.
Beginner to intermediateRead 5 minDoing: an hour6 steps
Two categorical columns, cross-tabulated before you chart either one
Two columns look connected on the surface. A cross-tab is the only way to know if that's true or just a coincidence of group sizes.
Beginner to intermediateRead 5 minDoing: 20 to 30 minutes once the two columns already sit in one table you can query6 steps
The funnel step where most signups fall apart
Somewhere between the landing page and the final click, most people just leave. Nobody can say exactly where.
IntermediateRead 6 minDoing: an afternoon7 steps
Your blended retention number hides two very different groups of users
Overall retention looks flat at ten percent. Nobody can say whether every group is stuck at ten percent, or a few groups are propping up a much worse average.
IntermediateRead 5 minDoing: an afternoon7 steps
Split customers into RFM segments instead of emailing everyone the exact same message
Every customer gets the exact same email, whether they bought yesterday or vanished a year ago.
IntermediateRead 7 minDoing: an afternoon7 steps
Run a full EDA pass before anyone opens the dataset
You've never seen this dataset before. Everyone wants a chart from it by this afternoon.
IntermediateRead 6 minDoing: 45 minutes to an hour on a table you can already query9 steps
One region's number, explained by a sales-cube slice and dice
One region's revenue moved. The dashboard just shows the company-wide total, so nobody can see where.
IntermediateRead 6 minDoing: an afternoon8 steps
Decide if an A/B test result is real or just noise
The new button design is 'winning' after two days. Is that real, or would it flip back tomorrow?
Intermediate to advancedRead 6 minDoing: an hour7 steps
Tell seasonality apart from a real trend
Revenue dropped last month. Is that the usual January slump, or the start of something worse?
Intermediate to advancedRead 5 minDoing: a couple of hours7 steps
A seasonal pattern turned into a next-quarter forecast
Leadership wants next quarter's number. A straight line through last quarter ignores the fact that this business always spikes in December.
Intermediate to advancedRead 7 minDoing: an afternoon9 steps
Star schema or wide table: which does this report need?
Every new question from the business means rebuilding the same join again. Something about this table design has to change.
AdvancedRead 5 minDoing: a day8 steps

Cookbook badge

Finish all 12 recipes to earn this.

In these roadmaps

Data Analyststep 14 of 14

Data Analytics cookbook

Give your KPI one company-wide definition

Choose a chart type that won't mislead your reader

Two categorical columns, cross-tabulated before you chart either one

The funnel step where most signups fall apart

Your blended retention number hides two very different groups of users

Split customers into RFM segments instead of emailing everyone the exact same message

Run a full EDA pass before anyone opens the dataset

One region's number, explained by a sales-cube slice and dice

Decide if an A/B test result is real or just noise

Tell seasonality apart from a real trend

A seasonal pattern turned into a next-quarter forecast

Star schema or wide table: which does this report need?

In these roadmaps

Data Analytics cookbook

Give your KPI one company-wide definition

Choose a chart type that won't mislead your reader

Two categorical columns, cross-tabulated before you chart either one

The funnel step where most signups fall apart

Your blended retention number hides two very different groups of users

Split customers into RFM segments instead of emailing everyone the exact same message

Run a full EDA pass before anyone opens the dataset

One region's number, explained by a sales-cube slice and dice

Decide if an A/B test result is real or just noise

Tell seasonality apart from a real trend

A seasonal pattern turned into a next-quarter forecast

Star schema or wide table: which does this report need?

In these roadmaps