Caylent Catalysts™
Generative AI Strategy
Accelerate your generative AI initiatives with ideation sessions for use case prioritization, foundation model selection, and an assessment of your data landscape and organizational readiness.
Jungle Scout Turns Amazon Market Intelligence into Agent-Ready Data on AWS
Generative AI
Artificial Intelligence & MLOps
At a glance
Jungle Scout is an Amazon Market Intelligence provider that has helped more than a million sellers over its lifetime to use data to innovate and grow more quickly. For years, that intelligence lived behind a dashboard that customers had to manually click through. As Jungle Scout's customers looked to unlock insights and deeper analysis from their data, Jungle Scout partnered with Caylent across three sequential engagements: a ML-based competitive intelligence engine, a Bedrock-powered Insights Copilot proof of concept, and a production Model Context Protocol (MCP) server, to make its data something an agent could query directly. The result is a fully launched platform built on data once locked inside bulk exports.
Solution Implemented
→ XGBoost classifier predicts brand winners across Amazon subcategories, paired with unsupervised clustering, plus SHAP explainability so every brand score comes with a defensible why.
→ Moved Jungle Scout's LLM inference layer onto Bedrock, using Claude and Amazon Nova Pro models, to build the security and cost foundation a customer-facing agent needed before going to production.
→ Bedrock-backed agent using Amazon Athena data tools with persona-aware routing across Brand Manager, Analyst, and Executive views, testing dual orchestration between deterministic and agentic workflows.
→ FastMCP server running alongside Jungle Scout's production AI service, authenticating customer agents through the same Amazon Cognito identity layer that secures the user interface.
→ Market Analysis tools for live data queries, workflow-as-tools that expose Jungle Scout's own multi-step diagnostic processes as single agent calls, and an early set of opinionated, actionable tools built for agents that take action on a customer's behalf.
→ An architecture where every MCP call is logged and attributable, something the prior bulk-export model never had.
Outcomes
→ Fast-tracked from beta to general availability across Jungle Scout's entire Cobalt customer base after early customer feedback.
→ Competitive diagnostics that once required significant analytical effort now take minutes.
→ Multiple one-call deal closes that were attributed directly to live MCP demonstrations.
→ Jungle Scout's engineering team can now extend and operate the platform independently, adding data sets and capabilities on its own.
Bringing Order to Amazon’s Infinite Shelf
More than a million brands and sellers have used Jungle Scout’s data to understand what's happening across Amazon's marketplace, tracking 600 million products in 24 subcategories to understand pricing, keyword trends, market share, and where competitors stand.
New sellers and shifting algorithms can reshape a category overnight, and Amazon's marketplace works like an infinite shelf where a brand's toughest competitors aren't always obvious. That intelligence separates a brand leading its space from one just reacting to it.
Business Challenge
For a decade, Jungle Scout delivered its intelligence in two ways. A UX product gave customers dashboards and custom analytics. A Data Cloud product handled bulk delivery into a customer's own business intelligence systems. Both meant logging in, clicking through, exporting, then doing the work of turning numbers into decisions.
“Our data is so powerful and so unique, this view into Amazon is something that no one else really has,” said Sam Johnson, Chief Technology Officer at Jungle Scout. “But customers didn't have the time to activate it and make informed decisions with it.”
That gap widened as Jungle Scout's most sophisticated customers started asking harder questions like who was gaining share in their category or where the next ad dollar should go. Jungle Scout's data could answer those questions, but a dashboard built for human clicks couldn't keep pace. Data Cloud had its own limits too. Once a customer exported data in bulk, Jungle Scout lost visibility into how it was used, with no way to price by consumption.
“The release of Claude’s Opus models in December of 2025 really changed a lot of people's perspective on what was possible with these agents,” Johnson said.
Jungle Scout had tried AI before and shelved it when customers didn't see the value. This time felt different, and the Jungle Scout and Caylent teams sat down together to put a plan into action.
"Caylent brought the expertise that got us to the frontier quickly, so we weren't spending six to nine months just trying to catch up. We got there much faster because of Caylent.”
Sam Johnson
Chief Technology Officer
Solution
Jungle Scout brought in Caylent, a partner from earlier exploratory projects, and the work unfolded as three connected engagements rather than a one-off project.
The first phase built the analytical foundation helping predict brand winners across Amazon subcategories and provide explainability for brand scores. Built to power premium advisory work for Jungle Scout's professional services team, the engine became the backbone the rest of the platform draws on.
The second phase moved Jungle Scout's LLM inference to Amazon Bedrock, running Claude and Amazon Nova Pro, and used that foundation to build an Insights Copilot proof of concept. The Copilot paired Amazon Athena data tools with persona-aware routing, tuning responses for a brand manager, analyst, or executive. That work gave Jungle Scout's engineers hands-on experience with Bedrock before any of it touched customer traffic.
The third phase, and the one Jungle Scout took to market, is a production MCP server built on the Model Context Protocol, the standard the industry has settled on for connecting data to large language models, whether through a customer's own chat experience or an outside provider.
The FastMCP server runs alongside Jungle Scout's production AI service, authenticating customer agents through the same Amazon Cognito layer that secures the user interface. Its tools fall into three groups. Market Analysis tools handle live data like sales estimates and keyword intelligence. Workflow-as-tools package Jungle Scout's own multi-step processes into a single agent call. A newer, opinionated set is built for agents that will eventually act on a customer's behalf. Every call gets logged and attributed, giving Jungle Scout the metering layer Data Cloud never had.
“It gets as close to the wow moment you’ve always hoped for in software when you're building something for customers,” Johnson said about the first time a customer connected an external LLM to the server. “Within a prompt or two, we could show them insights they'd been struggling to pull together on their own, and competitive opportunities their own company had considered before but never built a full business case around.”
That shift shows up in how customers diagnose competitive pressure. Isolating what's driving a market share gain or loss on Amazon used to take a level of expertise most brands didn't have. Now it takes minutes, and customers keep iterating in natural language until they find whether the issue is pricing, advertising, content, or reviews.
Working with Caylent
Johnson discussed the deciding factors in picking a partner for a space that's moving in real time. “One of the biggest criteria for me is flexibility, having a partner who can adapt as your needs change,” he said. “Caylent has always delivered on that.”
He also pointed to the people Caylent put on the project, saying Caylent holds a high bar for who it hires. Jungle Scout wanted a team that had done this work before, not one learning AI delivery on Jungle Scout's budget.
Speed mattered as much as expertise. AI vendors everywhere are racing toward the same frontier, and models change month to month. Jungle Scout couldn't afford months spent catching up. “Caylent brought the expertise that got us to the frontier quickly, so we weren't spending six to nine months just trying to catch up,” Johnson said. “We got there much faster because of Caylent.”
Johnson wanted more than a working product. “One of my goals was to have my team become experts,” he said. “They learned a lot and got to a point where they could drive this project on their own, adding data sets, adding skills, all the next-generation things we're building now. That's possible because Caylent helped us build that foundation. This was a great engagement and a great partnership.”
Technologies Used
Jungle Scout's MCP platform runs on Amazon Bedrock, using Claude and Amazon Nova Pro for LLM inference, with Amazon Cognito handling identity and authentication for agents connecting through the FastMCP server. Amazon Athena powers the data tools behind the Insights Copilot proof of concept. The competitive intelligence engine underneath draws on an XGBoost classifier, K-Means, GMM, and DBSCAN clustering, and SHAP explainability. Together, this stack lets Jungle Scout put one of the largest e-commerce intelligence datasets in the world, tracking 600 million products, directly in reach of AI agents built on the Model Context Protocol.
