Dashboards & BI

Business intelligence and dashboard tools convert a database into something a person can look at, and the category's persistent problem is that looking at something is not the same as learning from it. Most organisations have more dashboards than questions, and the ones built for a specific decision are the only ones that get used past the first month.

The practical split among the launches here is between self-service tools, where non-technical users build their own views, and code-defined tools, where analysts define metrics in version-controlled files and everyone consumes them. Self-service is faster to adopt and produces the classic failure: four dashboards showing revenue, none of which agree, because each author defined it slightly differently. Code-defined semantic layers solve exactly that by making the definition of a metric a single reviewed artefact, at the cost of setup and a dependency on analysts.

If there is one thing to insist on, it is that number should mean one thing. A shared definition layer — even a documented one in a wiki — resolves more analytical confusion than any visualisation feature.

On visualisation itself, the good defaults are boring and well established: bar charts for comparison, lines for change over time, axes starting at zero for bars, no dual axes implying a relationship that is not there, and colour used to distinguish rather than decorate. Tools that make bad charts hard are doing you a favour. Anything defaulting to a three-dimensional pie chart is not.

Query cost is the operational surprise in this subcategory. Dashboards that refresh on a schedule against a warehouse charging per query scanned can quietly become a large line item, particularly when they are refreshing hourly for an audience that looks weekly. Caching, incremental refresh and simply reducing the schedule are the levers.

Access control deserves a check: a dashboard tool connected to a warehouse frequently sees everything, and row-level restrictions are a feature rather than a default. The salary table is in there somewhere.

The most useful dashboards answer one question, are looked at on a rhythm, and have an owner who removes them when the question stops mattering. See [analytics](/categories/analytics) for product and web measurement specifically.