Product Design

After Effect
Illustrator
Sketch
Edge Animate

Palantir products' analysis process had always been very linear, users have to keep drilling down to create a smaller subset of data. When I first joined Palantir and their map team, I noticed that users have to go back and forward to test their hypothesis. They can also easily lose their analysis if they accidentally click on the blank part of the screen. 

In December 2013, we had to ship a new version of Raven (our map product) to a completely new vertical client - insurance risk engineers. They want to user Raven to understand causations of huge policy payouts so they may price their policies better in the future. 

 
 

Design Problems

  1. How can we make it easy to explore data without losing the established context the user already created?
  2. How can we make these filters transferrable?
  3. How can we make it easy to add, remove, enable and disable filters without confusing the users.

 

 
 

"Balance the ease of exploring data without losing the established context"

I want to help our users to develop an intuition,
and help them to set up their experiment to understand causation.

 

 

First iteration of filters

Creating sets of filters to compare and contrast

 

Breaking

 

Dual-selection system

Some designers were against the unified selection system, because it maybe be overwhelming when there are a lot of data. So they initiated a dual-selection system where if you select something on the graph, the histogram will update to only the things you have selected. Because you have a new histogram, you can click on certain values, and that would introduce a new state called sub-selection.

I was not convinced because introducing another state would make it confusing. So we end up user testing it (see below video), and we discovered that the sub-selection paradigm really confused our users as they are unsure to anticipate what should happen if they want to do a secondary action (e.g. moving the nodes) - does the sub-selection move or the selection?

Inertia for change

Changing the entire selection system for Raven was difficult, but changing it for the entire code 33 suite was almost an impossible task. Many people in the company is already accustomed to how the java desktop client operates. I needed to bring proofs and user research (see below) to the director of product and engineers to explain the awkwardness of a duel-selection system. 

 

 
 

Making

Compare & Contrast

To many analysts, the actual count of values isn't that important. What is more interesting for them is being able to see the proportion of the selected objects compared to the overall data set. I introduced partial highlight in the december version of Raven and managed to get this feature included in the Code 33 release

Slicing data just the way you want

Most analysts infer insights from the data via trial and error. They first establish a couple of possible variables and tweak them one by one to test. Under the filter system, it is now easy to manipulate variables to understand causation.

Treat data as the first class object

I got rid of the inconsistency between the timeline, histogram and canvas panel, and made a holistic selection system that works consistently across selection. When the analysts wants to constrain their data set they can easily create a filter by double clicking, right click, press enter or click on the create filter button. 

Below is my proposal, a web prototype made with edge animate. Screen recorded - that's why it's quite jerky.