In a famous episode of Seinfeld, Cosmo Kramer pretends to be the “Moviefone guy” to callers who dial Kramer’s home by mistake (because his phone number is so similar to Moviefone’s). Kramer pulls off a decent impression of the Moviefone guy’s signature voice, but is unable to decode the touch-tone signals that callers use to select the films they’re interested in hearing about. After Kramer bungles an incoming call from George Costanza — randomly guessing films like “Agent Zero” and “Brown Eyed Girl” in response to the unintelligible beeps and boops of George’s keystrokes — he gives up and asks, somewhat desperately: “Why don’t you just tell me the name of the movie you’ve selected?”
Kramer meant well, but impersonating a machine in order to make sense of and respond to information is hard work. Two decades after that episode was filmed, thanks to the internet and mobile devices, we have more information to make sense of and respond to than ever. There are “analytics” for everything, which means there are also “dashboards” full of charts, stats, and data visualizations for everything, too. Apple, for instance, assumes that you really want a dashboard for tracking your personal health.
It looks great. It looks like data should look: dense, accurate, and futuristic. So why do I feel like I need to impersonate a machine in order to actually make any sense of it? Why doesn’t it just tell me what I need to know, in words, so I don’t feel like Kramer pretending to be the Moviefone guy?
“I don’t hate dashboards — I think they just crushed my spirit,” says writer/programmer Anil Dash. In 2013 he wrote a much-debated blog post arguing that “all dashboards should be feeds” — in other words, that brief written reports can convey actionable insight more efficiently than an array of dials, gauges, trendlines, and statistics. “The bias in dashboards is towards what’s easy to output from a database,” Dash says. “It’s very literal: here’s a bunch of numbers. For anything other than quickly glancing to see that nothing’s in the red, [dashboards] are pretty lazy about constructing any kind of narrative that is useful to a person.”
Dash’s social-media analytics startup, ThinkUp (co-founded by Gina Trapani), practices what he preaches. It digests the activity on a user’s Twitter or Facebook feed and reports any salient changes in short written messages.
If dashboards treat you like a CIA analyst, ThinkUp’s feed treats you more like the president: a very important, very busy person who just wants the bottom line. Dash has implied that most dashboards are in the business of flattering their users more than informing them — “they demo well and look great in investor pitch decks…But they don’t actually help me make decisions,” he writes — but the user experience created by ThinkUp’s “data verbalization” approach is undeniably flattering too, in its own way. Words are personal. It’s no surprise that apps which “intelligently” filter and integrate data to fit a user’s specific perspective in time and space — like Dark Sky, which delivers curt “hyperlocal” rain forecasts, or Google Now, which can tell you how long your evening commute will take before you think to ask — often privilege text in their user interfaces.
Datasets with an abundance of “pre-existing insights” lend themselves to verbalization, says Dino Citraro, co-founder of Periscopic, a data visualization firm based in Portland, OR. Pre-existing insights are simply meaningful declarative statements (or combinations of them) that are extant in the data: for example, a message from Dark Sky saying “rain for the next ten minutes”; or an update from ThinkUp reporting “@ev tweeted 15 links to Medium.com this month, more than to any other site.” “If your data has a high amount of pre-existing insights, then you can create a very structured narrative experience for it,” Citraro explains.
Ineffective dashboards or visualizations don’t lack for insights; instead, they fall short on “structured narrative experience.” They fail at directly describing the context that would make their pre-existing insights meaningful. They show, but don’t tell.
The data verbalizers’ own “pre-existing insight” is that in many predictable use cases where analytics are involved, the value to users is in the telling, and the showing may very well be beside the point. When I open up Dark Sky, I’m not interested in visually comparing multiple weather-related variables in order to discover some novel pattern. I simply want to be told if there is unpleasant weather entering my vicinity in the very near future so that I can avoid it. ThinkUp’s insights are slightly more sophisticated, but Dash and Trapani are betting that they are still simple enough to be adequately communicated in plain English. “If my social-media dashboard makes me a literal analog to the pilot of a plane, we think there’s a middle ground which is like flying by wire,” Dash says. “I want the info I need to make decisions to guide the plane where I want it to go, but I don’t need to know the barometric pressure the entire time I’m in flight.”
According to Dino Citraro, highly visual dashboards like Apple’s health tracker could simply be hidden behind a simpler verbal readout —with visualizations available on demand, but not by default. “If I were tracking my blood count, it’s fine for an app to just say [in text], ‘Everything’s OK,’” Citraro says. “But under certain circumstances I would want to know more. Is it worse than last week? I don’t want to be surprised with ‘It’s not OK!’ all of a sudden. But I think people get enamored with visualizing data for data’s sake. That becomes really distracting. You can’t just throw up a bunch of stuff and say to the user, ‘Make sense of this yourself.’”