By Mico Yuk and Anita Gibbings
Ask yourself this question, “What’s the first step in building a successful business intelligence (BI) dashboard?”
If you’re like 95 percent of our readers, you said, “Scoping, of course!” But you’re wrong!
The starting point of any successful dashboard is identifying the who, what, and when. This planning phase is actually before scoping – and, unfortunately, it’s one of the most overlooked phases in the dashboard development process.
To help you get started, Mico Yuk, Founder of the Xcelsius Gurus and SAP Mentor, is offering a free download of her BI Dashboard Planning Guide, one of the 10 templates she designed to help you avoid some of theses pitfalls.
- Who: Who is the dashboard for, and who will pay for it?
- When: When is the dashboard due? Is there a major milestone attached to it?
- What: What problem will it solve, and what high-level key performance indications (KPIs) do you need to track to reach them?
In this article we will focus on “what.” As part of the planning process, “what” is the second most important part of the dashboard planning process after “who.” Why? Some companies spend tens of thousands of dollars, with good reason, to hire management consultants to help them identify and analyze KPIs and transform their business. The real problem? Most companies simply don’t put enough effort into determining their KPIs.
Often companies use generic KPIs, like revenue, costs, and margins, which are lagging indicators that don’t always result in business transformation. You might notice a drop in revenue, but unless you identified the right metric, you might not realize that you need to revolutionize the way your sales people handle their accounts. A dashboard with revenue by product would only indicate a pre-existing problem; whereas leading-indicator KPIs would likely track where sales people spent their time or how long accounts were spending in one stage of the sales cycle. This type of information would allow you to correct the issue before it had a major impact on revenue.
Depending on which KPIs are selected, a business can take an innovative path as individuals respond to what is being measured. This is similar to the company that wanted to grow sales and began tracking the number of net new customers. Without realizing that many of them were actually consuming more resources than revenue creation, the company’s margins dropped.
Successful Dashboard Stories
A successful dashboard should tell a four-part story for each of KPI:
1) Current State: Where you are today
2) Trend: How you got there
3) Forecast: Where you’ll end up over time at your current run rate
4) What-if: What you need to do to hit your targets
Sales Dashboard that “Tells a Story” – Credit: AnalysisFactory.com
Note the storyline at the top right: Scorecard, Performance, Trend, What If that allow you to first recognize that you have a problem, diagnose the issues and decide on the appropriate action.
Click images for larger view
Another thing to consider is that most KPIs are useless without the use of alerts or statistical functions such as probability. Successful dashboards contain KPIs that provide direct answers to questions like “Will I hit my target?” Instead of showing this in the form of a bar chart that displays actual vs. target, consider using a simple probability formula that displays a single percentage to target on a real-time basis throughout the month.
For instance, if it’s the middle of the month and the probability of you hitting your target isn’t at or above 50 percent, then a yellow alert should display on the dashboard for that KPI, indicating that unless you implement some changes to your business now, you’ll miss your targets at month end. This is way more efficient for running a business and measuring the appropriate targets as opposed to simply displaying data.
So before you bring in IT, business leaders need to spend significant time getting to the core business goals or challenges. This allows for a clear understanding of the purpose of each metric and how it will drive behavior in your organization once it’s published broadly.
Mico Yuk – Founder of the Xcelsius Gurus and popular weblog EverythingXcelsuis.com , SAP Mentor, and BI Influencer, will be teaching the above strategies and many more in her new online coaching series titled “The BI Dashboard Formula,” which will kick-off in September 2012. Don’t miss her early bird specials (including a free Blackberry PlayBook for the first 50 that sign up).
Register now and watch for the next blog post in August, covering the second step of Mico’s seven-step formula.
The Key First Step to Successful BI Dashboards,
RSS Feed
June 18th, 2012 at 8:09 pm
No, the first step to successful BI dashboards is to tell the truth with your graphs. I’m flabbergasted that you would publish this article with two bar chart that don’t start at zero, but at 20 and at 515. For a bar chart to be effective, and honest, we have to be able to use the length of the bars to make rapid comparisons. Starting the bars at anything other than zero wrecks its effectiveness.
Really, after all this time, you still chose to ignore Stephen Few and Ed Tufte?
John C. Munoz
June 25th, 2012 at 12:57 pm
Best practices are guidelines, not divine messages from the priesthood… Personally, I prefer something like “bars should generally start at a natural reference level, which is not always zero” – but implying the chart that doesn’t follow this is “dishonest” and “ineffective” is an exaggeration that implies a dim view of the viewer’s “chart intelligence” (which may or may not be realistic!)
August 13th, 2012 at 8:47 am
@Timo, why exactly do you want to break with single most important reason that makes bar charts effective? Unless you have a great reason to do that (and your rule is confusing, not clarifying), then don’t do it.
You stand by this example because you were/are SAP’s 8th employee, or because you think it’s an outstanding example of SAP’s visualization capabilities, or both?
And as for your “chart intelligence” reference, I’m confident that mine is quite high, but I’m not confident about yours. The 3d pie chart on your home page says a lot.
August 28th, 2012 at 11:29 am
Timely/timeless discussion by Naomi Robbins, PhD. on why bar charts should always start at zero. See link below.
You want to start your axis at something other than zero? That’s fine, just don’t use bars, there are better alternatives that are truthful to your data.
http://onforb.es/U8G8H0
August 29th, 2012 at 9:44 am
John,
Zero is in many ways as arbitrary a number as any other. Let’s say I have temperature data. I can show it in Celsius, or Fahrenheit. Why is my Celsius chart (which starts at zero = 32 degrees F) less “truthful” than the other one?
The Forbes article says “Not including zero in the baseline is one way to exaggerate a trend” and “Some graphs for the general public might be misinterpreted without zero.” I completely agree with both of those, but that’s very different from the “bar charts should always start at zero” (why not Kelvin, while we’re at it?!)
August 29th, 2012 at 11:38 am
Just to clarify, those are not actually bar charts, but more accurately combination charts.
Best practices are should be always be considered and used if they fit the business need. I personally would rather not look at a chart where the bottom 80% is always constant. I can find something else to display in that space.
I also think it is important to note that business dashboards aren’t typically targeted to a public-facing audience and that the people who actually use these charts do understand them. This isn’t misleading or incorrect.
August 29th, 2012 at 1:45 pm
I’m with Jamie on this, and widen the thought. We forget the information consumer’s (users) business needs and their and their abilities to interperate quantitative information at our peril
If we define a dashboard in the context of the examples above then I feel that starting the scale at Zero limits the information consumers ability to interperate the gravity of the differnces between bars “At a glance” which I think we all can agree on is a fundemental aspect of a “Dashboard” visualisation.
If we utilise screen real eatate along the lines of Tufte’s Data Ink ratios, I would ask why waste the ink when it delivers minimal value to the reader.
August 29th, 2012 at 4:05 pm
@Andrew & @Jamie,
I’m not saying ALL charts should start at zero, just bar charts. Why? Because we use the length of the bars to make comparisons, and lengths start at 0. Line charts or dot plots don’t need start at zero because, as Dr. Robbins tells us over at, http://onforb.es/U8G8H0
“We judge the value of a point on a line by its position along a scale. This judgment does not depend on zero. Even if you think you judge a bar graph by the position of the top of the bar, you cannot help but see the length of the bar.”
And @Jamie, putting a reference line on a bar chart doesn’t mean you can start the axis at 20 (as the example above does). It just means you’ve given your readers something to compare to, which is great. A better alternative would be to use dots rather than bars, and that would allow you to start your axis at somethig > 0. See my posting at http://bit.ly/JIJwUL that covers a similar topic, that is, why you shouldn’t log scale your bar charts.
August 29th, 2012 at 4:23 pm
Readers compare heights in bar charts. It’s a matter of perception, not of “chart intelligence”. “Chart intelligence” may tell you to pay close attention to the scale, but that’s not the point.
Most of the time, this means that you should start at zero, but the rule is more general: the heights must be proportional to the data, and that doesn’t happen in the example above.
That said, if variation is low you end up wasting precious screen real estate. In this case, the solution is not break the scale or start at a value different from zero. The solution is to use a different chart.
August 30th, 2012 at 12:34 am
@Timo:
I believe you are not disputing that bar charts should as a rule of thumb and as of best practice start at 0.
Bars do encode the value by length, therefore not showing the complete length invites, almost forces a user to misinterpret the data, when comparing bars to each other.
Your point about temperature charts is taken, although I typically would represent temperature (which is often shown across time) as a line chart anyway.
Why the Americans use Fahrenheit is a mystery to me anyway ;-) I think 0 degree Celsius (=when water freezes under normal conditions/pressure) and 100 degree Celsius, when water turns to steam (under normal conditions) makes more sense, but.. I am European (from the continent) after all, and we use the metric system.
August 30th, 2012 at 2:17 am
As an alternative to the bar chart with having a constant 80% bottom, you could use a data point chart. That way you don’t waste space and fix the ‘issue’ of the Y-axis zero starting point.
For people not familiar to this type of chart, it will only take a minute to explain. Issue solved.
I must confess I do apply many of the concepts of Stephen Few in customer dashboard projects but the most important thing to me when designing dashboards or other data visualisation solutions, is that I can say that I’ve thought well about the choices I’ve made and have looked over alternatives to find the best solution in which ‘science and art’ are balanced.
In the end, you need a dashboard that is going to be used by the users. If you have a Few and Tufte ‘proof’ dashboard, but is less engaging to the people who need to work with it, resulting in a low usage rate, you don’t achieve your goal. I tend to throw in some art, without breaking the laws of Few and Tufte but by challenging and stretching them now and then.
In the end, I think it’s good to have invested in knowledge on data visualisation, which can be provided by books from for example Few or Tufte, so that you can make a well considered decision instead of just throwing stuff random on the canvas, just because it looks nice or someone asked you to.
In practice, I’ve never been unable to convince users about design choices, mainly because I have thought well about it and am able to back them up.
One final note: I never use pie charts :)
August 30th, 2012 at 2:53 am
It might be possible under special circumstances to make a case for *not* starting at zero, but both of these really should – particularly the right-hand chart.
The right-hand chart doesn’t give you anything that the table underneath doesn’t do better. The same 2 data points are there, but the table at least does a better job of showing the relationship between Actual and Goal. The chart at best is redundant, and if we’re following “data ink” principles, ought to go. Personally, I think it’s misleading too.
The left-hand chart isn’t quite as bad, in that it does depict ups and downs from one period to the next, which isn’t quite so easy to pick out from the associated table. But the chosen origin is only 20, so why not make it start at zero anyway?
To be fair to the author, you do see this all the time in BI dashboards, and I suspect the software has been set up with some generic rule that causes it to pick its own origin based on the data associated with the selections. It’d be nice if the graphing routines wouldn’t allow a non-zero origin for a bar chart, but it would be a brave software company that would impose that on its customers.
August 30th, 2012 at 4:24 am
At this link, http://yfrog.com/h6jlsep, you will find a bar chart of maximum monthly temperatures in London for the last three years in degrees Kelvin.
Kelvin is the scale with a true zero for temperature so I guess the “must have a zero” rule means this is the right scale for this chart.
I can’t help feeling that in this case the rule is wrong, perhaps there are others cases.
IMO the issue is not with the rule but with its over-zealous application.
Interestingly, over-zealous application of rules seems to be a speciality of some members of the visualization community.
August 30th, 2012 at 10:55 am
looks like my previous post from today was deleted…
anyway:
@Timo:
I believe you are not disputing that bar charts should as a rule of thumb and as of best practice start at 0.
Bars do encode the value by length, therefore not showing the complete length invites, almost forces a user to misinterpret the data, when comparing bars to each other.
Your point about temperature charts is taken, although I typically would represent temperature (which is often shown across time) as a line chart anyway.
Why the Americans use Fahrenheit is a mystery to me anyway ;-) I think 0 degree Celsius (=when water freezes under normal conditions/pressure) and 100 degree Celsius, when water turns to steam (under normal conditions) makes more sense, but.. I am European (from the continent) after all, and we use the metric system.
August 30th, 2012 at 1:33 pm
@Steve,
Brave and unbrave software companies alike enforce zero baselines for bar charts. Tableau, SAS, Excel and Qlikview all default to a zero baseline for bar charts. I’m not a SAP user, but I suspect they default to a zero baseline too.
For the example above, the creators of the charts made the unwise decision to start their bars at 20, rather than 0. By doing this, they’ve made it more difficult for their audience to make comparisons in the chart (how am I doing compared to goal, how am I doing compared to month X?) Good chart makers spend a lot of time thinking how to get their point across. Violating the bar chart’s main strength should never be done because there are better alternatives out there, not because we should follow some rule blindly.
By setting the baseline to 20 rather than zero, a value of 60 looks 4x larger than 40, not 50% larger. Might as well start calling orange yellow and expect our audience to adapt.
Again, this rule only applies to bar charts because we use the length of the bars for rapid comparisons, and lengths have a zero baseline.
Finally, I’m curious what the folks on this thread think about the example above, not just the bar charts. Is the example above something that, you, my fellow commenters think is a good example of how information should be displayed?
August 31st, 2012 at 12:22 am
It’s nice to see that more and more people care about the effectiveness of dashboard design and there should be room for more points of view and those views should be respected.
If the topic of this post was a dashboard design competition or ‘rate this dashboard on design principles’ we could further discuss this but right now, I’m leaving the arena.
A great weekend to all of you!
August 31st, 2012 at 1:01 am
John,
I am thoroughly confused now. I think people are saying that line charts are okay not to have a zero but bar charts must have one. This seems odd given the obvious issues with the line charts in the Forbes article http://blogs-images.forbes.com/naomirobbins/files/2012/08/Huff.jpg
this clearly suffers from the no-zero issue.
In fact I find the whole Forbes article confusing but worst of all it contains the statement :-
“We judge the value of a bar by its length. Any length begins at zero. We judge the value of a point on a line by its position along a scale. This judgment does not depend on zero.”
but there is no research quoted to back this up. Given it is such a key point in the article some evidence to its accuracy should be given. Particularly given that just three lines above seems to be a line chart suffering from the very problem it is supposed not to suffer from !
August 31st, 2012 at 8:13 am
@John
No, I’m with you on the whole “bar charts should start at zero” thing. But while applications may well default to a zero baseline for a bar chart, that’s not the same as enforcing it.
(Incidentally, I just made a simple column chart in Excel, and it didn’t even *default* to a zero baseline, it chose its own non-zero value to start the vertical axis. Bad Excel.)
September 4th, 2012 at 4:34 pm
@Donald MacCormick:
I think most important is to try to understand that with bar charts indeed the values are encoded by length, same as values in pie chartes are encoded by area (or angle). Once you agree with this stateement it makes sense to start a bar chart at zero.
I have done a couple of presentations about “DataVisualization and dashboards” showing a non-zero scale bar chart, and I asked the audience to compare one bar to another (as % of value of the other bar that is), most audiences failed, which to me is not surprising as it shows that people expect a bar chart to start at zero.
Now, leave out the funny temperature discussion about Celsius, Fahrenheit and Kelvin.. for now ;)
September 5th, 2012 at 1:25 am
Andreas, I agree that bars should usually start at zero and if they don’t then they can lead to misleading results. I also believe that the same is true for lines and possibly for dots (but I have done no research on it so cannot say for sure). The “funny” temperature example is not meant to be funny, it shows that there are times when these rules do not apply. In fact that is my only really point here is that the dogmatic application of rules and the was against “sexiness” both get in the way of effective dashboards. Before considering visualization “rules” there are two things you should consider beforehand :-
1) What are the business drivers behind this dashboard ?
2) How can I maximize the use of my dashboard (unused dashboards are totally ineffective)
Both of these can lead you to breaking the so called “rules”, but to great effect.
September 6th, 2012 at 8:42 am
Another thought experiment: golf. You could chart total number of strokes, or the number above/below par. St. Andrews is a par 72 course. Is the two-under-par golfer (70 strokes) twice as good as the one-under-par golfer? Or twice as good as the golfer that took 140 strokes? Does the vision ability to compare bar lengths really mean anything — if it’s twice as long, does that mean it’s twice as “good”?