Dataiku Case Studies Rely on Automation for Scalability

Edit This Case Study Record

	Rely on Automation for Scalability Dataiku

Rely on Automation for Scalability

Dataiku

Technology Category	Analytics & Modeling - Big Data Analytics Analytics & Modeling - Predictive Analytics
Applicable Functions	Sales & Marketing
Services	Data Science Services
Challenge	A large national media organization wanted to provide high-quality recommendations for users of their app. Their goal was to target consumers with content that they would actually be interested in based not only on what they previously consumed, but how exactly they interacted with topics in which they previously expressed interest. For example, if someone chose to listen to a report on Topic A but then fast forwarded through much of the piece (as opposed to actually listening to the piece in its entirety), the app should take that activity into account for future recommendations. However, with a very small team and limited resources, the organization wanted to accomplish this in a scalable way. Not only would the system have to be mostly or entirely automated, but the team itself would have to be able to build the recommender easily in a way that would allow for quick tweaks and adjustments in the future. Read More
About Customer	The customer is a large national media organization that produces both news and public interest (cultural) pieces. They serve as a syndicator to a network of more than 500 stations across the United States. Additionally, their content is available directly from a variety of sources, including on the web and via their mobile application (which they develop and maintain in-house). The organization has a small team and limited resources, and they wanted to build a recommendation system for their app in a scalable way. Read More
Solution	The team behind the national media organization’s app leveraged Dataiku’s collaborative data science and analytics platform to efficiently and easily string together data cleaning operations and processes and then ultimately use that prepared data in the built-in supervised learning algorithms to build their recommender system. Out-of-the box without having to write code, the team was able to seamlessly try out different models and choose the one that produced the best recommendations for the app according to their business goals. At the same time, when they did want to use code for data processing, Dataiku enabled them to use a combination of languages (mainly R and Scala), choosing the tool that would be easiest for the task at hand. And with all of the data processing required (sometimes more than 25 recipes for a single processing operation), if any part of their process failed, it was simple and efficient with Dataiku for the small team to tell exactly where the issue was and to fix it quickly. Read More Log in to view content
Contents

Technology Category

Analytics & Modeling - Big Data Analytics

Analytics & Modeling - Predictive Analytics

Applicable Functions

Sales & Marketing

Services

Data Science Services

Challenge

A large national media organization wanted to provide high-quality recommendations for users of their app. Their goal was to target consumers with content that they would actually be interested in based not only on what they previously consumed, but how exactly they interacted with topics in which they previously expressed interest. For example, if someone chose to listen to a report on Topic A but then fast forwarded through much of the piece (as opposed to actually listening to the piece in its entirety), the app should take that activity into account for future recommendations. However, with a very small team and limited resources, the organization wanted to accomplish this in a scalable way. Not only would the system have to be mostly or entirely automated, but the team itself would have to be able to build the recommender easily in a way that would allow for quick tweaks and adjustments in the future.

About Customer

The customer is a large national media organization that produces both news and public interest (cultural) pieces. They serve as a syndicator to a network of more than 500 stations across the United States. Additionally, their content is available directly from a variety of sources, including on the web and via their mobile application (which they develop and maintain in-house). The organization has a small team and limited resources, and they wanted to build a recommendation system for their app in a scalable way.

Solution

The team behind the national media organization’s app leveraged Dataiku’s collaborative data science and analytics platform to efficiently and easily string together data cleaning operations and processes and then ultimately use that prepared data in the built-in supervised learning algorithms to build their recommender system. Out-of-the box without having to write code, the team was able to seamlessly try out different models and choose the one that produced the best recommendations for the app according to their business goals. At the same time, when they did want to use code for data processing, Dataiku enabled them to use a combination of languages (mainly R and Scala), choosing the tool that would be easiest for the task at hand. And with all of the data processing required (sometimes more than 25 recipes for a single processing operation), if any part of their process failed, it was simple and efficient with Dataiku for the small team to tell exactly where the issue was and to fix it quickly.

Impact #1	With automated data cleansing contributing to a data-driven recommender system, the media app automatically scores 200 million predictions every evening.
Impact #2	Dataiku has made it feasible for the small team behind the media organization’s content app to build and maintain this recommended system in a scalable way.
Impact #3	This has allowed for an increase in productivity by 50%, therefore allowing the data science team to work on other projects in parallel.

Impact #1

With automated data cleansing contributing to a data-driven recommender system, the media app automatically scores 200 million predictions every evening.

Impact #2

Dataiku has made it feasible for the small team behind the media organization’s content app to build and maintain this recommended system in a scalable way.

Impact #3

This has allowed for an increase in productivity by 50%, therefore allowing the data science team to work on other projects in parallel.

Benefit #1	The media app automatically scores 200 million predictions every evening.
Benefit #2	Increase in productivity by 50%.

Benefit #1

The media app automatically scores 200 million predictions every evening.

Benefit #2

Increase in productivity by 50%.

Download PDF Version

Overview

Rely on Automation for Scalability

Operational Impact

Quantitative Benefit