“Alright, Mr. Tableau, I’m Ready for my Close-up”
Can the success of COVID-19 dashboards be replicated in BI and analytics?
Post COVID-19, are dashboards finally poised for their big moment?
Generalizations are dangerous, but I’m comfortable stating that as a planet we’ve never cared more about data than today, as we fight the COVID-19 pandemic. We’ve been conditioned over the past decade for immediacy, and our constant connectivity drives us for feedback the moment we feel the urge for answers. From the current storm came a visualization that has become the global pandemic standard.
The Coronavirus dashboard created and maintained by Johns Hopkins is the iPhone of visualizations. People from all walks of life, across the entire planet, are pinging the site at a rate of over 1.2 billion times per day. That’s 14,000 hits per second! The dashboard has evolved from one that did not exist on January 21 to what Google is to search in two months. In the press, in the television news, in social media … the dashboard is everywhere.
The importance of this visualization can’t be understated. It has delivered an entirely new audience to analytical visualization. The world is different in important ways than it was a month or two ago, but for anyone who has toiled for adoption, acceptance, and adoration of a corporate dashboard deployment, it may be now or never!
There are some very important design aspects of the site worth appreciating. It’s intuitive. It starts in the top left with THE metric (total confirmed cases). Next to the number is a map visualizing those cases using the same color (red). Below the summary metric is a ranked list of countries with somewhat obscure secondary lists of cases by province/state and a US list by county.
The next most important metric for many is across the map: total deaths (in white). It’s followed by total recovered (green). If you want to know active cases, you’ll have to do a bit of math. Across the bottom, there is information about updates, a count of locations included, and a chart that plots confirmed cases and daily changes. Simple, effective, ubiquitous.
Not so fast…
While there is no doubt that the planet is more attuned to visualizations than ever before, anyone thinking they should copy the layout to count sales, customers, phone calls, or some other KPI du jour should consider carefully. The Hopkins dashboard is great because of its simplicity in a constantly changing environment, but that simplicity comes with severe, potentially catastrophic limitations within a corporate setting.
The Hopkins dashboard can slice data at levels already mentioned: by status (confirmed cases, deaths, recovered) and location (country, region, and US county). What’s missing is an ability to understand why the numbers are changing, and most importantly, what that means to ME, the user. For COVID-19 ME includes my loved ones. In your business organization, ME is your division, your geography, your product. ME can be 100 things to 100 people. It’s getting from the standard that everyone agrees down to ME that makes all the difference in visualizations.
ME is the single biggest reason corporate deployments fail. Users can’t investigate the questions they have by examining attributes and looking for relationships in the data. To scratch that itch, users leave the dashboard and look elsewhere. When this happens, the perceived (and in reality, the actual) value of the visualization falls dramatically.
Why bother with a dashboard if I need to go elsewhere to answer my questions? Why don’t the numbers match? Is it the same line of business? Is it the same period? The user (often subconsciously) dismisses the utility, and the dashboard never regains its original appeal.
Moving toward detail
If you’re like many people, the closer Coronavirus migrated toward your location, the data available on the Hopkins dashboard provided fewer and fewer answers. The first concerns were likely proximity. If there are confirmed cases in my state, are those near my home? Is there any chance someone frequents my gym? Are those affected elderly, and are my small children safe? Did those who passed away have preexisting medical conditions (like mine) that made them more susceptible to the virus? What percentage are male or female? Are there assumptions associated with the data (might it be potentially biased)?
The answers to these questions can all be found, but the information comes from different sources, at different times, and with different numbers. Confidence in any set of numbers weakens, as 1+1 never equals 2. This is what happens to most corporate dashboards. The details are missing, so “offline” investigations occur, and those details never quite align with the initial summary. Reconciliations are needed, but who has the time?
To deliver meaningful visualizations to your users, you have to deliver not just KPI measurements, but the ability to answer questions about ME.
Building Success
The only way to truly deliver ME is to build reports from the details. For business organizations, it means going to the transaction level systems for the information. If data is coming from the accounting package, it probably lacks the attribution I need to see ME. If your organization has multiple systems with varying levels of detail, it’s critical that data be standardized as much as possible and available through the dashboard. Every missing detail reinforces to the user “they don’t know ME”, and those dissenters will eventually sink your visualization across the organization …meeting by meeting, reporting period after reporting period.
To get your Viz to sing … Make it MUSICAL
To deploy a visualization that will have staying power and become a shared information platform for users, try to build focus in these areas:
M — ME. Make sure that when the user engages, they can see their group, their team, their location and themselves in the presentation layer.
U — Unbiased. Be sure to use all of the data, not just the data that tells the best version of the story. The data has to be credible.
S — Symmetrical. Data on one chart should support evidence for all charts.
I — Intuitive. The titles, axis, and labels should make the information obvious. If it’s not easy to understand and discuss, it’ll never get traction.
C — Consistent. Changes should be limited whenever possible, and users should be able to replicate views and reports they generated previously.
A — Automatic. Data feeds, cleansing, organization, and population should be automated. Manual interventions have data and knowledge risks that can destroy credibility in a flash. Second chances often prove to be elusive.
L — Layered. Periodic information should build layers like sheets of paper. You should be able to go back through the layers to compare/contrast.
The opportunity for dashboards and visualizations has never been higher. With good data and a commitment to strong design principals, your organization might finally be ready to accept the insights and understandings you’ve been offering. It’s “go-time” in the visualization and analytics space!
