A retail analytics team ingests clickstream events into a single Kinesis Data Stream with 10 shards. Three separate consumer applications (real-time dashboards, fraud detection, and an S3 archiver) each read from the stream. The team notices that as they add more consumers, each application experiences increasing read latency and frequent ReadProvisionedThroughputExceeded errors, even though the producers are well within write limits. Which change best resolves the read contention with the least ongoing operational overhead?