Mario Vinasco currently works for Facebook as data scientist in the consumer marketing group; in this role he is responsible for improving the effectiveness of Facebook’s own consumer-facing campaigns. Key projects include ad-effectiveness measurement of Facebook’s brand marketing activities, and product campaigns for key product priorities. Prior roles included VP of business intelligence in digital textbook startup, people analytics manager at Google and eCommerce Sr manager at Symantec.

He has over 18 years of progressive experience in data driven analytics with emphasis in database programming and predictive models creatively applied to eCommerce, advertising, customer acquisition/retention and marketing investment. Mario specializes in developing and applying leading edge business analytics to complex business problems using big data platforms, including Hadoop, columnar and traditional relational models.

Mario holds a Masters in engineering economics from Stanford University.

Here is my interview with him:

Anmol Rajpurohit: Q1. What are the main responsibilities of the Marketing Analytics team at Facebook? What benchmarks does the team use to measure its success?

Mario Vinasco: We strive to understand people everywhere to inspire the development of indispensable products and high impact marketing programs. We foster a culture of continuous learning and accountability to help Facebook improve and grow.

We measure our success with a combination of product metrics and brand survey questions that track user satisfaction, trust and favorability.

AR: Q2. What are the most common tools and technologies used by the Marketing Analytics team to achieve its goals? How has the experience been to move from relational databases to Hive?

MV: We use a combination of tools, technologies such as Presto and Hive (both open sourced), Experimentation Tools (internal), Survey deployment technologies (internal), Ads manager (available to advertisers), MS Excel, Tableau, Oracle and Vertica among others. The facebook data warehouse was built on Hive, however there are many needs for relational databases. We use Vertica for interactive analysis.

AR: Q3. Today, most consumers are having a multi-channel experience. Moreover, the lines between various channels are increasingly becoming blurry. How is this trend impacting Marketing Analytics' ability to judiciously attribute the performance across channels?

MV: Facebook has developed a sophisticated technology called Atlas that identifies users across devices; still, offline channels such as TV, radio, Out of Home, are difficult to measure precisely and we perform multi-channel attribution analysis. This attribution assigns probabilities of a user exposure to a campaign.

It is still part science and part art, and that is when a good analyst brings value to the table.

AR: Q4. What factors (both business and technical) play a key role in determining the success of Predictive Analytics projects?

MV: We have accumulated experience and benchmarks by performing many campaigns that use advanced segmentation, such as lookalike models and other classification models; these campaigns are measured according to their objectives, in some cases are product metrics and in others are awareness and sentiment.

AR: Q5. Even in the Big Data era, one of the consistent issues is incomplete and/or inaccurate data. How do you deal with this challenge?

MV: We use sampling techniques and cohort analysis and extensive testing and apply statistical techniques to create confidence intervals; in other instances, we need to inter or extrapolate data to fill in gaps, and in some cases make educated guesses.

AR: Q6. What were the key lessons for the Marketing Analytics team from the launch, monitoring and analysis of Lookback video feature on Facebook platform?

MV: Lookback was a huge success in terms of user satisfaction, fun, relationship enhancements and other key customer dimensions; after Lookback, we launched Say Thanks, Year in Review and other products that continue to help our mission to connect the world.

Lookback in particular, motivated friends to like and comment substantially more on this video than in any other piece of content.

We also learned a great deal about the technical difficulties of monitoring and measuring these products and we moved to network experimentation techniques for SayThanks.