Democratizing access to gaming analytics using Amazon Bedrock with Voodoo

Voodoo uses data to guide decisions across its business, from mobile game development to marketing and user acquisition. Teams needed timely insights to assess performance and decide where to focus their efforts. Nontechnical staff submitted data requests through a messaging application, often waiting days or even weeks for data analysts to respond.

The mobile game publisher saw an opportunity to use agentic AI to translate natural language requests into a complete, explainable analysis. The company wanted to give employees greater autonomy by helping them query its entire data lake—covering game performance, player retention, and monetization metrics—without SQL expertise. This would help teams get insights faster, accelerating data-driven decisions.

Voodoo launched DataWhisper, an agent that helps nontechnical stakeholders explore and query live data in natural language through conversational interfaces. The company first introduced DataWhisper as a side project and incorporated it into the messaging application so that employees could ask questions within their existing workflow. The tool returns analyses alongside traceability of how it reached results so that technical users can review the analysis accuracy.

The company then began rolling DataWhisper out to 30 early users, expanding access beyond the messaging application by exposing the agent through Model Context Protocol (MCP). “The first product was putting it in the place where everyone is already working so that they have answers to their questions at their fingertips,” says Fabien Traventhal, analytics engineering manager at Voodoo. “Then, it quickly evolved into an MCP. This is because the second tool that people spend time with is Claude Desktop, which integrates with other sources of information to support more complex workflows.”

To build and run the agent, Voodoo used Amazon Bedrock, a service for building generative AI applications and agents, and Strands Agents, an open source toolkit. Amazon Bedrock gives the company access to hundreds of foundation models from leading AI companies, including Anthropic’s Claude. “We implemented the agent with the Strands Agents Python framework, which works seamlessly alongside the AWS services we use,” says Traventhal. “This made it easy for us to get started, speeding up our time to market.”

The mobile game publisher implemented Amazon Bedrock AgentCore, a service for production AI agents, to simplify agent operations in production. Using Bedrock AgentCore Gateway, a fully managed AI gateway, Voodoo can securely expose the agent and its tools through MCP at scale. The company runs the solution on AWS Lambda, which provides serverless compute, and stores outputs in Amazon Simple Storage Service (Amazon S3), an object storage service.

Using DataWhisper, Voodoo reduced time to insight from days or weeks to an average of 2 minutes. The solution responds to 1,000–1,500 requests from 120–150 users daily—with around 220 unique users per week, about one-fifth of the staff.

Adoption spans executives, studio directors, developers, data analysts, marketing teams, and interns. A/B test analysis informs which game features to keep, while marketing and user acquisition teams investigate creative performance and campaign results. “If you reduce the time you take to get insights, you reduce the time to make a decision—and you probably make better decisions too,” says Traventhal.

Voodoo has also extended DataWhisper to more teams and products—including social networks BeReal and Wizz, and ad network Helix. “One of the benefits is that it showed everyone that AI can produce great internal products,” says Traventhal. “It has kick-started many other AI-related projects that we have right now.”