Adotas presents a Q&A with Luis Sanz, to discuss how machines bring in revenue for big brands and advertisers.
Q: What value can machine learning offer to brand marketers?
A: Machine learning can be applied in a wide range of applications to benefit brand marketers. Machine learning can analyze customer information, transactions, interactions, sentiment and so much more, to give marketers critical data that can inform their marketing strategies.
One of the most impactful ways that Olapic sees machine learning working for brand marketers is by analyzing visual content. User-generated content (UGC) is becoming a powerful force in brand marketing, and machine learning can help marketers determine what content is resonating, and why.
Machine learning can help find out why certain images are better at getting people to click on ads, down to the specific characteristics about the image itself. For example, machine learning could analyze an image of a bowl of soup and understand why it’s resonating with customers, calling out the color of the bowl, the amount of space it occupies in the image, the color of the soup, the size of the spoon and more.
Armed with that knowledge, marketers can then design campaigns and programs that lead to desired outcomes (such as higher conversions) with a higher degree of accuracy.
Q: How do brands filter out off-brand content, and surface imagery and videos that are most relevant?
A: When it comes to UGC, marketing teams are often looking to amass a set of on-brand content that matches the desired aesthetic and provides a consistent look and feel to their campaigns. But the nature of UGC means that brands have little control over the content that users create, so marketers need to figure out a way to separate the wheat from the chaff.
Brands can turn to solutions and platforms built on machine learning to search a set of brand guidelines before even starting to collect UGC. When those guidelines are applied, the solution is able to exclude off-brand imagery, or imagery that includes things marketers don’t want in their images. This eliminates the need to manually filter images, removing an enormous burden.
For example, if a brand sets parameters to exclude children, cigarettes, logos for other brands, selfies or blurry images, then those visuals will be excluded, vastly improving the odds of delivering only on-brand, high-quality imagery that can be activated in marketing programs.
Even though this is a highly technical process, there is still a human element involved. For example, at Olapic, our moderation team works with brands to apply training data to the machine learning processes to identify trends or standard imagery exclusions. The input of our moderators strengthens the machine learning algorithm immensely.
Q: How can brands add context and relevance (through things like metadata) to their content assets? How does the combination of machine learning and human
moderation deliver results?
Once a brand marketer has a collection of on-brand UGC, the next step is to map products, metadata, and tags to each asset, depending on where and how it will be used in the customer journey. Here again, we’ve found that machine learning combined with human moderation produce more accurate tags. This process involves using machine learning to identify objects in a picture, and enabling human moderators to manually tag and label photos.
Over the past few years, machines have become good at some computer-vision tasks, like identifying common objects in images, so machines are great for dealing with large volumes of images.
Humans, however, are still better at identifying the subjective qualities of images, like sentiment, as well as other visual subtleties that machine learning can’t recognize yet.
While this method can be intensive, this combination of human and machine also enables brands to better activate their content, and making it easier to bring a customer from a point of inspiration to a point of purchase.
Q: How can brands identify content that is most likely to outperform other assets
based on a variety of factors?
A: Once content has been curated and appropriately tagged, a final machine learning process can analyze how UGC will impact conversation rates, and where it would be best activated. Ideally, the machine learning element would analyze complex asset characteristics and recommend which content is most likely to perform at various points along the customer journey. For instance, one type of image may work well in social media, but a different type of image could work in an email campaign, while yet another image might be best suited for the product description page on an e-commerce site.
By analyzing a number of data points throughout the images themselves, and using its knowledge of what images have worked to increase conversions in the past, the machine learning algorithm is able to yield a powerful recommendation set.
Q: Does the increased use of machine learning mean that the human element is no longer needed in marketing?
Absolutely not. Machine learning is just one part of the equation for using UGC in marketing. We have proven that the combination of machine learning with human moderation produces much more accurate and powerful results from the use of UGC.
Even as machine learning becomes more efficient and accurate over time, and is able to analyze more data, there’s no way to completely replace the experience and the intelligence of humans. But when paired together correctly, they make a compelling and formidable team for brand marketing.
Q: How can marketers start using machine learning for UGC?
A: Before leveraging machine learning for UGC, marketers need to create a foundation on which the machine learning can occur, and that should be done by analyzing the performance of assets. So marketers should start by collecting assets now, and manually testing what types of UGC work in different environments.
Once marketers have a healthy volume of UGC, as well as some data on how it performs in different parts of the marketing ecosystem, then they have a solid foundation on which to deploy a machine learning solution.
Q: What is the future of machine learning and UGC?
A: Over the last few years, a number of trends have emerged. On the consumer side, just about everyone has a smartphone that they use to take pictures and access social media. This has led to the volume of UGC increasing. On the brand side, machine learning capabilities have advanced significantly, thanks in large part to the growing amount of UGC being created.
As these two trends continue to feed one another, we’ll see a cycle form, where brands optimize their usage of UGC through machine learning, which will lead to better and more accurate data, which will lead to the activation of better quality UGC, which will push more users to engage by creating more UGC, which will create more data, and so on.
As the cycle picks up momentum, the quality of the UGC, as well as the analysis of that UGC, will improve, bringing marketers insights that were previously thought to be impossible.
Luis Sanz is Olapic’s Chief Operating Officer and head of technology and product development. Prior to founding Olapic, Sanz was a consultant for Accenture, working for aerospace/defense projects in Europe and also worked for Ericsson, responsible for all support teams for the Billing Gateway and Service Provisioning departments across Europe, Africa, and the Middle East. Sanz and his Olapic co-founders have been recognized by the Eugene M. Lang Entrepreneurial Initiative Fund at Columbia Business School and were recipients of the E-cademy Award. Sanz earned a Master’s Degree in electrical engineering from the University of Zaragoza and holds an MBA from Columbia University in New York, where he co-founded Olapic in 2010.