Universal Containers (UC) uses Apache Kafka on Heroku to stream shipment inventory data in real time throughout the world. A Kafka topic is used to send messages with updates on the shipping container GPS coordinates as they are in transit. UC is using a Heroku Kafka basic-0 plan. The topic was provisioned with 8 partitions, 1 week of retention, and no compaction. The keys for the events are being assigned by Heroku Kafka, which means that they will be randomly distributed between the partitions.
UC has a single-dyno consumer application that persists the data to their Enterprise Data Warehouse (EDW). Recently, they've been noticing data loss in the EDW.
What should an Architect with Kafka experience recommend?