An AWS case study on agentic AI for Indian SMB business intelligence — onboarding automation, agent runtime, and persistent memory at SaaS scale
Wisemelon is a new-age business intelligence platform purpose-built for India’s 63 million small and medium businesses (SMBs). Unlike traditional BI tools that demand a dedicated data team, an analyst, and weeks of setup, Wisemelon delivers a daily briefing experience to SMB owners — a clear, prioritized list of what to focus on today and why. The platform functions like a Chief Strategy Officer on retainer, available 24/7, at the cost of a SaaS subscription, empowering business owners to move from gut-feel decisions to data-driven action without standing up an analytics function.
Business Intelligence and SaaS for SMBs (MarTech, FinTech, RetailTech)
Amazon Bedrock, Anthropic Claude (Sonnet and Haiku), Amazon Bedrock Knowledge Bases, Amazon Bedrock Guardrails, Bedrock AgentCore Runtime, AgentCore Memory (short-term and long-term), AgentCore Identity, AgentCore Gateway, AgentCore Observability, Amazon EventBridge Scheduler, Amazon OpenSearch Serverless, Amazon DynamoDB, AWS Lambda, AWS Fargate, Amazon API Gateway, Amazon S3, AWS Secrets Manager, AWS KMS, AWS IAM, Amazon Cognito, Amazon CloudWatch, AWS X-Ray.
Wisemelon required a fast, intelligent, and reasoning-first onboarding experience to support its rapidly growing base of Indian SMB customers. The system needed to seamlessly walk every new customer through deep business research, goal benchmarking, entity modelling, and data pipeline setup within hours rather than weeks, ensuring sub-second tool-call latency and zero drop-offs across long onboarding sessions. Achieving high availability and persistent context across multi-step onboarding conversations was a critical requirement, necessitating an agent runtime that could scale horizontally without burdening Wisemelon’s engineering team with self-managed inference infrastructure.
Onboarding precision and contextual understanding were key priorities to maintain a competitive edge against generic BI tools that rely on rigid templates. The platform required a sophisticated mechanism to interpret nuanced business descriptions across vague and complex inputs (such as “we’re a D2C apparel brand at ₹40L per month, 18 months old, mostly Instagram-driven”) and translate them into a tailored entity model, peer benchmark set, and connector configuration. Additionally, automation was critical to eliminate analyst dependency from the onboarding flow, reduce time-to-first-briefing, and enable thousands of new SMBs to activate concurrently without proportional headcount.
To balance cost and performance at SMB-scale economics, Wisemelon needed an architecture that scaled agent compute on demand and avoided always-on infrastructure costs that would erode unit margins on lower-tier plans. The architecture required intelligent guardrails to prevent hallucinated business profiles or benchmarks from contaminating customer accounts, while significantly reducing customer acquisition cost (CAC) and onboarding overhead. Furthermore, the solution demanded a secure and isolated environment for processing sensitive financial, customer, and operational data submitted during onboarding, ensuring a compliant foundation for expansion across regulated SMB segments.
Multi-Week Onboarding as the Primary Growth Bottleneck: Wisemelon’s manual onboarding flow took two to three weeks per customer — questionnaire intake, analyst review, entity model design, customer confirmation, connector setup, and pipeline validation. Activation drop-offs were highest in the first 7 days, and the analyst-led model could not scale beyond a few hundred concurrent onboardings without a proportional increase in headcount, capping the platform’s growth.
Generic Templates vs. Business-Specific Modelling: A logistics company, a salon chain, and a D2C apparel brand each need fundamentally different entity models, KPI definitions, and benchmark sets. Existing BI tools force customers into rigid templates, but Wisemelon’s differentiation depends on tailoring the data model and benchmarks to each business individually — something previously possible only with analyst time and impossible to deliver in hours rather than weeks.
CloudZenia supports you at every stage, delivering reliable, high-quality services at competitive prices.
CloudZenia worked with Wisemelon to design and ship an agentic onboarding pipeline that uses Amazon Bedrock as the reasoning core and Bedrock AgentCore for production-grade agent runtime, memory, and tool orchestration. Each onboarding stage is implemented as an autonomous agent, with structured handoffs between stages and grounded retrieval at every customer-facing step.
Wisemelon implemented Amazon Bedrock as the reasoning core of its onboarding experience, with Anthropic Claude Sonnet and Haiku handling deep reasoning and low-latency conversational tasks, while AgentCore Runtime hosts each stage as secure, isolated serverless agents.
A research agent enriches customer inputs using external signals, structured datasets, and knowledge bases, generating accurate business profiles in minutes instead of requiring time-intensive manual analyst research and validation workflows.
The benchmarking agent uses semantic search over peer-cohort datasets to generate contextual KPI targets with detailed rationale, replacing manual analyst benchmarking efforts and improving accuracy of performance expectations.
An intelligent agent generates tailored data models and refines them through guided multi-turn dialogue, replacing manual analyst-led modelling sessions with faster, accurate, and owner-validated schema creation workflows.
Connector setup is fully automated through AgentCore Gateway, enabling seamless integrations across multiple platforms with managed authentication, eliminating repetitive engineering effort and accelerating onboarding timelines significantly.
AgentCore Memory enables persistent context storage and structured coordination across multiple agents, ensuring seamless onboarding continuity while improving personalization and efficiency across all future customer interactions.
Guardrails ensure safe and reliable outputs by filtering prompt injection attempts, preventing hallucinated claims, and enforcing grounded responses across all onboarding interactions and customer-facing generated insights.
EventBridge Scheduler enables periodic agent execution for post-onboarding workflows without requiring always-on infrastructure, ensuring cost efficiency while maintaining reliable automation for recurring data processing and reporting tasks.
Secure infrastructure and observability tools ensure safe multi-tenant onboarding with strict data isolation, encrypted credential handling, and full visibility into agent performance, workflows, and real-time system behavior.
The AWS solution delivered transformative improvements across Wisemelon’s onboarding, commercial, and operational metrics within the first two quarters of production deployment.
Modern times need Modern Infrastructure Deployment
Embrace DevOps, Enhance Automation, and Deliver with Confidence.
Simplify, Scale, and Save with Cloud-Native Application.
Spend Smarter, Scale Faster with Our Cost Optimization Solutions.
Your Business is in the cloud? Secure it before its too late.
Amazon Bedrock provides foundation models (such as Anthropic Claude) via API, while Bedrock AgentCore is a separate set of managed services for deploying production AI agents on top of those models. AgentCore includes Runtime (serverless agent hosting), Memory (short-term and long-term context), Gateway (turning APIs into MCP-compatible tools), Identity (agent authentication), and Observability (tracing and monitoring). Wisemelon uses both: Bedrock for model inference and AgentCore for production agent infrastructure.
AgentCore uses consumption-based pricing (per-second of CPU and memory consumed during agent sessions, plus per-event memory pricing) rather than always-on infrastructure costs. Combined with serverless Lambda and EventBridge Scheduler-driven invocations, infrastructure cost scales linearly with active customer count rather than peak provisioned capacity — a critical requirement for SMB unit economics.
At a target scale of 10,000+ SMBs each running daily intelligence pipelines, the operational overhead of self-managed agents (memory persistence, multi-agent routing, OAuth handling, scheduled execution, observability) would have consumed Wisemelon’s engineering team. AgentCore provides these as managed services with consumption-based pricing, allowing the team to focus on the intelligence layer rather than infrastructure.
Yes. The pattern — Bedrock for reasoning, AgentCore Runtime for hosting, Memory for context, Gateway for connector authentication, Knowledge Bases for grounded retrieval, and Guardrails for output safety — generalises to any SaaS product where onboarding requires business-specific data modelling, multi-connector authentication, and personalised configuration. CloudZenia delivers similar agentic onboarding implementations as part of its AgentOps on AWS practice.
Opcodera is a software product company building developer-led, cloud-native solutions for digital marketing, link management, and customer engagement. Its flagship product, EQRGen, is a leading QR code generation and URL shortening platform designed to empower....
A data organization tool is designed to process large volumes of raw data from diverse formats such as PDFs, images, and documents. Using advanced data extraction techniques, the tool analyzes and extracts meaningful information from unstructured sources...
CloudZenia can help you wherever you are in your cloud journey. We deliver high quality services at very affordable prices.