Automating Coaching Insights with GenAI & AWS

Professional coaches often struggle to scale because they spend hours manually analyzing fragmented client data and task completion rates. To solve this, we developed a serverless AI pipeline that automatically flags performance trends and drafts personalized messages based on real-time sentiment.

ClientB2B platform

PublishedApril 2026

IndustryB2B productivity

This case study explores the implementation of an event-driven architecture using Amazon Bedrock and the Nova model family to transform raw JSON data into actionable coaching summaries.

The client is an innovative B2B productivity and personal development platform designed to help professional coaches systematize their workflows and track client progress. Their platform bridges the gap between coaching sessions by keeping clients engaged and accountable through daily routines, goal tracking, and weekly reflections.

Coaches using the platform were spending an excessive amount of time manually reviewing fragmented client data—such as task completion rates and qualitative weekly reflections. Without a centralized, automated analytical tool, it was difficult to quickly gauge a client's true progress or identify underlying risks, such as burnout or declining engagement. The client's objective was to build an intelligent engine that could save coaches time, automatically flag critical insights, and proactively draft personalized messages tailored to the client's current performance and emotional state.

To address the client's challenges, we designed and implemented an architecture featuring a fully serverless, event-driven architecture on AWS.

From an infrastructure perspective, the solution leverages AWS managed services to ensure high availability, scalability, and strict cost control.

Amazon S3: Acts as the secure landing zone for incoming data snapshots and the storage for the generated AI insights.
AWS Step Functions: Serves as the central orchestrator of the entire pipeline. It manages the state and sequence of the multi-step analytical process, providing built-in error handling and retries.
AWS Lambda: Provides serverless compute power. Individual Lambda functions are deployed to handle specific, isolated tasks (e.g., parsing the JSON, communicating with the AI models, and formatting the final output).
Amazon Bedrock: The foundational generative AI service. It securely processes the prompts using the Amazon Nova Premiere model, applying necessary Guardrails to ensure the final generated messages adhere to strict safety and tone-of-voice policies.

During the development, we uncovered several crucial technical insights that heavily influenced the efficiency and cost-effectiveness of the final solution:

Bypassing RAG for Structured Data: Initially, a Retrieval-Augmented Generation (RAG) pipeline utilizing a vector database was considered. However, because the client's input data was provided in a highly structured JSON format, we determined that a complex RAG setup was unnecessary. Processing the structured data directly reduced both latency and infrastructure overhead.
Client-Driven Model Adoption: The client specifically requested the use of the Amazon Nova model family. We successfully integrated this via Amazon Bedrock, proving it to be highly capable of handling nuanced coaching instructions and delivering excellent cost-efficiency for the required workloads.
Optimizing Bedrock Guardrails: Security and content safety are paramount in coaching. While Amazon Bedrock Guardrails are essential, our cost analysis revealed they accounted for approximately 84% of the total cost per inference request. To optimize operational expenses, we architected the pipeline to apply Guardrails only to the final output generated for the client, rather than during intermediate drafting steps.

The implementation was a resounding success, validating that Generative AI can securely and effectively scale personalized coaching.

Cost Efficiency: Through model selection and Guardrail optimization, the total cost of processing and analyzing a weekly snapshot was reduced to roughly $0.014 – $0.017 per analysis.
Dynamic Contextualization: The AI engine successfully demonstrated the ability to adapt the tone and focus of its messaging based on specific performance criteria and sentiment, acting as a true co-pilot for the coaches.
Clear Path to Production: The client received a fully functional demo environment and a concrete integration roadmap for their Minimum Viable Product (MVP), utilizing Amazon DynamoDB and Amazon EventBridge to seamlessly connect the AI engine with their existing application.

AWS SERVICES

Amazon Bedrock

AWS Lambda

AWS Step Functions

Amazon S3

Key Takeaways

100%

Serverless

$0.016

Final cost to process

0

Lag

Igor is our go-to person for cloud services.

Ask him anything.