“Performance Maximizer is an AI marketer exclusively for advertisers. “The more you employ it and the more data you accumulate, the upper your promoting performance will be.”
Buzzvil Product Manager (PM) Hong Dae-gi expressed Buzzvil’s vision as “creating an optimized AI reward promoting expert.”
“The name Buzzvil could also be unfamiliar, but 17 million people, one-third of your entire population, have seen Buzzvil reward advertisements, and the reward amount has exceeded 25 billion won,” he said. Which means that it’s already a standard promoting method.
Particularly, he said that “the acquisition conversion rate in comparison with the quantity of ad exposure was improved by about 270%, and the profit in comparison with the ad cost was improved by about 400%” with the Performance Maximizer that grafted AI technology onto existing reward ads. Performance Maximizer is a technology that analyzes user behavior data and infers users who can create the very best value.
Buzzville’s AI model learns from first-party user data provided by advertisers in each industry, and uses models optimized for every advertiser.
First-party data refers to user interest and behavior data held by an organization through membership registration. Within the case of a shopping center, this refers to product searches, adding to cart, and purchases.
The bottom model is an open source artificial neural network model (NNM). Greater than 600 million Buzzvil user response variables and predictor variables were analyzed.
A small variety of users were expanded into similar user groups through a seed generation model (M1), and the means of prioritizing promoting users by determining similarity through a worth inference model was automated with AI.
PM Hong Dae-gi said, “Because the data provided by advertisers varies greatly by industry, once the model training is complete, it may be seen as a very different model.” He also explained that they’ve an internal solution that automates this data preprocessing process, which leads to fast and accurate promoting execution.
Based on this technology, it is feasible to filter out ‘cherry pickers’.
Cherry pickers are users who simply receive rewards and don’t take part in value creation activities when running reward advertisements. As an advertiser, it’s important to filter this, and Buzzvil solves this by applying a performance maximizer to ‘multi-reward’ and ‘dynamic reward’.
Multi-reward is a technique of providing rewards in stages to distinguish between high-value and mid-value users and induce mid-value users to take actions desired by advertisers. Within the case of online shopping malls, rewards will be repeatedly provided at stages comparable to product search, adding to cart, purchasing, and repurchasing, inducing continuous purchase of a single product.
Dynamic rewards are also a technology that gives different rewards depending on the potential value of the user. High-value users who make purchases, etc. are given more rewards, while cherry pickers are given fewer rewards.
The performance maximizer will be applied even when there may be little prior data. Within the case of a brand new game launch, there isn’t any user behavior data, but expected goal data from the starting stage will be used.
“After applying AI, the efficiency of the promoting operation team has increased significantly,” he explained. “We’re improving promoting performance by changing promoting optimization settings in real time.”
Along with enhancing the functionality of every solution, the corporate said it’s specializing in cost optimization. “We’re supplementing ML Ops personnel to optimize cloud costs,” he added.
Hong Dae-gi, product manager, said, “Buzzvil has been within the reward promoting business for 10 years, and advertisers have been showing a 90% re-execution rate,” and “We expect the re-execution rate to extend further after introducing AI.”
Reporter Park Soo-bin sbin08@aitimes.com
