Case Study

Supply Chain SaaS Startup Compresses a Nine-Month AI Build into Seven Weeks with Claude on Amazon Bedrock

At a glance

A supply chain planning SaaS startup gives operational teams at its customer companies exception-driven tools for inventory decisions. But every specific question, like why one item’s forecast missed, required a data engineer to write custom SQL. As planning vendors race to add AI-native, conversational capability, that bottleneck slowed how fast customers got answers.

NMD turned the company’s proof-of-concept into a hardened, tenant-aware AI Copilot pairing a natural-language-to-SQL agent with a retrieval-augmented knowledge assistant on Amazon Bedrock. Business users now ask supply chain questions directly and get grounded, exportable answers in seconds.

What the CEO estimated as a nine-month internal build shipped in about seven weeks with NMD, at 84% AWS-funded cost, putting the company on a path toward an estimated $240,000 in annual AWS consumption at full production scale.

Industry

Use Case

Agentic AI, RAG (Retrieval Augmented Generation), Analytics & Business Intelligence

Solution implemented

The value equation

Company Snapshot

A supply chain planning SaaS company serving businesses of all sizes with AI/ML-based, exception-driven planning applications.

Location

United States

Customer Situation

This company sells AI/ML-based supply chain planning software to companies of all sizes, competing against both legacy planning giants and a wave of AI-native entrants. That position depends on how fast its own customers can turn platform data into decisions.

In practice, every specific supply chain question, like why one item’s forecast missed or which orders needed exception handling, required a business user to route the request to a data engineer for a custom SQL query. The company’s own CEO estimated that building this capability internally, without outside help, would take roughly nine months given competing engineering priorities.

Across supply chain planning software, vendors face growing pressure to replace static dashboards with conversational, self-service access to data, a shift analysts call the sector’s move toward AI-native planning.

NMD Solution

NMD reviewed the client’s existing proof-of-concept and found a working natural-language-to-SQL and RAG agent architecture that lacked the security, tenant isolation, and guardrails needed for production use. The solution required hardening the chat surface, wiring single sign-on with tenant-scoped enforcement, and adding SQL AST validation, intent triage, and loop prevention around the existing agent logic. NMD also productionized the RAG pipeline with automated ingestion, chunking, and versioning in Amazon OpenSearch, then deployed the full stack on Amazon Bedrock using Claude, Amazon Bedrock Agents, AWS Lambda, Amazon DynamoDB, and Amazon S3 running on ECS.

What We Delivered

The client’s business users ask supply chain and inventory questions in plain language and receive grounded, human-readable answers with CSV or Excel export, drawing on both live database data and a curated knowledge base. The copilot enforces read-only, tenant-scoped access through single sign-on identity handoff, with SQL guardrails, intent triage, and semantic validation preventing unsafe or out-of-scope queries. Runbook documentation and knowledge transfer equip the client’s technical team to operate, extend, and roll the solution out to additional customers as a standard, repeatable platform capability rather than a one-off engagement.

Ready to Transform Your Software Business?

See how New Math Data can transform your software business with AWS-powered innovation.