top of page

"Home experiments" and stacked bar graphs

If someone asked you which graph would be best for visualizing the customer types for each account manager based on the table in the picture, what would you suggest?

It is likely that you will offer a stacked graph like the one shown in the picture.

Let's look at what motivated our choice and whether it is indeed a graph that will convey accurate messages to those who read it.

Here are some reasons why this type of visualization is preferred:

1. The graph on the right is one of the "recommended" graphs provided by tools (both Excel and Power bi).

2. It seems logical: there is a height comparison between case managers, clear separation between client types and the graph looks colorful and appealing.

Now let's look at whether the graph really provides insights and answers questions.

Look at the graph again and answer the following question:

Which account manager has the most at-risk clients according to his portfolio and how does he differ from others?

You will probably be able to answer the question, but it will be challenging, it will take some time, and some of us might be wrong.

What makes this "pretty simple" graph challenging?

1. Since all parts except the first do not start from the same point, it is very difficult to compare their sizes. Here is an experiment I did on myself: On the board, I arranged five paper rectangles of different lengths, one below the other, each starting at a different point. Trying to estimate the sizes of the rectangles, I could only identify the smallest one for sure. My biggest surprise was discovering that even the rectangle I thought was the longest wasn't the longest. In order to understand the size rating, I had to arrange them a second time from the same point.

2. The process of processing excess colors involves a lot of work, beyond that, the colors have meaning. Their use in multi-stacked graphs is random only for separating data, raising unanswered questions, for example, why are repeat customers in orange? What's wrong with them?

3. Our short memory is indeed short, so most of us will have trouble remembering all the categories of colors. Due to space constraints, this information can only be found in the legend. So, reader has to play "ping pong" to figure out what color belongs to which category. These are additional resources that the brain needs to read and decipher the graph's message.

What is an alternative?

An option is to use a "table graph" (like in the picture below) that clearly separates customer types. Warning colors should only be used for customers who are at risk and need special attention.

Using this visualization, it is possible to compare portfolios by customer type and identify areas of improvement in seconds.

POWER BI's dynamic features also let you sort each item by customer type and see how the picture changes.

* I created all graphs using Power BI.


2 Kommentare

Guy Glantser
Guy Glantser
28. März 2023

Rita, I agree with you about the challenges imposed by stacked graph.

The main problem with such use cases is that you have two dimensions (i.e. account managers and customer types), and you need to decide on which dimension you would like to compare the measures.

Do you want to compare between customer types of each account managers? Or do you want to compare between account managers for each customer type? These are two different business questions.

A stacked graph tries to answer both questions with a single visual, and usually the result is not being able to answer any of the questions easily.

Your suggested of a table graph is definitely better, but I think it's stil confusing. Because…

Gefällt mir
Rita Fainshtein
Rita Fainshtein
28. März 2023
Antwort an

Thanks for your comment, Guy. You're right, one graph - one question / message. Measures on graph can be expressed in monetary terms, quantities, percentages, etc.

Based on the question I asked in the example, I ranked parts (percentages) of customer types. If I had asked the question differently: Which account manager has the most at-risk clients, and where are the gaps compared to others? I would build the graph based on absolute values. Almost every data set had several options of messages and corresponding graphs.

There is always a translation of the specific words in the graph. Semantics play a large role, a graph is based on words, not numbers :)

Gefällt mir


Get New posts delivered straight to your inbox

Thank you for subscribing!

bottom of page