Data analysis, done properly.
What I do
I work with data — spreadsheets, databases, exports from whatever system a business happens to use — and turn it into something people can actually trust and use.
In practice that means finding where two numbers disagree, cleaning up what’s messy, and building reports or dashboards that answer a real question instead of just looking impressive.
How I work
How a number gets out of here.
None of these are clever. They’re the things that stop a wrong number reaching somebody who’s about to make a decision with it.
Assume the import lied
A single blank cell is enough for a loader to read an entire date column as text. I check what every column actually came in as before building anything on top of it — types, encodings, duplicate keys — because that kind of error doesn’t announce itself. It surfaces three joins later looking like something unrelated.
Say the number before you run it
Predicting the row count before a join and then checking it catches what counting afterwards never does; once a result is on screen it’s easy to rationalise. A join that fans out, or quietly drops a third of your rows, still returns a total that looks entirely reasonable.
The wrong version is the documentation
Six months on, nobody remembers why there’s a DISTINCT in that CTE — including me. The approach I abandoned, and the reason it broke, is what keeps the working version maintainable instead of load-bearing folklore.
Name the decision first
Every piece of work starts with which decision changes depending on the answer. If nothing changes, it’s decoration. And if the data can’t carry that decision, you hear it in the first conversation rather than in the final report.
Direction
Where this is going.
Right now the focus is analysis: making the numbers a business already has clear and trustworthy. Over time I want to move a level deeper — building the pipelines that keep those numbers reliable without someone rebuilding a report by hand every month.
The work says it better than I can.
If you’d rather see how I think than read about it, the projects section walks through a full analysis — including the ranking I built first, which pointed at entirely the wrong sellers.