Documenting BI reports for marketers

Most of my documentation work has served one reader: a marketer who has been given a Power BI report and no data team to stand behind it. This page lays out the approach I have developed for that reader, what order to teach things in, where the documentation should live, and what it must never assume, and the worked, end-to-end demonstration of it, an open-source web app whose documentation set follows every principle below.

Assume the tool is scary and the theory is missing

Two assumptions shape everything else, and both came from being proven wrong the polite way, by real readers.

The first: a marketing manager once asked me a question that revealed they were not struggling with my reports at all. They were missing the fundamentals of their own discipline that my documentation silently took for granted. Documentation answers the questions its audience is most likely to ask, but there can be no questions where there is no underlying knowledge to spark them. So the documentation has to supply that knowledge itself: discipline-level explanatory pieces, what the email send funnel is, why the list beneath it matters, sit alongside the product documentation, and the reader is routed through them before they ever meet a metric definition.

The second: the first time I documented a semantic model, I published the documentation in the format I knew best, a Power BI report. The BI-phobic marketing readers it was written for bounced off the Power BI service environment before reading a word. The reading surface is part of the writing. For this audience the documentation (and ideally the product walkthrough itself) lives on a plain, friendly web page, never inside the BI tool it explains.

Structure by the reader’s question, not the tool’s features

I structure this material with a user-pull approach: a landing page routes each reader by discipline, email, paid search, organic social, and within each discipline every page opens with the marketing question it answers, never with the feature it happens to describe. An email marketer arriving at the docs is first shown the funnel view their discipline is read through, and only then the definitions of the metrics that measure it.

The reference layer is a governed glossary: every metric and dimension gets exactly one canonical definition, in exactly one place, and every audience-facing page links back to it rather than restating it. Click-through rate and click-to-open rate are each defined once, which is the only reliable way to stop two teams arriving at the same meeting with different numbers bearing the same name. The glossary also carries a kind of negative knowledge that BI documentation usually omits: which dimension-metric combinations are semantically valid, and which produce a number that means nothing. Mapping the meaningless pairs, and keeping them out of every presentation, prevents the wrong conclusions before they are drawn.

Teach by doing, starting from blank

For this reader, the tutorial cannot be a tour of a finished dashboard, since a finished dashboard is precisely the thing they do not yet understand. The tutorial starts from a blank report template: empty selectors, an empty canvas. The reader picks a date axis, a legend, a first set of email metrics, and watches the plot and the table assemble themselves. Then they save what they built as a bookmark, switch to a different one, and come back, and in that round trip they learn, in their hands rather than in prose, what a metric, a dimension, and a bookmark actually are: a bookmark is nothing more than a saved pre-selection that answers one recurring marketing question.

The how-to guides then cover the recurring tasks of the role, and they are allowed to branch on what the data shows: check deliverability against the benchmarks first; if it is healthy, go read the click; if the list is eroding, fix that before anything else. A guide that pretends every reader’s data looks the same is a guide that abandons half of them at the first fork.

The shape underneath

Readers of my approach page will recognize the Diátaxis frame underneath all of this: explanation supplying the missing theory, reference as the governed glossary, tutorials that teach by doing, how-to guides for the tasks of the role. What this page adds is the calibration for one specific, underserved reader: the non-technical marketer. For them the order of arrival matters (explanation first), the surface matters (never the BI service), and the glossary must govern, not merely describe.

The demo: report-template-site

report-template-site is a small open-source web app (React + TypeScript, deployable to Vercel as a static site) that puts all of the above in one place: one combo chart and one matrix, driven entirely by three field-parameter selectors, plot axis, legend, measures, with a hamburger menu of built-in views and four bookmarks, each a saved pre-selection that answers one recurring marketing question (“is deliverability slipping?”, “which sends earned their keep?”). The interaction model replicates, in web form, the field-parameter-and-bookmark pattern professional BI teams build in Power BI; it runs on twelve months of fictional data for an unnamed specialty tea-and-homeware shop, seeded with deliberately weak sends and a quietly eroding list segment so that every bookmark has a story to tell.

The app’s second page is the tutorial: empty selectors, an empty canvas, and six numbered, coach-marked steps in which the reader picks an axis, a legend, and their first measures, watches the chart and matrix assemble themselves, saves the result as a named bookmark, switches to a built-in view, and returns to their own. Each step gates its “Next” on the reader actually having done the thing, instructional design as UI state. Four short explanation guides (the deliverability funnel, CTR versus CTOR, list health and erosion, campaigns over time and revenue) live both in the repo’s docs and inside the app itself, one behind each bookmark, so the discipline knowledge arrives at the moment the view raises the question. The report also enforces its own glossary: it declines to plot meaningless dimension-metric combinations, saying why in a notice, and objects when a per-send benchmark is asked of a monthly aggregate.

The documentation set is Diátaxis-shaped end to end: two tutorials (using the demo, and rebuilding the same template natively in Power BI Desktop with field parameters, bookmarks, and the selection-pane menu), three how-to guides (reading the report, swapping in your own data, deploying), a four-page reference layer (glossary, data model, bookmark shape, code map), and five explanation pieces, including one arguing the template’s own thesis, with its trade-offs stated honestly: one canvas plus bookmarks replaces a stack of static report pages, but only if the metric catalog stays governed.