Parrot Analytics scales smarter content intelligence with Amazon Bedrock AgentCore and Amazon Nova

Parrot Analytics measures audience engagement across video platforms, social media, and research datasets to understand global demand for films, television series, and sports content. These insights help studios, investors, and streaming platforms evaluate performance and decide what content to produce, acquire, and invest in.

As the platform expanded, the volume and complexity of incoming signals increased significantly. Much of this data arrives in unstructured formats and must be classified before analysis. For example, a video about a sporting event may reference multiple teams, players, tournaments, and locations that must be accurately identified and linked to internal databases.

Previously, some of this work relied on manual verification and internally developed ML models. Processing 10 million signals required up to 12 weeks of human verification and cost approximately $100,000 per batch, slowing analysis cycles and limiting how quickly insights could be delivered as signal volumes grew.

At the same time, Parrot Analytics continually refines its demand classification algorithms to keep pace with changing audience behaviors. Rebuilding and improving these models while processing large volumes of signals increased development timelines and added operational complexity. The company needed a scalable way to automate metadata extraction and signal classification while reducing cost and processing time.

To modernize its signal classification workflows, Parrot Analytics implemented an agentic AI architecture on AWS using Amazon Bedrock AgentCore, an agent development platform that helps organizations build and deploy agents for complex workflows, and Amazon Nova 2 Lite, a fast and cost-effective reasoning model used to analyze unstructured media signals and extract relevant metadata at scale.

AWS experts worked closely with Parrot Analytics during implementation to design and optimize the architecture for large-scale AI workloads. The new system integrates with Parrot Analytics’ existing data infrastructure, which ingests signals from global sources across more than 100 markets. Together, the teams refined agent-based workflows capable of analyzing incoming signals, extracting relevant metadata, and linking the information to Parrot Analytics’ internal knowledge base. These workflows allow the platform to interpret unstructured inputs—such as videos referencing television series, films, athletes, or sporting events—and automatically match them to the correct entities within the company’s demand analytics system.

Amazon Bedrock AgentCore Runtime orchestrates these agents across multiple workflows while running them in isolated MicroVM environments for secure execution at scale. This helps the platform process large volumes of signals efficiently, while Amazon Nova 2 Lite provides the underlying AI inference.

“Amazon Bedrock AgentCore helped us rapidly deploy AI agents that orchestrate intelligent workflows to classify millions of content signals accurately while scaling our analytics platform globally,” says Wared Seger, CEO of Parrot Analytics.

By implementing AI agents on AWS, Parrot Analytics significantly improved the speed and efficiency of its signal processing workflows. The new system processes media demand signals 10 times faster than previous methods, enabling the company to analyze large volumes of incoming data more quickly and generate insights sooner.

Operational costs also decreased substantially. Automating metadata extraction and classification workflows reduced operational expenses by approximately 60 percent. Using Amazon Nova 2 Lite further improved cost efficiency by delivering comparable performance to alternative models while lowering inference costs by up to five times versus Claude Haiku 4.5.

The platform now supports high-volume AI workloads, processing up to 25 transactions per second (TPS) and 20 million tokens per minute (TPM) for media signal classification and analysis.

In addition to scaling throughput, automation reduced manual verification. AI agents running on Amazon Bedrock AgentCore now handle approximately 85 percent of signal classification tasks, allowing engineering teams to focus on improving algorithms and developing new capabilities for the platform.

“Amazon Bedrock AgentCore with Amazon Nova 2 Lite transformed our audience demand intelligence pipeline, cutting manual verification by 85 percent, processing signals 10x faster, and reducing costs by 60 percent,” says Seger.

With these improvements, Parrot Analytics can continuously refine its demand models and scale its analytics platform to support growing global datasets while delivering faster insights to studios, investors, and streaming platforms.