GumGum is a leading computer vision company with a mission to unlock the value of every online image for marketers. Its patented image-recognition technology delivers highly visible advertising campaigns to more than 400 million users as they view pictures and content across more than 2,000 premium publishers.
Founded in 2007, GumGum invented the In-Image advertising category and is used by the majority of Fortune 100 companies. GumGum ads consistently achieve an 81% viewability rate and deliver 10 times better engagement than traditional display options, while still prioritizing high-quality inventory, brand safety, and user experience.
The company also offers GumGum Visual Intelligence, a real-time visual listening platform that helps brands identify and activate online pictures relevant to them, as well as engage their top influencers on social media.
Headquartered in Santa Monica, GumGum has 10 offices on three continents, including the US, the UK, and Australia.
About a year and a half ago, GumGum began work on a new initiative: expanding the company’s computer vision, or image recognition, capabilities beyond In-Image advertising. GumGum’s goal was to utilize image recognition on social media, based on the fact that at least 80% of images shared on social media relating to a brand do not include a text-based mention of the brand. With no hashtag or other identifying text, brands have no way of knowing about or finding this user-generated content—content that is critical to capturing and understanding marketing effectiveness.
Traditional “social listening” platforms do not, and cannot, solve this problem—it requires computer vision. GumGum developed a solution: an application for marketers that provides insights and visibility, as well as enables them to take action (pivot marketing strategy, refine messaging, engage with influencers, etc.) on social media image content.
While clients were intrigued with the initial product, the company noted a common request. Clients not only wanted GumGum to find these images, but they also wanted to understand the context in which the images were posted. Was the user at a baseball game when he or she posted the picture? At a sponsored event? Is the brand’s product on a table? In a refrigerator?
Based on this information from their customers, the GumGum team knew that for their new product, they needed to build in additional functionality that automated the process of understanding context.
To train the neural network of GumGum’s computer vision algorithm to identify relevant images and understand the context in which they were shared, the company began the search for annotated training sets to be used as ground truth. The team began with pre-existing public datasets, but found a large gap between the types of labeled data available and the type they needed: social media data.
From there, GumGum determined the best route was to build their own training datasets. A month’s worth of searching and trials netted them inadequate results from both outsourcing to contractors and working with third parties on their needs for annotations—neither could handle the fast pace GumGum required while maintaining high levels of accuracy. The team needed accuracy, scalability, and velocity and wasn’t finding it.
The Move to Mighty AI
It was this predicament that made GumGum interested in learning more about Mighty AI. The Mighty AI Training Data as a Service™ platform—with its advanced UIs, library of customizable task templates, qualified community of specialists, and proprietary 3-layer QA process—was a promising solution to GumGum’s challenges. After integrating via API (setup took 3-4 days), GumGum’s image annotations task set was up and running, and opened up to the Mighty AI community. GumGum started with a small volume of tasks to feel out the platform, and was up to full volume within 10 days.
GumGum’s goals were a solid match for Mighty AI’s Training Data as a Service™ platform and image annotation solution. While the Mighty AI team customized their image labeling template for GumGum, the GumGum team integrated with the API, then began posting images. Next Mighty AI distributed the images to qualified and trained community members to produce high-quality annotations, calibrated to GumGum’s vigorous standards. Throughout the process, GumGum was able to monitor task velocity and analytics through the customer portal.
When the task set was complete and results passed review checks, the data was posted to a web endpoint specified by GumGum. GumGum evaluated the accuracy of results as well as turnaround time to judge success—Mighty AI was able to achieve a 20-minute turnaround time. This process continues today.
In working with Mighty AI, GumGum is now able to expand on its product’s features, and provide richer insights to clients. The team is confident that with this new layer of additional datasets, their offering will clearly advance their position as a leading computer vision company.
Will you be our next success story?
Let’s discuss how Mighty AI can help.