The chart is not the analysis.

In 2012, Kandel, Paepcke, Hellerstein, and Heer interviewed 35 data analysts at 25 organizations. They wanted to see how analysis was done in daily work, not in a demonstration. What they heard was not mainly about charts. It was about the data before the chart. Finding and correcting the data was often the slowest part, and visualization tools did little for that task.

One analyst put it in terms of their own week: more than half the time went to integrating, cleansing, and transforming data, and they were lucky to get to any analysis. Analysis, which is what the job is named for, was left for when there was time left. It was not the center of the day. It was what happened at the end, if it happened.

Six of the most technical analysts used interactive charts only to present a story they already knew. They did not explore with them, because they could not work the data inside those tools. They did the work somewhere else and came back to the chart to show the result. The chart was the slide. It was not the place where the thinking happened.

The design point holds for a reason other than frequency. If the slow part is getting the data into usable shape, a tool that starts only once that is done skips the difficult part. An analysis system has to run on the data the business already has, where that data already is. Asking for a second copy inside a charting application creates another place to clean, another place that goes stale, and another argument about which of the two copies is the valid one. The slow part does not disappear. It is done twice.

An example makes this easier to see. If the analyst spends the first half of the week joining and cleaning data, the chart in the second half is not the analysis. It is a way of telling an analysis that has already happened. Starting from the chart means waiting until that first half is finished somewhere else, or copying the data in order to begin. Both options leave the slow part outside the tool.

A dashboard that starts with the data already clean starts after the work. It can look finished. What it shows was done earlier, somewhere else.

The plain consequence is this: do not start with the chart. In those interviews, most of the time went to getting the data into shape, and the chart arrived once the story was already clear. Ask the question of the data the company already has, where it already is. If the tool asks for a separate copy, you will clean the data twice and then argue about which copy is the valid one.