Not Everything Needs to be a Dashboard

I want to make this clear at the top. Dashboards are super important. They are often the product of a data scientist or data analyst’s work and the medium through which value can be scaled and repeated.

But even when you’re driving a car with a great dashboard, you still want to stick your hand out the window and feel the wind.

In data science, you need to have a feel for your data — you need an open environment to go into your data and just try things.

Clean. Pivot. Filter. Dirty. Graph. Pivot again. Delete everything. Start again.

The order does not matter, but the ability to freely explore your data, unencumbered by the scalability considerations of your process, is an important part of data analysis that I think gets overlooked by some of the premiere data tools.

In a previous post, I likened this process to jotting down notes on a blank piece of paper. This is crucial for exploring ideas, and would work much worse if all paper had mad-libs on them already.

In talking with people in the data field, I have found that Tableau showcases this problem. Yesterday, I was talking to the head of an FP&A team, who described Tableau as “over-productionized.” Because of this they are turning to Jupyter Notebooks for their team to do ad hoc analysis on data from their snowflake database.

I have also spoken to many people who use Tableau as a means to pair down their datasets so that they can fit into Excel. To me, this workflow exposes two things:

1. Tableau does not provide a flexible enough environment for scratch work, while Excel does.
2. People need to be able to do this scratch work on datasets that do not fit into a normal spreadsheet.

These two points are the impetus behind Mito, a spreadsheet interface for Python, which allows you to complete the flexible workflows that exist in a spreadsheet on large datasets that would not normally fit into a spreadsheet. Every edit in Mito generates the equivalent Python.

I have described Mito more in detail in previous blogs.

I hope this notion of data tools neglecting the power of scratch work resonates with you. It is something I think about a lot and plan to explore more.

I’d love to hear any thoughts in the comments or shoot me an email: jake@sagacollab.com :)