Client work sits behind NDAs, so here are seven things you can inspect end to end: six Power BI builds, each mapped to a sector I have worked in, with their models, measures and data-quality work in public — one of them measured against a deliberately bad version of itself — and one product you can run against your own tenant export.
Three years of the COREP and FINREP returns 129 EU banks file with their supervisors. The sector’s CET1 ratio is 16.2%; average the banks’ own ratios and you get 22.7%. Every ratio is recomputed from its components, every item code mapped across three renumbered exercises, and the year-to-date P&L taken apart bank by bank.
Explore the data →
A decade of a bank’s card book: 1.7m transactions, what was declined, and what turned out to be fraud. Two thirds of the file is adjudicated and the rest is not, so the obvious way to compute the fraud rate is wrong by a third — the report shows both denominators. The same rows are then modelled a second time, badly, and measured against this one.
Five years of HR records: headcount, turnover, engagement and training. The HR system’s status field contradicts people’s own dates 1,146 times, the recruitment file turns out to belong to someone else, and 44% of training spend is booked against people who did not work there.
Explore the data →
Three years of a retailer’s orders to 164 countries: net sales, margin and delivery against the lead time each shipping mode promises. First Class is late on every order, and the mode that costs more delivers no faster — only the promise differs. The column called Sales isn’t revenue, and the last four months come from a different generator.
Explore the data →
Four years of a US retailer’s orders, rebuilt as a star schema whose product key turned out not to be one. 793 customers who almost never leave: retention, not acquisition, is the story, and every measure was checked against pandas before it was drawn.
Explore the data →
A full profit & loss statement built from a real general ledger. Subtotal and detail lines share one matrix without double counting, and a single measure drives three column views.
Explore the data →Turns a Power BI workspace export into a governance audit in seconds: over-privileged access, admin hygiene gaps and orphaned workspaces, flagged and explained.
Open Pulse →