Let each finding become the next question

From a dashboard I designed for wholesale sales reps, who take retailers’ orders for manufacturers, and the agencies they work for.

Asking someone to set up the right report before they start assumes they know where their questions will lead. They don’t; that’s why they’re looking. Each thing they find changes the question, so a product has to let a finding become the next question without making anyone rebuild the view that raised it.

I saw this in research: when we asked reps and agency owners which numbers they wanted on the dashboard, the answer was often silence, or a vague idea about totals. A few power users showed us exactly how they added up their data, and their examples shaped the numbers we offered everyone else, but most people couldn’t name their questions until they had something in front of them.

In one of our weekly research sessions, a rep’s sales for the year were behind last year’s, so the first question was which months, and a chart of this year against last showed July and August well below. The next question was whose orders were missing, and the orders behind July’s bar opened in the table below the chart. The rep knew one retailer had ordered every month last year and, sorting the orders by retailer, found it wasn’t there. It wasn’t in August’s orders either.

The chart stayed as it was while the rep read those orders, and that mattered for the next question: how much of the drop did this one retailer explain? The bar for July of last year opened its orders the same way, so the rep could see what the retailer usually ordered and weigh that against the gap between the bars. The retailer’s usual orders came to about a third of the gap, enough to be worth a closer look without being the whole story. If opening the orders had replaced the chart, as drilling down does in many reporting tools, the rep would have had to rebuild it to check. At that point it’s tempting to skip the check and assume the first odd detail explains everything.

The missing retailer raised a question the rep hadn’t started with: was this new? The rep found the retailer well down the table that ranked their retailers by orders and, done with the sales chart for the moment, chose to chart that retailer’s orders month by month in its place. The retailer’s orders had been falling since late last year, a drop the rep had never noticed inside the mix of all their retailers. Getting there took fewer than ten clicks, no typing, and no spreadsheet. The dashboard couldn’t say why the retailer had slowed down, but the rep could call the retailer already knowing what the orders showed.

Every step was one move from the last, because any number could open into the records behind it and any record could become the subject of a new chart or table. That depended on each part of the dashboard keeping its own timeframe, so opening July’s orders left the chart showing the whole year.

People found separate timeframes confusing, and some asked for one timeframe across the whole page, which would have turned the year’s chart into a chart of July the moment July’s orders opened. What people could do was star a timeframe as their default: the chart, the ranking table, and the table of orders opened on it whenever the dashboard loaded, and any of them could then change on its own. Cards kept their own timeframes, and every part labeled its period, so nobody had to remember what each one showed. A few people went on asking for one locked timeframe, and the separate timeframes stayed, because the investigation depended on them.

An investigation rarely ends at its first question, and each answer gets checked against the view that raised the question. Opening what’s behind a view shouldn’t replace it; the person should decide when they’re done with it.