Unyson Blog

Back

Visual Analytics: From Data to Dashboards

07/25/2017

One of Unyson’s key strengths is the visibility we provide. We build highly customized Tableau dashboards for our customers to support our top-notch analytics and operational services. No two dashboards are the same because no two customers are the same. However, there are some similarities between all dashboarding projects, and I’ve learned these ones the hard way.

Whether you are a consumer or creator of dashboards, I hope you will find these notes useful.

Begin With the End In Mind

Stephen Covey said it best. At some point in the dashboard building process, both generalities and details are going to have to be covered.  My advice is to not get wrapped up with details early in the process.  Parse out the important questions (“How are my carriers performing?  Are my vendors routing on time?  Where are all of my orders in the lifecycle?”), and help us get familiarized with the reporting you rely on already.  We’ll head for the specifics as necessary: your analyst will eventually need to know precisely which dates, reference numbers, and location information matter to you, but it is very easy to waste time answering the wrong question with a dashboard.

The same goes for the look of the dashboard itself: workflow should go from function to form, not the other way around.  An analyst can spend days making a dashboard pretty, but if you as the customer don’t care about the question the dashboard is answering, the work was for nothing. If, as a customer, you want to see a key performance indicator over time, or relative to another, make that clear.  Analysts tend to be literal people who will give you what you ask for.  Helping us understand the heart of what you’re asking lets us be of maximum use to you.

Data Integrity is Key

The Business Intelligence Golden Triangle is “Fast – Right – Pretty.”  Our primary goal as analysts is “right:” if you can’t trust the data, then what are we doing here? If you are a new customer with a complex data configuration, then expect at least a third of the development time to be solely dedicated to getting the underlying query right. Even if your analyst is building from a well-established data pipeline, there are plenty of obstacles that can stand between the data warehouse and a finished dashboard. This lead time only increases if you have a need to employ R or integrate advanced analytics.

Once “right” is taken care of, the choice between “fast” and “pretty” is up to you: if you want it now, your first pass isn’t going to be pretty, but it will be functional. If it has to be pretty out of the box, you’re going to get it relatively slowly.  However, neither “fast” nor “pretty” matter in the face of “wrong.”  Unyson’s analysts take pride in providing high-quality, trustworthy numbers.

Good Work Takes Time

New customers are always excited to get their hands on their dashboards, and with good cause. Allow me a moment to manage expectations: a day-one dashboard is a guaranteed disaster.  Why? Because like I said, data integrity is key.  Personally, it takes at least a billing cycle before I’m fully comfortable presenting a dashboard.

But why a whole billing cycle? That could be a month!  Two reasons: first, most dashboards are inherently backwards-looking anyway.  Second, after a billing cycle we should have an end-to-end example of your business process. While all the possible scenarios may not have played out, there is at least a complete example. Don’t think this means we only start when we have all the data we need; we need to (and do) build out prototypes.  But in order to maintain trust, we need to present a full slate of production data.

While UAT/Dev/Test data might be acceptable for building out Excel-grade reporting and the skeleton of the dashboard, it is nowhere near good enough to build a full product, let alone put in front of the customer. People tend to anchor on the first thing they see, and if we present incorrect numbers even a caveat of “the real numbers are going to be different” is not going to alter the first impression. This is also why making a dummy dataset and presenting that is a bad idea as well.  Our goal is always to be useful. We want to build useful tools that our customers actually use. Testing data has its uses, but acting as stuffing for dummy reports is not one them.

Get Buy-In From the Top Down

Every project has a top decision maker, and every dashboard has end users. The best-case scenario is that they are the same individual. We can get stuck in a very long, very painful “guess and check” loop without the opinion of the highest-ranking user. Not only does that clarify the specification, it helps us condition other users’ input. Part of the art of dashboarding is getting everyone on the same page.

The Golden Triangle still applies, but you as a customer can make “pretty” go faster if you present a clear idea of what you’re expecting to see. This is where it helps to share current reporting with us.  Quite often, we can collapse four or five spreadsheets into a single page on a dashboard.

As with everything, good communication is crucial.  Ideally, there will be a single point of contact filtering all of the end-users’ input and aggregating it into what we need to build. However, it’s very much the analyst’s job to gather, distill, and prioritize requests. A good analyst knows that the job is one of giving customers what you want and need, not necessarily what they ask for.

 

If you’re not acquainted with the power of dashboards or have any questions, I’d love to hear from you!

Andrew Roby

Business Intelligence Analyst II

Aroby@unyson.com