How I work

Every case on this site follows the same six phases, adapted from the Google Data Analytics process. What I added is the part that makes it repeatable: each phase has an exit gate, and I do not move to the next phase until every item on it is met. When something fails, I go back rather than improvise forward — finding out during analysis that the data cannot answer the question means returning to Ask, not writing a softer conclusion.

The six phases

#PhaseQuestion it answersWhat it produces
1AskWhat is the real problem, and whose is it?A business question, the decision it unlocks, and metrics with operational definitions.
2PrepareIs this data any good, and can I trust it?A record for every source: origin, licence, period, known biases.
3ProcessIs it clean, and can I prove it?A cleaning log where every transformation carries its reason and its row counts.
4AnalyseWhat does the data actually say?Results, plus the checks I ran to try to break them.
5ShareWill the audience understand it?Charts whose headline is the finding, not the variable name.
6ActWhat should the business do?Recommendations ranked by impact against effort, with their limits stated.

Three rules I do not break

Everything is documented as it happens, not afterwards. In three of the six phases the documentation is the deliverable. Reconstructing decisions from memory produces weak case studies, and that is exactly the detail an experienced reviewer notices.

No invented numbers. If a figure did not come out of the data, it does not enter the report. Where there is a gap, it is declared as a limitation.

Raw data is never edited in place. Every transformation produces a new file and an entry in the cleaning log, so the whole chain can be re-run from the original source.

What you can read, and how deep

A case study produces far more material than anyone wants to read at once, so it is published in three layers. This page and the case pages are the first two; the third is one click away in each case's repository, for anyone who wants to check the work rather than take my word for it.

LayerTimeWhat it is
The card15 secondsHeadline, key chart, and what the case demonstrates. The home page.
The case5 minutesContext, data, process, findings, recommendations, limitations.
The evidence30 minutesCleaning log, scripts, source records, the full case record. In the repository.

How I choose what to work on next

A portfolio is an argument about what you can do, not a folder of projects — three cases demonstrating different things beat eight demonstrating the same one. So before starting a case I check what it would add along four axes: the type of problem, themain tool, the kind of data, and whether the result has ever contradicted my starting assumption. If a candidate case fills none of those gaps, the effort is better spent elsewhere.