Case Study: Goyal Brothers Prakashan

Despite its scale, GBP’s marketing operations depended on manual, chat-and-call-based support for its field force. A small central operations team fielded pricing, catalogue, dispatch, and demo-asset requests over WhatsApp, phone, and email — a model that could not keep pace with the seasonal nature of the education business.

Key challenges included:

• Manual, unstructured support over WhatsApp, calls, and email, causing response delays of more than 24 hours during the school-adoption season.
• An approximately 3x seasonal spike in request volume (60–200+ requests/day) that the operations team could not absorb without adding headcount.
• Frequent misrouting between editorial, creative, logistics, and digital teams, since most requests carried no structured ticket, category, or priority.
• Institutional knowledge concentrated in a small number of senior staff, creating single-point-of-failure risk and slow onboarding for new field representatives.
• No unified audit trail — requests and resolutions lived in personal chat threads and inboxes, limiting visibility and compliance review.

These constraints directly reduced field-sales productivity, delayed customer engagement, and created a measurable risk of lost school adoptions during the business’s most revenue-critical window.

To address these challenges, i2k2 Networks designed and deployed GBP Marketing Sahayak — a production agentic AI helpdesk built on Amazon Bedrock AgentCore. Unlike a scripted chatbot, it runs a genuine multi-agent loop: a Sahayak Supervisor agent interprets each request, plans the work, and delegates to four specialist agents — Knowledge, Diagnostics, Action, and Routing & Escalation — built with the open-source Strands Agents SDK.

Key Features and Capabilities:

For Field Sales & Marketing Representatives:

• Multilingual (Hinglish, Hindi, and English) chat assistant for instant, cited pricing, catalogue, and dispatch-status answers.
• Autonomous read-only lookups against CRM, ERP/dispatch, and campaign systems to diagnose issues before the rep even asks.
• Guided, confirmed ticket creation, collateral requests, and demo-link generation — no more waiting on the central operations team.

For Marketing Operations & Management:

• Every request becomes a structured, de-duplicated, categorised ticket with full diagnostic context already attached.
• Intelligent escalation routes only genuinely ambiguous or out-of-policy cases to the correct human team, with full context carried forward.
• A complete, replayable audit trail of every agent decision, tool call, and confirmation for compliance and responsible-AI review.

• Amazon Bedrock AgentCore Runtime: Secure, serverless, session-isolated hosting for the multi-agent Strands crew — each session runs in its own microVM, destroyed and sanitised at session end.
• Amazon Bedrock AgentCore Gateway: Converts GBP’s CRM, ERP/dispatch, helpdesk, and campaign APIs into governed MCP tools with per-tool authorization, split into separate read-only and write-only gateways.
• Amazon Bedrock AgentCore Memory: Short-term session memory and long-term rep profile/preference memory that persists across sessions.
• Amazon Bedrock AgentCore Identity: Validates inbound Amazon Cognito identity and manages secure outbound credentials to enterprise systems.
• Amazon Bedrock AgentCore Observability: End-to-end tracing of every agent reasoning step, tool call, and model invocation into Amazon CloudWatch.
• Amazon Nova (on Amazon Bedrock): Multilingual foundation model powering planning, reasoning, and reliable structured tool selection.
• Amazon Bedrock Knowledge Bases: Managed RAG returning grounded, cited answers from GBP’s pricing and policy documents.
• Amazon Bedrock Guardrails: PII masking, denied-topic filtering, and contextual grounding checks that block ungrounded pricing/policy answers.
• Amazon Aurora PostgreSQL Serverless v2 (pgvector): Auto-scaling managed vector store underlying the Knowledge Base, reached via the RDS Data API.
• AWS Lambda: Integration middleware for enterprise APIs, plus event-driven supporting workloads such as Knowledge Base sync and the weekly responsible-AI quality evaluation.
• Amazon API Gateway: Secure entry point for the web chat channel and the backend integration APIs.
• Amazon Cognito: Field-force authentication, issuing JWTs with role and region carried as verified attributes that scope agent entitlements.
• Amazon DynamoDB: Session state, ticket-reference mapping, and other fast-access application data, on on-demand capacity.
• Amazon S3 and Amazon CloudFront: Static frontend hosting and the operational knowledge corpus, with cross-region replication into a secondary AWS region for disaster recovery.
• AWS IAM: Least-privilege roles scoped per agent and per tool.
• AWS Secrets Manager and AWS KMS: Secure third-party credential rotation and customer-managed encryption keys.
• GitHub Actions and Amazon ECR: CI/CD pipeline building and deploying the AgentCore container image.

• Amazon CloudWatch: Monitors platform health, latency, tool-call success, and error logs, with alarms on latency and error-rate anomalies.
• AWS CloudTrail: Full API-level audit logging, archived to Amazon S3.
• AWS IAM Policies: Least-privilege, per-agent, per-tool role-based access to safeguard sensitive marketing and enterprise data.
• AWS WAF + Amazon CloudFront: Perimeter protection for the web-facing channel.
• End-to-end encryption: AWS KMS customer-managed keys and TLS in transit protect rep and enterprise data at rest and in motion.
• AWS Secrets Manager: Rotates third-party enterprise credentials — no credentials in code or prompts.

• Marketing-operations response time cut from 24+ hours to near-instant for self-served queries, with early production data already tracking ahead of the 6-month target.
• Field reps get consistent, always-available support in the Hinglish, Hindi, and English they actually use — including outside office hours, when school visits actually happen.
• Every request is now a structured, de-duplicated ticket with full diagnostic context, giving operations leadership visibility it never had with chat-based support.
• Reduced dependency on a small group of senior staff for institutional knowledge, lowering single-point-of-failure risk and cutting new-rep ramp-up time.
• Serverless, consumption-based architecture absorbs the ~3x seasonal adoption-window spike with zero added headcount.

i2k2 Networks Pvt. Ltd. is an AWS Consulting Partner based in Noida, India, specializing in cloud managed services, migration, and applied AI solutions, including agentic AI systems built on Amazon Bedrock and Amazon Bedrock AgentCore.