·
Fewer dashboards, clearer answers.
A fintech had 240 dashboards and still could not say why a number moved. None of the 240 was kept. The business now runs on eight new dashboards built around its decisions, and they open in 3 seconds instead of 15.
240 dashboards became eight.
- A new dashboard for every new question
- Nothing was ever retired
- Typical load time of 15 seconds
- Opening one rarely settled anything: ask an analyst, wait, then Excel
- Eight dashboards, each built around a decision
- Each has a named owner
- Typical load time of 3 seconds
- The number and the decision sit on the same screen
240 dashboards, one big licence, no answers.
Nobody planned this. It built up one reasonable request at a time. Someone needed a number no dashboard showed, so a new one was built. The old one stayed, because retiring it was nobody's job. Repeat until 240 dashboards sat on a six-figure licence.
- Nothing was ever retired.
- The licence grew with the pile of dashboards.
- Some dashboards repeated another with a different filter. Some had not been opened in months.
- They loaded in 15 seconds. Above ten, people stop bothering.
- Opening one rarely ended the argument.
- Every new question gets a new dashboard and nothing comes off.
- Some of your dashboards repeat each other, and some are never opened.
- Yours take more than ten seconds to load.
- Opening one rarely settles anything, so you ask an analyst and wait.
Why it mattered. Speed decides whether a dashboard is opened at all. Under a few seconds, people check a number mid-conversation. Above ten, they stop bothering.
Start from the decisions, not the dashboards.
The tempting move was to lift all 240 onto something cheaper. Our view was different: the problem was never usage. It was that opening a dashboard rarely ended the argument. So we started from the decisions the business makes, and treated speed as a reason people would use the dashboards, not a nice extra.
A test first, then eight dashboards built new.
- Week 1 test: every dashboard was asked four questions. Does it carry a decision? Does it have a named owner? Is it a duplicate? Is it still opened?
- Duplicates were merged and dead dashboards retired, not moved.
- Eight new dashboards, each built backwards from a decision.
- A properly tuned Superset stack, not "installed and left".
- A cut-over off the old licence in weeks 5 to 6, bringing load time from 15 to 3 seconds on the same data.
