In this series we will explore the latest thinking about AI, machine learning, and data science through a series of interviews with Ipsos data science leaders. In our first post, we talk with Benoit Hubert, Chief Data Science officer at Ipsos Social Intelligence Analytics.
Tell us about your data science journey.
I fell into data science following a computer science path during my teenage years. When I was 13, I used the money I earned from my first student job to buy a Commodore 64. This nifty machine helped me develop my first gaming applications (Spyke Control), cartographic representations of France and Europe, and even fake news streams that I broadcasted on my family’s TV.
After high school, I pursued electronic computing and mathematics courses in college. Eventually, I got my degree in statistical and digital engineering and became a data scientist. After a stint at a large financial firm working with banking network data, I was drawn to market research and joined a company focused on customer satisfaction and experience measurement (and was involved in implementing the first dynamic web server for survey results in France!).
Fast forward a few years and I was testing and implementing techniques that pushed the limits beyond what humans could achieve with things like decision trees, random forests, and shallow neural networks. These data mining algorithms were my first foray into machine learning – and a powerful complement to classic tools like inferential statistics and data analysis that were already delivering powerful results on small data sets.
Now at Ipsos, I work with the latest evolutions in the field of data science every day. Combining extraordinary developments in computing power with the massive flow of qualitative data is leading us down new, exciting roads.
How are brands thinking about data science today?
Brands have always been receptive to and interested in data science. They’ve always needed data to assess drivers of demand, their competitive landscape, and their positioning in the market – and to forecast future performance. Marketing science and data science offer multiple methods for getting this kind of information. Yet brands’ perceptions of data science have evolved in recent years.
With the exponential growth of data (everywhere!), brands are starting to grasp that they need to tap into consumer-generated data sources. To better understand consumers and their behavior (across millions of posts), brands need machine learning, AI, big data, automatic language processing, image processing, sentiment and emotion analysis, and more.
Recently, brands have started to understand consumers’ spontaneous expressions on social networks. The development of multi-language processing techniques that use bottom-up approaches allow insights pros and marketers to get complete, unbiased pictures of what consumers are saying about their brands, competitors, and markets. This is valuable both upstream and downstream from traditional research methods and provides valuable insights that more brands are starting to turn into action.
Lastly, video and image analysis are on everyone’s minds – this is an area in which we’ve undertaken significant research and development work. It is really the new frontier, and already we have been using video analysis to help clients with everything from commercial performance modeling (i.e. what’s retaining viewers’ attention and what’s not), to CPG packaging performance (i.e. analysis of the images and graphics on a package, logos, and messaging compared to others in their category).
What challenges do insights and marketing pros face – and how are we helping them?
Among the top challenges for market research is the ability to offer decision-making tools to address pressing issues like economic growth or climate change. To achieve this, data science needs to be transparent and explainable (and trustworthy!). Similarly, improving the accuracy of predictions will continue to be a core challenge. Can we trust prescriptive actions based on data-science analysis?
For Synthesio and Ipsos, our challenge is to embrace data science more broadly to cement our position as an industry leader. Our hybrid approach combining data science and AI draws on our human, methodological, scientific, and technical assets. Thus, our ability to create high quality, versatile knowledge bases (training samples) from big data environments like social networks or surveys is critical. It will also be important to study, exploit, expose, market, and maintain the knowledge we gain from our data science methods, and further anchor ourselves in the digital economy.
Over the last decade, there has been a race to higher performance in data science models, especially with the emergence of deep learning. However, models won’t always be 100% correct. And when they’re wrong, it’s crucial to be able to explain why (or else the credibility of the model will be questioned). And of course, if you’re able to understand why something went wrong, you can work on fixing the mistake and improving the future quality of the model.
A related challenge for applying data science is to avoid bias when building models. For example, some biases can lead to reinforced stereotypes (especially gender and racial). One recent study showed that Facebook’s job ad algorithms were disproportionately targeting certain genders: software engineers for Nvidia and sales associates for cars skewed male, while Netflix and jewelry skewed female.
Consequences like these are why it’s crucial to build unbiased models – and where working with large research companies like Ipsos that have decades of experience building representative panels can help. This knowledge, along with our understanding of local markets and how consumers express themselves online, are all built into our algorithms to yield the most accurate, trustworthy results.
Want to learn more about Synthesio and Ipsos’ data science capabilities? Request a demo with one of our experts.