Data Scientists are expected to develop dashboards and reports
Data Scientists are expected to develop dashboards, reports, and visualizations. Visualization is the routine of showing the final calculations, clues, and predictions.
Unfortunately, many times Data scientist fail to deliver a good dashboard or report that will be broadly used by their audience.
I keep on hearing frustrated comments how the dashboards are too complicated, they aren’t using precise terminology, there too many filters and the reports take too slow to load.
While talking with the audience, I often realize that many users of the reports and dashboards give up, frustrated from waiting for the report that never loads.
Data scientist users can often ignore their tools and reports.
Instead of using the tools provided by their data scientists, the audience often opts for creating their version of a report or their data pull, that works for them at the time of creation and it allows them to finish their task fast.
That is bad. If that trend continues, it turns out that Data Scientists are going to end up creating tools, dashboards, and reports that are not used by anyone. All the effort and time that data scientist spends on building those tools and dashboards is spent for nothing.
To overcome that, there are some excellent practices some good practices.
Good practices to develop highly usable dashboards and reports as a Data Scientist
No matter what tool you are using for your data visualization, Python, R, Tableau, Adobe, Marketo or Qlikview, the principles are the same: keep it as concise and straightforward as possible
Who is going to use your report?
The most crucial question is: Who is going to use my report?
Do you build this report for sales, marketing managers, field marketing, C- level people, IT, support or maybe HR?
It is essential to identify the audience that should use this report. I see a lot of data scientist building reports without considering their audience. They say, everyone should use this report, but that’s hardly the case.
Once you have successfully identified your audience, you should try to stick to their way of looking and working with information.
For example, a marketing manager will not want to look at data in the same way as your C- level audience.
Try to avoid generalized reports and dashboards.
Create a story
Everyone loves to read interesting stories, especially if they are supported by great pictures.
You want to tell a story with your dashboard and reports in it.
Imagine you are writing a really interesting story about a new discovery you just made using your favorite machine learning model.
Write a couple of sentences for each of your findings
Now, try to translate that story on a whiteboard or paper.
Make data summarization
Instead of having all raw numbers displayed in a report, try to keep everything on a summary level. Even better, hierarchical summary.
Let’s say you have to develop a financial report. You can have drill down of fiscal hierarchy: Fiscal Year > Fiscal Quarter > Fiscal month. Starting off with Fiscal Year
Similar if you are developing a report that can be summarized by geography.
Have great visualization
Once you make successful summarization, you can put great visualization.
Nowadays is easy to make animated visualizations. I strongly suggest you go for it. People are already bored by looking at numbers and plain charts. They are easily distracted by something that is moving, looks cool and it’s flashy. This is one good way to keep their interest longer on your report or dashboard.
Simply, you want to make your users play with your charts as much as possible.
By doing that, your users will be more attracted to your reports.
Give more explanation in your reports
It is really easy for a normal user to get lost in your report.
Yes, as a data scientist, you will ask yourself how is that possible. this information is so clear, simple and straightforward to read. But, remember, not everyone thinks like a data scientist, and not everyone understands your logic.
That is why it is crucial that while developing your report, you put as much as explanation possible.
Explain what your dashboard is about
Give a concise summary of what kind of information can be found on your dashboard. Giving just a title is not enough most of the times. It really helps when there is a short intro about each kind of information the dashboard can provide.
Give a short explanation for each of the graphs
Altho graphs should be self-explanatory, often helps to have some text that will give extra information. Explain about your metrics, what the numbers represent, why you choose specific techniques.
Remember, not everyone is number savvy as you are.
Include information icon
Again, to avoid the clutter of all extra information, a good practice is to add small info icon next to the chart that you want to explain. Click or hover this icon should show the new window with the additional explanation.
Give less filtering options
Filters can be really a nightmare in a dashboard and report.
I can’t count the times when I saw filter overload on dashboards.
Filters can make your report useless very fast if they aren’t organized the right way.
If your report suffers from a filter overload, you might want to consider redesigning your report and most probably separating your report in multiple reports.
Use precalculated functions for displaying data
Reports can be slow for many reasons, but most common one is that the report contains too much data in the background.
Consider moving only the needed data in your report.
Create functions or procedures on the database level to restrict the amount of data. For example, consider creating one procedure for displaying Sales by geography, whereas input parameters you will have fiscal period and geography.
Arrange the report in sections
Make sure to divide your report into more logical sections. A good idea is to have this sections divided with different colors of dividers if possible.
Don’t mix apples and oranges together, you will confuse your user even more with displaying too much information without proper division.
This is few but really important notes that I found really helpful in building my dashboards and reports.
I realize that most of the data scientist I talk to on conferences and meetups have the same problems on daily basis, so I hope it will help you too!
Furthermore, I have already written an article that should give you more idea of How can I use the data?