Optimizing Digital Marketing Campaigns with Amazon Bedrock by Adin.ai

Use case general description 

The use case aims to develop an optimization suggestion mechanism for enhancing the performance of ongoing digital advertising campaigns across Google Ads, Meta, and TikTok platforms. The project utilizes campaign performance data collected from these platforms, manually labeled data provided by domain experts, and trusted plain documents related to digital advertising optimization. 

Various foundational models such as Llama 2-70b-chat, Claude 2, Claude 3, Command, and Titan Express were trained using these datasets. Fine-tuning, continued pre-training, and knowledge-based methods were employed through-out the process. 

Important results in numbers 

L'Oréal 

In the study conducted with the Maybelline New York brand of L'Oréal company, with the generative AI-based smart optimization system, there was a 67% decrease in cost per reach, a 71% decrease in cost per ad recall and a 3% increase in ad recall(people). In total, 40% cost saving was achieved after only 2 days from smart optimization. 

Papara 

As a result of our work with Papara, we achieved a 30% reduction in unit costs through smart planning and optimization, successfully reaching an active customer base. 

Model selection 

Various foundation models available within Amazon Bedrock were utilized for the smart optimization modeling process. Models such as Llama2-70b-chat, Claude 12, Claude 3, Command, and Titan Express were selected based on their respective strengths and employed within the scope of the study. Additionally, fine-tuning and continued pre-training operations were conducted based on accessibility for the Llama 2-70b-chat, Claude 3, and Titan Express models. Throughout the study, the emphasis was not on utilizing a single model, but rather on leveraging the strengths of multiple foundation models in conjunction. 

Experiment matrix

Figure 1: Experiment matrix

Implementation

This section outlines the implementation details of the scoring system, including the production environment setup, and system architecture. 

Production system description 

At the heart of the optimization system lies the Amazon Bedrock service. Additionally, AWS Lambda, Amazon S3, Amazon API Gateway, Amazon Cognito, AWS Developer Tools (including AWS CodeCommit, AWS CodeBuild, and AWS CodePipeline), Amazon ECR, and AWS Amplify services are leveraged to ensure the system’s smooth operation in the production environment. 

System architecture 

The architecture of the system is built upon the premise of models providing recommendations in output format, adhering to specific standards and formats, based on input data sets. Input data for the model is stored on Amazon S3. Customized prompts tailored to individual models are prepared within the Amazon Bedrock component and implemented within the Amazon Bedrock section specific to each model. The outputs obtained are also stored on Amazon S3. Within the architecture, the suitability of optimization recommendations made by the models to predefined standards based on domain knowledge is taken as a checkpoint. 

Figure 2: Smart optimization architecture

UI screenshots 

Figure 3: Smart optimization screen

Conclusion 

Our study delved into various methods using different Amazon Bedrock foundation models, focusing on crucial metrics like performance, cost efficiency, and success rates. Employing a model mix approach, we strategically combined strengths from multiple models to effectively address output variability. Furthermore, specific models were leveraged to enhance the training of others, fortifying the system’s robustness. Model-specific prompt engineering ensured outputs were tailored to meet specific requirements.

This case study represents a significant advancement in digital advertising, introducing a pioneering suggestion mechanism powered by generative AI within Amazon Bedrock. Highlighting the importance of Amazon Bedrock, our study underscores its pivotal role in driving future innovations in the industry. Through Adin.ai’s holistic approach and leveraging powerful generative AI models, we aim to revolutionize the sector by delivering tailored suggestions derived from cross-channel data analysis facilitated by AWS’s architecture. These suggestions are meticulously crafted to optimize campaign performance and drive results.

In conclusion, our study serves as a testament to the transformative potential of generative AI within Amazon Bedrock, promising to reshape the industry and deliver exceptional results on a global scale. As we remain committed to innovation, we’re poised to lead the way in digital advertising optimization, paving the path for future advancements and setting new industry standards.