My friend George P. Burdell, a man of profound insight, recently pointed me to Dr. Bissantz's article, "Can we steer banks like cars?", at Bissantz's blog: Me, myself, and BI.
After reading the article, I called my friend, George, and exclaimed, "Wow! I didn't even know banks had to obey traffic signals."
There was a long pause, and then a sigh.
George said, "Here's what I thought was the key take away: 'People who deliver green lights instead of the underlying numbers have made a decision instead of supporting the decision-making process.' You ought to think about that. That's worth generalizing into a decision support visualization principle, something like: 'Your data visualization should inform and support the decision process; it should not make the decision.'"
So I thought about it for a awhile. I called George back the next day.
"It seems to me that the problem decision support systems are trying to solve is this: How do you support a decision without forcing the decision maker to understand too deeply the model used to support the decision making? 'Green light' is obviously too high level… at the same time, we can’t expect someone to be an expert in quantitative analysis to use the models we develop for them, or our services won’t be useful in the first place. So, I’m curious, how do we decide where to draw the line between providing raw data and digesting it for the user?"
George responded, "Actually, green light/red light dash boards simply communicate a state of affairs against a predetermined preferential threshold. They are not just data digestions but interpretations as well. Unfortunately, the way most of them work, a system could run just under the line of a threshold and indicate 'good' when it is actually very close to a 'bad' situation. This is why Dr. Bissantz says the machine is doing the deciding. You might could argue that another light could be added (i.e., yellow) that suggested that one was approaching the threshold. However, I would say that the machine is still being given the role of human interpreter against a predetermined bias."
"So what should we do?" I asked George. "Don't managers want simple, quick answers so they can run their business on autopilot and go play golf?"
"Oh heavens! I hope not. The way around this is to provide analyses that do digest the raw data without imposing judgments through such icons such as green light or red light, go or no go, good or bad. We should simply report the digested form in such a way that quick, yet deep, insight can be gained and lead one to more questions. An example of such data digestion might be a risk profile or a tornado plot. They don’t make judgments about the state of things, they merely inform with data digested from the models."
And that seems smart to me.
The next time your dashboards indicate "green," you might want to ask what that really means.