Russell Ratshin, solution architect with Teradata Aster

Russell Ratshin, solution architect with Teradata Aster

Tech2Tech

Ask the Experts

Know Your Customer

The Teradata Aster digital marketing optimization solution simplifies the process of uncovering unique customer insights.

The explosive proliferation of customer interaction channels is challenging data-driven marketers to better understand and serve customers within this expanding universe of touch points and data sources. With the momentum for digital marketing optimization (DMO) solutions on the rise, Teradata Magazine spoke with Russell Ratshin, solution architect with Teradata Aster, about what makes the company’s offering unique among DMOs.

What are the special characteristics of diverse data types like email, digital advertising and social media, and what are the implications for big data analytics?

Three things you didn’t know about Russell Ratshin

{ 1 } He is the co-founder of Wooloo.org, an online resource that brings together artists from around the world who are interested in pursuing collaborative work.

{ 2 } A songwriter and musician, Russ plays rock ’n’ roll, folk and rap music.

{ 3 } Before embarking on a technology career in Silicon Valley, he studied political science at UCLA and Columbia University.

Ratshin: A successful digital marketing solution often involves interacting with customers and potential customers through multiple channels. This translates into a variety of data sources and, very likely, a variety of different data types.

Relational databases are built to store relational data and address the demands of analysts who work with relational data. But data types like those found in clickstream logs, advertising logs and social media streams require more than a relational data store. Big data analytics also requires the ability to parse and understand semi-structured data that is frequently found in log files and formats such as XML and JSON.

What big data analytic approach does the Teradata Aster platform incorporate into a DMO strategy?

Ratshin: Interactive or digital marketing encompasses every digital interaction that an organization has with an individual. Emails, banner ads, search engine optimization and website content are all critically important in shaping a successful DMO strategy. Teradata Aster’s SQL-MapReduce approach to big data analytics enables the user to easily track and predict behavior using multi-channel path and pattern analysis, statistical analysis and complex marketing attribution algorithms.

The volume and complexity of new data sources requires advanced analytics beyond last-touch or last-click attribution. What is unique about Teradata Aster’s attribution analysis?

Ratshin: The solution provides several tools that go beyond a last-click approach. Besides the out-of-the-box attribution operator, which is used specifically for scoring various events that precede a conversion or any other event of interest, there are operators available for sessionization and path and pattern analysis. Additionally, custom operators can be developed and installed in the database to handle any needs not addressed by the ones included with the platform.

What attribution models are offered by the Teradata Aster DMO solution and how easy are they to implement?

Ratshin: The operator is executed by an analyst as a simple extension to standard SQL. By altering a single parameter, a conversion-related score can be assigned to specified events using any one or a combination of these models:

  • First- or last-click.
  • Uniform. The score is distributed equally among preceding views or clicks
  • Exponential. Scoring decays exponentially to favor the most recent events preceding a conversion
  • Weighted. The score is based on a defined number of events prior to a conversion or in relation to specified periods of time

How does your unique SQL-MapReduce architecture support the attribution model development, and how does it differentiate the solution?

Ratshin: Some types of data analytics are best handled using standard SQL. Others are best handled with a “programmatic” MapReduce approach. Teradata Aster’s SQL-MapReduce framework brings together the benefits of these two approaches by allowing the analyst to seamlessly switch between standard SQL and SQL-MapReduce, depending on which one can best answer a specific question.

Complex attribution models, such as exponential or weighted models, can be effortlessly applied in the Teradata Aster platform, often in just a single pass of the data. Analysts with a SQL skill set and SQL-based BI tools can configure and utilize models that were developed using MapReduce without having to alter or even understand the underlying programmatic MapReduce code.

Why DMO?

Digital marketing optimization (DMO) provides a detailed understanding of customers, which enables marketers to improve their reach and performance. DMO also helps organizations maximize their messaging across all channels to better engage customers and improve campaign effectiveness.

How is the solution incorporated into the DMO workflow?

Ratshin: Improving the website experience or the effectiveness of advertising campaigns is a cyclical process driven by action and reaction. The enormous amount of data that can be drawn from websites and campaigns allows a data scientist to analyze visitor flow and then compare and contrast results from previous and live campaigns.

A rapid and thorough investigation of this information can yield immediate insights that can then be fed back into the cycle. As an exploratory platform capable of answering data questions using standard SQL and SQL-MapReduce, Teradata Aster enables this type of investigation.

What might companies expect when they start using it?

Ratshin: Companies using the marketing attribution operator are seeing an immediate boost in marketing effectiveness. The goal of introducing more complex attribution scoring models is to discover whether there are advertisements or events that are influential and important, but are effectively ignored by the traditional last-click model.

The Teradata Aster attribution operator provides two important pieces of information for every relevant event: the percentage of the total score assigned to it, and how much time passed between the event and the conversion. For example, an exponential scoring model may uncover that certain campaigns, such as early product announcements, may not have led directly to a purchase or conversion, but they paved the way for other advertisements or events that followed. A last-click or last-touch model will only recognize the impact of the most recent event and not the ones preceding it.

Also, time-sensitive campaigns, such as those executed during the holiday season, may receive lower overall scores compared to other campaigns. But a lower time-to-conversion value may outweigh the lower score when determining if the campaign was a success.

How is digital source data from social channels such as Facebook and Twitter incorporated?

Ratshin: Depending on the volume of data and the desired refresh rate, social channel data can be loaded into the Teradata Aster MapReduce Appliance in its semi-structured format using the standard loading tool or the respective application programming interfaces [API] made available by sites such as Twitter or Facebook.

Rather than performing any ETL processes prior to bringing in the data, the Teradata Aster platform enables any required data transformations to be performed in-platform in a parallelized manner. Once the data is loaded, a number of operators are available to conduct marketing- and consumer-related research, such as graph and social influencer analysis.

Can different sources of data be combined and associated to the customer?

Ratshin: To the end user, the Teradata Aster platform is a relational database. Because of this, all types of disparate data sets can be pulled together using simple SQL join techniques, just like with most other databases. When simple joins are not enough, the SQL-MapReduce platform allows for “fuzzy matching” between data records for customer identification across disparate data sets. Algorithms not possible with standard SQL can be built to identify individuals by using data such as retail store transactions or social networking aliases and activity.


Your Comment:
  
Your Rating:

Comments
Datawatch Q4-2014