Q&A with Asaf Shemesh, Supply Product Manager, and Roy Sadaka, Machine Learning Team Lead
There’s a lot of buzz out there about machine learning or the practice of building highly accurate applications that learn from data over time to help drive business decisions. It’s a branch of artificial intelligence (AI) that makes a big difference behind the scenes in the consumer insights industry. Our in-house experts Asaf Shemesh and Roy Sadaka explain how, as well as what it means for customers.
- Q: How is machine learning integrated into the Toluna experience?
- A: We design a deep learning model that’s constantly being trained across our respondent panel data from about the last two years. The idea behind the model is to leverage that power to make accurate predictions about how panelists would match up to certain surveys — and even how they would answer specific questions. We’re constantly fine-tuning the system to get to its best potential, trying different sets of parameters to see how they perform on KPIs, and testing and refining to improve accuracy.
- Q: Can you offer an example of machine learning that we see in daily life?
- A: Sure — think about Netflix. They’ll ask for some basic information when you sign up, then monitor your behavior and patterns over time to better determine who you are. This helps the streaming service recommend shows or movies you might like by using predictive data.
- Q: So why is machine learning important for Toluna panelists and customers?
- A: It allows us to make better matches between respondents — or personas, using the information we’ve gathered about them — and the surveys that would be most meaningful to them. This keeps them motivated as members of our panel, and we want our members to be as happy and satisfied as possible. On the customer side, machine learning helps drive value through efficiency, automation, better intelligence, and prioritization to deliver increased levels of speed and quality for our clients. Our respondents are presented with surveys that are relevant for them which keeps members engaged and responsive.
- Q: What do you mean by prioritization?
- A: Surveys have different priority levels and senses of urgency attached to them. In the past, flagging them in terms of priority was an all-manual task. Now, machine learning and advanced algorithms can help make these decisions about what is urgent and what would be the best survey to offer a member at a particular time. It helps bring the surveys we need to complete forward.
- Q: Are there any other benefits machine learning provides in the consumer insights space?
- A: Yes, there are. When developing these data-driven personas, we’re also able to detect similarities between our panelists. These segmentation capabilities allow for a new set of features that were not possible before, such as enhanced fraud detection, or anticipating panelist sentiment towards various topics. Machine learning also helps us with traffic management by understanding and predicting the pace of traffic in a given time. Pace prediction helps control traffic volume, creating a more optimized respondent experience.
- Q: How do you see machine learning progressing in the consumer insights space?
- A: As we look into the future, we may see a world where we direct 100 people to a survey with a target of 1,000 respondents, gather their data, and use machine learning to predict the outcome of the other 900 panelists with as high of an accuracy as possible. There may be a scenario where clients can opt for virtual respondents using data inferences rather than executing a full survey. This may be especially relevant if it’s almost impossible to survey certain very hard-to-reach groups. In the near term, quality improvement is a big benefit of machine learning. It can help us do things like remove and even prevent gibberish and toxicity. Our platform already has classic solutions to flag these types of things, but machine learning can block them ahead of time as the content is recognized by a machine.
As we continue to invest in our platform, we also prioritize technological investments. Machine learning helps us create better respondent matches, create a deeper understanding of our respondents, flag priority surveys, and better detect fraud while maintaining quality. Stay tuned for where it will take us next.