What's Wrong with Pie Charts and 15 Top Visualisation Tips

Dashboards Are Dead, Long Live Dashboards

The push for visualisations in modern BI products means that more and more products are available to make our data look pretty and engaging. However one analyst’s taste for bright colours and crowded dashboards might not suit the engineer’s taste for simple outlines and clear labelling, while the marketer’s taste for neutral tones and bold graphics would be fulfilled by neither.

I want to look at some simple guidelines for data visualisations that will provide consistency and ease of interpretation to your dashboards and reports.

Which Chart is Good for Which Job?

If you’re not a statistician or mathematician then you probably don’t work with charts every day of your working life and haven’t sat through hours of lectures on origins of axes, logarithmic scales, trend line computation etc. You just want your reports on sales, customer engagement or service usage to look appealing and get the core message across.

Understanding what your message is, is key to designing a good visualisation:

  • Are you trying to say something has got better or worse over time?
  • Are you wanting to show these baked beans sell better than this other kind of baked beans?
  • Are you describing cohorts and their differences?
  • Do you have many different categories to display or just a few?
  • Do you want to emphasize how a few data points are better than average?
  • Have you got a story to tell, or just one big number to be proud of?
  • Do you want to focus on the details or is an impression of scale more important?

The answers to these questions will help you chose which type of chart is best to display your narrative, as some are better than others for all these circumstances. I’m going to discuss a few basic scenarios and make suggestions for how to display them, and I’m going to discuss why some people say pie charts are useless, and why some people just can’t do without them.

Trend Analysis

A classic need for any analyst is to display a chart of performance over time. This could be sales performance, service use, costs, profit – anything that is measured at distinct points in time and subsequently can be compared to its historical values.

Many people will instinctively know that the right chart type for this analysis would be a line chart:

English speakers read left to right and it is part of our psychology that we imagine time on a left to right axis as well. A line chart with time along the x axis feeds into this shared concept and we can automatically agree on what the chart is telling us.

However we can still improve on this basic concept with some more guidelines:

  • It is the direction of the axis that is most important for trend over time analysis, not the actual chart type – a column chart can still represent time
  • Lines that connect data points imply a relationship between them – as if the previous month’s result had an impact on the next or directly led to it. A continuous line may give the impression of continuous samples and imply data points that do not even exist. A column chart breaks this idea up a bit and shows individual data points as ‘stand-alone’. A good compromise is to add indicators to the line to show the sample points.

Additional analysis can be added to show average performance via a calculated line, algorithm-based trend lines to smooth peaks and predict future performance, multiple lines to compare categories over time, or area charts that shade in the space under the line to draw greater attention with block colour.

Proportional Analysis

This is where you are comparing cohort sizes with each other. In these charts you want to give the correct impression of the size of one population compared to another. It is very easy to misrepresent data with the types of chart many analysts will chose for these datasets – they can be hard to read and easy to over simplify.

A pie or donut chart (the pie with a hole in it) is a classic visualisation that we can all recognise and we all think we can read. However there are a few problems with it as an analytical tool.

  • They are very difficult for us to judge comparative sizes from – without labels can you tell which is bigger out of Nottingham and Cambridge in the chart above?
  • With too many categories they become unreadable – the slices are too thin and data labels overlap.
  • As they display proportion they give no indication of true-values – Bristol looks like it has outperformed all the other cities, but how much revenue is that exactly? You could add data labels, but without an axis for comparison this is still hard to interpret at a glance.
  • There is little scope for additional analysis such as average-lines or added measures, unless you turn the pie 3D, and even then the height of one slice might obscure one behind it:

All of this is not to say they should never be used – they should just be used carefully. They are attractive charts that draw the reader’s eye and can make bold introductory statements on a report as long as they are backed up by more specific and detailed visualisations later.

Newer forms of proportional visualisation can also offer the same attention grabbing, more impressionistic storytelling. Tag clouds will be familiar from websites where popular search terms are often displayed using them, but they can also represent sales data for example:

Heat maps or tree maps are similar to pie charts, except the pie tin is rectangular and the slices are a geometry teacher’s nightmare to calculate:

Category Comparisons

After time-trend analysis perhaps the most often required chart type is one that can compare performance across categories. You will have different products, different sales people, numerous teams, separate locations – but you need to see which one has done best, or worst.

This is where bar charts, or maybe funnel charts or radar plots can help. As we discussed above, left to right axes should be used for timelines, our brains want to read historical progression on a horizontal plane. A vertical axis helps to enforce the fact that the displayed categories are not time-related.

The bar chart has a few extra benefits compared to a column chart:

  • It is easier to read the dimension axis labels as they won’t overlap no matter how long they are, and will be horizontal, not at a crazy angle.
  • Bars can potentially be added to tables, whereas columns cannot
  • Data labels can be added which again won’t overlap into other sample points
  • Overall they make more efficient use of space.

Stacked and mixed bar charts add flavour to the concept and you can pack a lot of information into a small space:

One thing to watch out for is whether your BI tool of choice ranks bar charts with the highest values on the top or the bottom of the dimension axis – my preference is always for at the top.

Funnel charts and radar plots can also help to distinguish performance between categories with added relationships to either multiple axes or a narrative element to the data:

More Complex Scenarios

Some more standard business visualisation needs are contribution analysis, distribution/correlation, portfolio analysis and outlier analysis. These kinds of questions need specific tools to answer them with, and there are charts for each:

  • Waterfall charts or ‘bridges’ show contribution of successive values to a target, giving a feeling of progression (without time being a component on an axis) – which justifies their horizontal orientation.
  • Bubble charts enable you to show distribution over two measure axes, like a scatter plot, with the addition of bubble size to give a third value indicator. These can be useful for correlating two or more sets of values or analysing a portfolio with multiple indicies:
  • Box plots are charts which are very useful for monitoring statistics that change over time or categories that have multiple values – classically stock-market prices, but also potentially waiting times or levels of contaminants in water samples over time. They show the maximum value, minimum value, first and third quartiles, the median and any outliers.


The layout of a group of visualisations in what you could call a ‘dashboard’ is more of an art than a science. There have been studies that aim to show where viewers’ eyes look first on menus, websites etc which might give you tips on where to place your most important facts and analyses, but these studies are inconclusive and the fabled ‘menu sweet-spot’ of where to put your most expensive dishes has turned out to be untrue.

I will generalise for an English-speaking audience and say that most of us will naturally want to start digesting information by looking at the top left of a document or screen and continue to read from left to right and down the page from there.

This suggests that placing top level KPIs or totals, or the main message of your visualisation in the top-left position will give it prominence. You can then lead your audience through your visualisation’s narrative with successive charts and numbers placed on the page leading down and to the right from there.

As you get further from the starting point you might make your information more detailed and drill into the figures. Ideally each successive visualisation ought to add to the overall picture in a coherent fashion.

If you need to discuss two or more unrelated topics in one document or dashboard, then ensure there are clear demarcations – use a new page or draw literal lines around the unrelated information.

Negative Space

Negative space is an intuitive concept that helps dashboard designers, sometime even without them realising. It is the ‘white space’ around and between objects that helps to keep things readable and clear. Paragraphs feature negative space in the indentation and line separation that tells the reader a new subject or concept is starting.

Newspaper columns have negative spaces that separate stories or keep one whole article in multiple columns without us reading from one to the other accidentally.

You will need to handle negative space around charts and keep it relatively free from clutter – ensuring legends and titles do not overlap or appear to close to an unrelated graph.

Grouping your visualisations into columns or rows using negative space (or the odd simple line or box) tells your audience what objects should be considered together, and complement each other, and what is separate and bears no relation.


Your organisation has probably already spent a considerable amount of time and money on design and design assets. You will likely have a logo on your website, perhaps a colour scheme for corporate communications, and maybe even a corporate font.

Using these elements in your dashboard will help to breed confidence in your output – it will be trusted as part of a continuum of information instead of a hastily thrown together analysis. Using the corporate colour scheme or a logo helps people see that the information belongs to them and their efforts contribute to what is on the page.

If you don’t have direct access to the marketing or design department of your company, you can probably still re-use the assets from the website with a simple right click/save picture on the logo or using the ‘inspect elements’ (F12) option to trawl for images. Even the venerable Paint application on Windows can help, with the colour picker tool telling you what RGB values your website uses so you can replicate them easily.

Bringing It All Together with a BI Tool

Now that you have an understanding of different chart usage and some tips on dashboard design you can start to create some masterpieces in your visualisation application of choice.

SAP have a selection of great tools for doing this, from the new SAP Analytics Cloud offering with intuitive and predictive design options, to Lumira Discovery and Lumira Designer which offer multiple sophisticated options for visualisations and applications, and Web Intelligence which can produce great looking dashboards with interactive filtering, linking and navigation.

DSCallards can also offer you support with your Microsoft Power BI reports, offering courses on the desktop tool, query design, data transformations and data modelling.