You are here

DataSong’s Big Data Analytics Platform for Marketing Optimization Helps Clients Understand Buying Behavior and Improve Customer Targeting


Background

DataSong Software’s marketing optimization and attribution engine uses historical data about individual customers to predict and influence future buying behavior.   With many companies vying for consumer spending, DataSong, a San Francisco-based software organization is increasing the precision with which marketing is analyzed. DataSong has developed a unique way to understand attribution that takes into account sales from all of a company’s order channels as they come into their call center, web site, mobile sites, or retail stores as well as activity in all of the marketing channels (display/retargeting, email, social media, direct mail, etc).  It goes well beyond traditional methods and is significantly more accurate than simpler methods (last click, averaged, etc).  Additionally, DataSong is leveraging the attribution platform to offer solutions that impact the top and bottom line, such as marketing treatment recommendations. Examples include who to include in next month’s catalog mailing, or whom to exclude from your retargeting display campaign. As a result, DataSong’s clients save money, increase revenue per campaign and customize each customer’s relationship with the company. 
DataSong fig1
Figure 1:  Without precise attribution modeling, many marketing departments simply credit revenue to the “last touch” the customer had before a purchase.  DataSong’s analysis shows this approach can cost marketers hundreds of thousands of dollars in unnecessary spending.

Challenge

Attributing Marketing Spend to Customer Revenue

“Once you know what is driving your marketing, you can better direct what treatments you are using for specific people.” – Brandon Mason, CTO, DataSong.

DataSong understands its clients’ challenges very well.  Without quantitative data to guide them, marketers often attribute revenue incorrectly.  When marketers are unaware of the actual purchasing influencers, they spend too much on the wrong things, afraid to reduce any part of the marketing spend because they don’t know how it will affect sales. 

This problem is especially acute in consumer marketing companies that spend a large percentage of revenue on marketing.  Further complicating matters, many of these companies have separate groups for catalog, e-mail or other marketing.  Each group justifies its budget by attributing revenue to its efforts.  DataSong’s analysis has shown that many company marketing groups use flawed approaches such as the “last touch” before purchase. As a result, the aggregate claims of revenue attribution of all the departments often sums to greater than 100% of sales, which is impossible. 

Marketing departments need an analytics-driven approach to help deal with the increasing amount of complexity that has emerged in the last 10 years.  Here’s why:

  • It used to be about being able to quantify the results of what is now considered basic and rudimentary information: POS data, direct mail (including catalog), email distribution, and/or telemarketing. Through codes, marketers would be able to attribute POS data with the specific marketing campaign. This approach isn’t effective today. Marketers have a much larger assortment to influence purchasing (e.g. email, mobile messages, banner ads, search, catalog, in-store promotions, affinity credit cards, etc.).  These channels need to be considered alongside factors that aren’t in their direct control (and which are often excluded from a company’s own assessment of revenue attribution), such as seasonal habits or the length of time the person has been a customer. 
  • Operational complexity makes it difficult to have one view of the customer’s activity.  Many marketing tactics are executed using disparate databases and platforms. 
  • IT systems that customize offers for each customer have matured, making it easier to feature a new product, a price promotion, or loyalty points.  Unless marketers know what offers drives which customers, they’ll end up programmatically using the wrong tools for the job.  
  • The availability of reduced “cost per touch” methods tempts marketers to increase the frequency of touches, (i.e. mailing catalogs is orders of magnitude more expensive than an email or text message), potentially over-saturating or even annoying customers. 
  • Customer mindshare has become increasingly difficult to win due to the constant barrage of marketing being presented on television, in print, on websites and mobile devices. 
  • Marketers don’t know how to utilize the terabytes of data collected about their customers because of both its variety and range of formats.  The amount of data being stored in organizations (e.g. every mouse click, transcripts of conversations in call centers, customer demographics and history, and even third-party-sourced data such as credit scores), often called “Big Data,” has outgrown traditional analytical tools.  The value of this data is often untapped. 

DataSong's principals, self-described “data geeks,” saw an opportunity to create a platform – an analytics engine – that would utilize dozens of data types as inputs to individual buying behavior models for each of its clients’ customers.    Since they wanted the ability to analyze data at the granular level for each customer, the DataSong platform would quickly grow to rely on many terabytes of data.  DataSong looked at the current off-the-shelf applications and approaches and rejected them.   Their platform must allow for data exploration, allow them to customize the statistical methods used in the models, execute very quickly on big data and would need to scale without sacrificing speed. 

In order to be a game changer, they would need to build their own.
DataSong fig2
Figure 2:  Modern marketers must utilize a huge amount of internal and external data to attribute revenue per customer to specific variables (aspects of the marketing mix that are in their control and elements that are out of their control) in order to optimize the marketing mix and associated spend. 

Solution

Custom Analytics with Revolution R Enterprise and Hadoop
DataSong's purpose-built application is a modern, high-performance big data analytics engine, scoring 50 million records per day for each of DataSong’s customers.  It marries Revolution Analytics’ Big Data Analytics capabilities with Hadoop’s data management and computational power.  Since no two clients are exactly alike, the statistical methods that underlie the analytical models can be customized to meet each client’s exact requirements. 

From an analytics perspective, the company borrowed approaches from forward-thinking and more analytically mature industries. For example, DataSong adapted models used in the bioscience sector, where GAM (Generalized Additive Model) survival analysis techniques effectively measure differences in the outcomes in patients under different treatment regimens.  However, many of the methods that DataSong wanted to use had not been designed for Big Data analytics.  Using Revolution R Enterprise, which is based on the power of the R statistical platform, DataSong built a “big data analytics engine” utilizing multivariate statistics, time-to-event models and GAM survival analysis techniques, which tells its clients about:

  • Purchasing triggers
  • Who is most likely to buy
  • Cross-channel triggers – i.e. looking through a catalog, but making an online purchase
  • Recommendations – which marketing treatment to apply (and when) per user
  • Strategic allocation – Reallocating marketing funds as a result of modeling

DataSong’s platform utilizes Revolution Analytics’ RevoScaleR, a set of big-data statistical analysis capabilities offered with the Revolution R Enterprise software. With RevoScaleR, the scalable, high-performance “XDF” data file format optimizes the process of streaming data from disk to memory, dramatically reducing the time needed for statistical analysis of large data sets.   It runs on commodity hardware, which has helped reduce the cost of the platform. 

CEO John Wallace describes the results.   “We’ve combined Revolution R Enterprise and Hadoop to build and deploy customized exploratory data analysis and GAM survival models for our marketing performance management and attribution platform. Given that our data sets are already in the terabytes and are growing rapidly, we depend on Revolution R Enterprise’s scalability and fast performance – we saw about a 4x performance improvement on 50 million records.    It works brilliantly.”   

According to CTO Brandon Mason, “We like the fact that Revolution Analytics is bringing R to Hadoop, and has a strong Hadoop-related roadmap.  That kind of enterprise support is important to us.”


Figure 3:  DataSong has utilized Revolution Analytics’ Hadoop capabilities to leverage RevoScaleR’s Big Data Analytics capabilities and Hadoop’s data management and model execution capabilities.  The engine scores 50 million records per day per client using commodity hardware. 

Results

Reduced Marketing Costs and Revenue Uplift for DataSong Customers

With the help of Revolution R Enterprise, DataSong has created a big data analytics solution for marketing optimization that has been credited with saving one client $270,000 in just one campaign and delivering a 14% revenue uplift for another. 

DataSong enables marketers to plan, measure, and optimize marketing campaigns across all marketing channels using integrated software and customer-level, advanced revenue attribution. Whether it’s three or a dozen multi-channel touches to the consumer, marketers can reliably plan and allocate spend for each marketing channel based on actual performance.

About Revolution Analytics

Revolution Analytics was founded in 2007 to foster the R community, as well as support the growing needs of commercial users. Our name derives from combining the letter "R" with the word "evolution." It speaks to the ongoing development of the R language from an open-source academic research tool into commercial applications for industrial use. 

Though our Revolution R products, we aim to make the power of predictive analytics accessible to every type of user & budget. We provide free and premium software and services that bring high-performance, productivity and ease-of-use to R – enabling statisticians and scientists to derive greater meaning from large sets of critical data in record time.  

We also offer our full-featured production-grade software to the academic community for FREE, in order to support the continued spread of R's popularity to the next generation of analysts. 

For customers such as Pfizer, Novartis, Yale Cancer Center, Bank of America and others, our flagship Revolution R Enterprise product stands for faster drug development, reduced time of data analysis, and more powerful and efficient financial models.