Platfora

Behavioral Analytics

Project Name

Platfora

Team

Product Manager
Scrum Master
Front-end Developers
Back-end Devlopers
QA
User Experience Designer

Description

Platfora was a Big Data Discovery platform built natively on Apache Hadoop and Spark. Platfora enabled users to visually interact with petabyte-scale data in seconds. Data sources from hadoop, cloud, traditional databases went through data prep powered by platfora lenses and got translated to data products like platfora vizboards.

Platfora aligned scrum teams to feature deliverables. I was responsible for UX in following features: Behavioral Analytics, Vizboards, BI, etc.

Flow Problem Overview

Flow was a new visualization type that allowed users to analyze connections across a number of dimensions. This was particularly helpful for behavioral analytics.

As you can see in the image, flow was optimized for looking at event series, an order of actions taken by an entity. Sequencing of events helped answer questions such as 'Which page did users visit before leaving our site?'. To get this visualization, data had to be analyzed on entity-by-entity basis. (e.g., customer by customer)

Say an analyst was trying to create a flow to see what pages users visited after. Data had to be grouped by customers and ordered by timestamp. Then next 2 pages had to be output as columns based on position. For most analysts, writing an expression language for these steps was a challenge (see image below).

Solution 1

To assist analysts who were not comfortable with
writing expression language, we created event series builder
(from 2nd slide) where users could easily define event series,
select a pathing type and edit number of stages

This builder was used at a stage when users prepare and clean their data

Solution 2

After analysts were done with the data preparation process,
they brought these sequential event series fields (e.g., URL, URL+1, URL+2) to Vizboard to visualize as a flow

Because this was big data, users had to sort and limit fields to get an uncluttered flow.

We also introduced a new field type called complex field. Without this, users had to drag and drop each field to get a visualization. For one or two stages, complex field was not necessary. But, with more than ten stages, it became a must.

With complex field, all sub-fields lived under one parent field. Therefore when a parent field was selected and dragged, all sub-fields automatically followed.