Drill-Down

Clicking a summary number to see the detail behind it, and the trap of drilling without a question.

Also known as: drilldown, drilling into detail, drill through

Drill-down is navigating from a summary number to the detail behind it: a total breaks into regions, a region into countries, a country into accounts, an account into transactions. It is the difference between a number and an explanation.

Why it matters

A dashboard tile tells you something is wrong. Drill-down is the only thing that tells you where. Without it, every follow-up question becomes a new request to the data team, and the person with the question waits.

The trap

Drilling without a question. The path from “revenue is 8% below target” to “here are the forty thousand transactions” is technically possible and analytically useless. Drill-down answers a question that got narrower, not one that got bigger. “Which segment explains the gap?” narrows. “Show me everything about revenue” does not — it just moves the work from the analyst to the reader, who now has to do the narrowing themselves.

Designing one

Every level should answer a question somebody actually asked. If a level exists because the join was easy, delete it.
Keep the drill consistent with the metric definition, or the detail will not reconcile with the summary. Two definitions of “revenue” produce a waterfall that disagrees with the tile above it — see metric definitions.
Watch the grain. The commonest drill-down bug is a join that changes grain and quietly multiplies the detail rows, so the drill shows more than the total.
Put a row count on each level. A reader who drills into five rows needs to know that before they form an opinion.
Make the top level open in one click. A summary number you cannot open is a dead end, and readers stop trusting the numbers above it.

The trade-off

Each level is another query, and the deepest level is usually too slow to be interactive. Pre-aggregate the levels people actually use — aggregate tables and olap cubes exist for this — and leave the raw table for the rare case. Give readers the freedom and it becomes self-service analytics; keep it inside a page and it is dashboard design or a bi dashboard. And when the drill is a fishing expedition after an experiment, ab testing is the discipline that stops it.