Tableau vs. Power BI: My Two Cents


At the end of August, Tableau picked a fight with Microsoft, posting a slideshow titled “10 Ways Power BI Falls Short” (http://www.tableau.com/compare-tableau-power-bi). Their ten criticisms were:

  1. Power BI limits visualizations to 3,500 data points, which can lead to missing outliers.
  2. Relatively simple calculations require learning Microsoft’s DAX scripting language.
  3. Power BI has limited ability to do trending and forecasting.
  4. Users are unable to “slice and dice” their visualization by more than two categories.
  5. Power BI’s filtering capabilities are limited as there is no “keep only” option.
  6. Users are unable to customize popup content by adding different dimensions and measures or changing formatting. With Power BI, you get what you get.
  7. Power BI cannot group data together on the fly, forcing you to do your grouping ahead of time in data prep.
  8. Users cannot use another person’s visualization as a starting point for making their own changes.
  9. No story-telling capability.
  10. Power BI does not allow the input of data on-screen, limiting users’ ability to perform what-if analyses.

Of course, Microsoft did not take this lying down, almost immediately responding with their own top 10 list of features available in Power BI which are missing from Tableau (http://mspoweruser.com/microsoft-responds-to-tableaus-criticism-on-power-bi):
  1. Out-of-the-box content packs for dozens of data providers, making it easy for business users to connect to data
  2. Dashboards that collect important visualizations from reports into what Netz called a “higher-level view.”
  3. Natural-language queries to explore data and create new visualizations.
  4. Enterprise-class SaaS BI, including connecting to on-premise data.
  5. A data engine Microsoft says is 10 to 100 times faster than Tableau’s.
  6. Community-created custom visualizations that use an open visualization platform (Tableau’s platform, Netz pointed out, is closed and proprietary).
  7. Native integration with Cortana, Excel and real-time data feeds.
  8. Automated quick insights.
  9. Integrated ETL (extract, transform and load) tool.
  10. A data model that supports large numbers of tables and more complex relationships between tables. “Tableau does not support more than trivial data models,” according to Microsoft’s emailed response.
  
My Evaluation
Clearly, Tableau is the market leader, but the fact that they are directly calling out Microsoft shows that there is some concern about Power BI. So, how close is Power BI to matching the capabilities of Tableau? How many of the above points are accurate and how many are simply overblown? I’ve seen a few comparisons of the products, but I figured the only way for me to really understand the pros and cons of each package is to analyze the situation myself.

I’ve written a number of blog posts over the past few months and many of the visualizations included in those posts were created using Tableau Public. So what better way to analyze the capabilities of Power BI than to attempt to replicate those visualizations with the Power BI. I should note here that my experience with Power BI is limited; I’ve played with the tool a bit and created some basic visualizations, but I’m definitely nowhere close to being an expert, so much of this analysis will be initial impressions and it’s likely I’ll miss something here or there. If so, feel free to let me know.

Maps
I decided to start with my “Nate Silver Challenge” election map. Tableau’s implementation of maps is far better than any other product I’ve worked with, so I was really curious to see how well Power BI handled them. The results were not bad. Take a look (note: I did not take the time to bring Alaska and Hawaii into the map):



So what did I find in this analysis? Here are a few observations:

  • Tableau uses the state name as a geographic identifier, but Power BI uses the state code. In my data set, this field was named “Code”, but I could not get Power BI to recognize it as a geographic entity until I physically named the field “State”. Strange, but an easy change, so no big deal.
  • In Tableau, you have lots of options to change the look and feel of a map. Power BI does not appear to have many such options. For instance, the maps show highways, which can be really distracting and often unnecessary. I could find no way to turn these off (though I should note that they do go away if you zoom out).
  • As noted in Tableau point # 6, I could not customize popups. For instance, I would have liked to put the state name on the popup instead of the code and would have liked to remove the “Count of Prediction” value from the popup, but the lack of customization capability prevented this.
  • Perhaps the most annoying issue was that the visualization automatically zooms to the level it feels appropriate. In my case, it ended up cutting off parts of the northwest, Texas, and Florida.
  • Finally, the maps take a bit of time to load whereas Tableau is almost instantaneous.
  
I then decided to get a little deeper with maps and attempt to recreate my county-level drug overdose map. This was a complete failure because of Tableau point # 1. The data set has information for 3,000+ counties for each year from 1999 to 2014, a total of over 50,000 data points. Because Power BI is limited to 3,500 data points, it only mapped some of the counties (seemingly in random fashion) and presented me with a small—so small, it’s easy to miss—warning in the top left corner. When I click on this warning, I get the following message:




Essentially, I am forced to filter my data in order to get it to properly display. I attempted to filter down to just one year, which should have resulted in less than 3,500 data points, but could not get it to work. This is very disappointing, but more importantly, could be very problematic to a visualization because it could, as Tableau noted, cause an analyst to miss outliers in the data set. It’s easily noticeable in my situation because of all the uncolored counties, but could be missed in a data set with say 4,000 data points.

Hierarchical Data
Next, I decided to recreate some of the visualizations in my post on visualizing hierarchies. In that post, I used religion data to create a treemap, a sunburst, and packed bubbles in Tableau. So, I attempted to recreate those in Power BI. Here's the result. Note that there are multiple "tabs" because, at this time, Power BI does not allow you to embed a specific tab only.




The treemap was easy to create, just like Tableau, but Tableau point # 4 came into play as the treemap is unable to display any more than two levels at one time. Instead, Power BI forces you to “drill down” to see the next level.

The Sunburst (or radial treemap) I created in Tableau is not an out-of-the-box visualization. Instead, I used a template created by another developer, formatted my data in the right way, then created the visualization. This was a somewhat difficult process. But, with Power BI’s community-driven visualization gallery (Microsoft point # 6), I was able to download a sunburst visualization and use it just like standard visualizations. This definitely made it much easier to create a sunburst. The results are not bad, although it does seem to have some trouble understanding that the levels of the hierarchy are not consistent across the board (some have two levels, while others have three), but I think there is likely a fix for this that I am just yet to discover.

Packed Bubbles requires a community visualization as well, so I downloaded “Super Bubbles”. I attempted to display all three levels in the hierarchy but it didn’t show anything. I then tried to display only the lowest level and was able to get bubbles, but no text and no popup information. Eventually, I chose just the first level and was able to get bubbles, text, and popup info. Granted, this is a community visualization, so it’s difficult to point at Microsoft for this issue.

I did see one interesting take on the bubble chart though, which is worth sharing. A community visualization called “Aquarium” creates animated fish, the size of which is based on some metric. So I used it with the religion data. See the video below (or interact with the visualization embedded above).


Honestly, I’m not sure of how much utility there is in such a visualization as there are many other ways to look at this data which are more insightful, but it is a really fun visualization and something that is quite eye catching. Perhaps more importantly, it shows what is possible with Power BI.

Word Cloud
Finally, I used Power BI to replicate my word clouds of Donald Trump’s and Hillary Clinton’s election speeches. This again required a community visualization.



The word cloud was really easy to create and actually had some additional capabilities over Tableau, including the ability to rotate words. But there were also problems, including the fact that hovering over a word provides no further information. I would have liked, for instance, to include the number of occurrences. But, again due to Tableau point # 5, I am unable to customize any popup text. I also noticed that it doesn’t handle contractions properly. In the case of “doesn’t”, it created two words, “doesn” and “t” and “here’s” became “here” and “s”. Again, this is a problem with the community visualization, rather than Power BI itself.

Some More Positives
Much of the above focused on the limitations of Power BI in comparison to Tableau, but there are definitely some really nice features. Here are just a few I encountered:
  • Like Tableau Public, visualizations can be published to a server and shared with the public. Visualizations can also be easily embedded into web pages, blogs, etc.
  • As I’ve discussed previously, Power BI has a great community of developers creating new and interesting visualizations and publishing them in the Power BI Custom Visuals Gallery (Microsoft point # 6). These can be downloaded, along with samples, and used in your own visualizations. There are lots of really nice visualizations in the Gallery including some of the ones used earlier in this post and others I intend to try soon—Waffle Chart, Chord Diagram, Sankey Diagram, and Force-Directed Graph. I can’t overstate how much I like this feature. It is much easier than finding (and trying to dissect) the custom work of a Tableau developer in order to create a non-standard visualization. The only slightly annoying thing with this feature is that you have to re-import the visualizations into each document, rather than them being permanently added to the toolbox.
  • You can do much more with Power BI’s data modeling tool than what’s available in Tableau (Microsoft’s point # 9).
  • Now this may seem like a less than spectacular capability, but I found that Power BI does a really nice job refreshing data from Excel. While this should be a straightforward thing, Tableau has an incredibly annoying inability to do this on a consistent basis.
  • And finally, its free! (at least for now)

Final Thoughts
When you’re used to Tableau, Power BI gives you this feeling of reaching for the cookie jar but it being just out of reach—it seems to fall just short in most scenarios. Some of its flaws are pretty significant—its limit of 3,500 data points, lack of popup customization, limitation on the number of categories which can be analyzed, inability to change the look and feel, slowness, etc.—but I feel these can issues can be overcome in time. After all, this is only version 1.0. And, considering that, Power BI is really a great start. That being said, it doesn’t come anywhere close to the capabilities of Tableau right now and Microsoft has a lot of work to do to catch up. But, to be fair, I have yet to find any other product that has the level of functionality as Tableau—the bar is set pretty high. Ultimately, I expect the battle between Tableau and Microsoft to lead to further improvements and innovations. And I look forward to watching how things play out.

If you’d like to interact with the visualizations discussed in this post, here are some links:

Power BI
 Tableau

Ken Flerlage, October 11, 2016

8 comments:

  1. Great summary article of the differences, one of the best I have come across in the comparison between the two tools.

    ReplyDelete
    Replies
    1. Thanks. I had looked for unbiased comparisons and couldn't find any, so I decided to do one myself.

      Delete
  2. Hello Ken, this was a great write up, especially the example dashboards where you could actually see the differences.
    Since it has been a while, and PowerBI and Tableau both have made some changes, can you please post an updated comparison as things stand now?
    Thanks

    ReplyDelete
    Replies
    1. I've considered updating this comparison considering the improvements made in both software packages. At the time I wrote this, I was fairly unbiased. However, I have since gone sort of "all-in" with Tableau and it would probably be somewhat biased of me to compare the two at this point in time.

      Delete
  3. Hello Ken, I've read your article with great interest, since we are facing the exact same question: Power BI or Tableau. Having played around a bit with Power BI I am not sure if all of your mentioned short comings are still valid, especially when considering that your work has been carried out almost 2,5 years ago and that Microsoft seems to have put quite some effort in improving Power BI.
    Would you be in a position to post your impression of the current situation in the battle?
    Do the short comings (either way) still apply in your opionion?
    Thanks!

    ReplyDelete
    Replies
    1. Yes, I'd agree that Power BI has come a long way since this blog, so its analysis may no longer be entirely valid. I don't really have any plans to update this, to be totally honest. I will say that I have since come to the conclusion that Tableau is the best tool available, for some of the reason above, but more importantly because of the platform that they've built. They've built a platform that allows you to create anything you can imagine, without the need to write code, and with a consistency that applies across the entire platform. Happy to discuss further if you'd like to reach out to me directly. You can email me at flerlagekr@gmail.com.

      Delete
  4. Hi Kevin, this looks a bit old now and lot might have changed since then. Do you happen to have a new analysis between the two for versions in 2021.

    ReplyDelete
    Replies
    1. Yes, that's true. Both tools have grown quite a bit since I wrote this. Unfortunately, I don't have any intention of comparing the tools again.

      Delete

Powered by Blogger.