MeritKeep

SQL for product managers

If you already pull your own numbers with SELECT and WHERE, 3 of this subject's 26 skills are behind you and about 3 hours come off the estimate. Less than you hoped. What changes is not the length but the order: 5 skills are open to you today, and the ones that bite are about grain, not syntax.

Sources checked, not expert-reviewedSQL & Data AnalysisNo signup
Settle it — take the check, about ten minutes
This page assumes
  • You type your own queries into a dashboard's query box rather than asking an analyst for every number.
  • You can read a query somebody else wrote well enough to argue with the definition inside it.
  • You have never had to prove that a total was not inflated by a join.
If none of that is you, the ordinary SQL & Data Analysis path is the better page — it assumes nothing.
3Skipped, if we are right
5Already yours, renamed
18New to you
34–44Hours, not 47
The low end assumes what transfers really does transfer; the high end assumes none of it does. Nothing comes off for the skills we think you have until a check says so. Lay it out week by week at the hours you actually have.
01 · What we would skip

What you already have

Every one of these is a guess about you, and every one of them has a check behind it. Take the ones you doubt.

You choose columns, filter rows and sort them already — that is what every question you type into a query box is made of.

02 · What transfers

Things you already do, under another name

These stay on the path. What you bring is a head start on the idea, not the hours — and each one has a place where the resemblance stops being true.

You read averages and counts every week, so the functions themselves are familiar.

Familiar until a column has blanks in it. Aggregate functions ignore missing values rather than complaining about them,1 so an average over a column with gaps is an average of a smaller population than the one you think you are looking at — and different engines disagree about the edges.2

You already say per user per week in a spec, which is the hard half of grouping.

Saying the grain and proving it are different jobs. A result at the wrong grain looks exactly like a result at the right one — same columns, plausible numbers, no warning — and the only defence is being able to state what one row means and check that it does.

You define cohorts for a living; the retention curve in your review is one.

Defining a cohort is not building one. From a raw event log it means a first touch per person, a truncation to the period you care about, a join back to everything that happened later, and a decision about what counts as active — each of which can be wrong on its own.

You know what a wrong number costs in a review, which is the instinct behind checking one.

The instinct is not the method. Validating a result means knowing which totals have to reconcile, which row counts must not change across a join, and what order of magnitude you expected before you ran it.

The pivoted table in your weekly deck is conditional aggregation with a different name on it.

A spreadsheet builds those columns from whatever it happens to find. In a query you name every column in advance, which forces a decision the sheet made for you: what happens to a category that first appears next month.

03 · Where you start

Open to you today

Nothing below is waiting on anything else you have not done. This is read off the dependency graph, not chosen.
04 · The rest

What none of this covers

  • foundations2 skills · ~3h
  • aggregation2 skills · ~3h
  • joins5 skills · ~10h
  • composition3 skills · ~5h
  • windows4 skills · ~9h
  • analysisone skill · ~2h
  • performanceone skill · ~3h
The full skill map has every one of them, in the order they depend on each other.
05 · The graded work

Briefs you could not hand in today

Each one targets at least one skill this page has not credited you with. The marking checklist is published before you start.
06 · Where this comes from

Sources, and what each is worth

  1. 1. PostgreSQL, Aggregate FunctionsThe reference, and the place the null-skipping rule is stated plainly rather than discovered in a review.
  2. 2. NULL Handling in SQLite Versus Other Database EnginesShort, and useful precisely because it shows the rules are not universal — worth reading once so you stop assuming they are.
  3. 3. SQLBolt — interactive SQL lessonsFree, interactive, and the fastest way to check whether the skills this page credits you with are really there before we ask you.
07 · Also asked

Questions that come with this one

Do I need to know any SQL to take the check?

No, and it needs no account. It adapts as you go — if the early questions are easy it stops asking them, and if they are not it stops guessing and tells you where you actually are. SQLBolt is a decent way to test the same ground yourself first.3

How is this different from the ordinary subject page?

Same skills, same graded briefs, same marking checklists. What differs is the order and the starting point — this page names what we would skip for you and what we would not, and the check is what settles whether we were right about you in particular.

Why is so little of it skippable?

Because reading a number and producing one are different skills. The parts of this subject that go wrong quietly — grain, missing values, a join that inflates a total — are invisible from a dashboard, which is exactly why the estimate does not shrink much.

What if I already know more than this page assumes?

Then the check says so and the path gets shorter. Everything here is a hypothesis about you rather than a finding, and nothing in the product acts on it until you have answered questions that settle it.

Arriving with something else

Find out which half of this is true

Everything above is what we would assume about product managers. The check asks you instead — it adapts as you answer, needs no account, and what it settles carries into the path if you build one.

About ten minutes. Nothing is marked until you hand something in.Take the SQL & Data Analysis check

One cookie, so we can see which subjects people finish and where the site loses them — plus a replay of where you click. Nothing you type is recorded, and once you’re signed in the words on screen are blanked out too. No advertising, nothing sold. The detail.