A Generative AI Engineer built an LLM application using the provisioned throughput Foundation Model API. The application is ready for deployment, but the request volume is too low to justify a dedicated provisioned throughput endpoint. Which strategy should they use to ensure the best cost-effectiveness?