Show evidence, not verdicts
From a dashboard I designed for wholesale sales reps, who take retailers’ orders for manufacturers, and the agencies they work for.
Reps and the agencies they work for trusted a rule of thumb: a retailer who hasn’t ordered in a while probably hasn’t been visited in a while, since reps take many of their orders in person. So they watched the days since each retailer’s last order and made contact once the count passed a threshold, as if the number named both the problem and the fix. All it could show was how long it had been. The dashboard had to respect the gap between what a number shows and what people read into it: a product should calculate and compare, and show what its records establish, but the verdict about what that means belongs to the person.
Two labels for the same number show the difference: “Days since last order” states what the records say, while “Retailer needs a visit” builds the rule of thumb into the label and turns an interpretation into an instruction. I check labels like these closely, because an instruction directs somebody’s time. Color got the same care: a card changed color only when its number crossed a threshold people had set themselves, so the dashboard never turned a number red on its own judgment.
As we worked through the figures with reps and agencies, the rule turned out to be only partly true. Some retailers had gone quiet because nobody had been by. Reps told us about others that had gone quiet over repeated problems with their orders, where a visit to ask for the next order was the wrong conversation to walk in with. A dashboard that labeled every quiet retailer as needing a visit would have repeated the rule of thumb with more authority.
One threshold can’t fit every retailer, either. A retailer that buys almost everything at two shows a year goes months without ordering and is perfectly healthy, while a retailer that orders every month can be in trouble after a single missed month. The retailer’s own history is the best guide to whether its quiet is unusual, so the dashboard kept the neutral measure and put that history one step away from it: the retailer’s orders by month, this year against last. A comparison like that is still evidence, because the records back it up; it shows the orders falling without saying why or what the rep should do.
The pressure to label retailers anyway was real: at a session of our monthly council of agency owners and reps, an agency owner asked for an “at-risk retailers” list the dashboard would fill in on its own. Later, the product team started on a task list that would tell reps whom to call. Both would have put the rule of thumb on the screen as an instruction. I kept the at-risk list off the dashboard and showed the owner the retailers’ histories instead. For the task list, I argued that each task should open the records that raised it rather than just name someone to call. I also argued that the rep, not a change in the records, should decide when a task was done, and that’s the version we started building.
A retailer’s history changed the conversation a rep walked into. Reps on the council said they stopped opening with “You haven’t ordered in a while,” which gives a retailer little to respond to, and started with the trend instead. That let them ask what had changed rather than arrive with a diagnosis.
Everything the dashboard named, it could establish from its records, like how long a retailer had gone without ordering or which orders sat in a status like pending. Some explanations never reach the records at all, like a phone call, a new agreement, or a change in the retailer’s plans, so whether a pattern adds up to a problem is for the rep to decide.