Amazon’s reference sample turns the slowest part of deal evaluation into an orchestrated agent workflow. The AWS Machine Learning blog publishes a complete, deployable multi-agent system that screens acquisition targets, models their valuation, checks strategic fit, and audits every assertion against its source (Accelerating M&A due diligence with Amazon Bedrock AgentCore). This guide walks through the documented deployment so you can stand up the same architecture in your own AWS account.
What the sample actually does
M&A teams repeatedly pull financials, filings, and market data by hand, then reconcile them across deals. The reference post reports that the same gathering-and-summarizing loop, run by agents without human intervention, collapses work that previously needed weeks of analyst time into hours (Accelerating M&A due diligence with Amazon Bedrock AgentCore). The system is not a single prompt; it is a supervisor that routes work to four specialist agents, each grounded in citations the compliance team can trace.
Four operating pressures push teams toward this design. Slow review cycles delay every bid, fragmented data forces manual reconciliation, duplicated analysis wastes effort on each new deal, and governance requirements demand traceable, cited insights. The reference architecture answers each pressure with autonomous data gathering, intelligent routing, persistent memory, and built-in auditability (Accelerating M&A due diligence with Amazon Bedrock AgentCore).
The post presents two implementation paths. Amazon Quick is the integrated, fully managed option for standard business-intelligence patterns, while the custom AgentCore path gives fine-grained control over agent behavior, memory, and model selection (Accelerating M&A due diligence with Amazon Bedrock AgentCore). The open sample follows the custom path, which suits M&A teams with proprietary valuation methodologies or complex multi-agent coordination.
Prerequisites before you deploy
Before the stack builds, the post lists concrete requirements. You need an AWS account with Amazon Bedrock model access enabled for Anthropic Claude and Amazon Nova, the AWS CLI at version 2.15 or later, Python 3.11 or later, Node.js 20 or later, and the AWS CDK v2 installed globally (Accelerating M&A due diligence with Amazon Bedrock AgentCore). The agent container image is built on AWS CodeBuild, so Docker is not required locally. The documentation states the full deploy-run-cleanup cycle costs under USD 5.00, which keeps experimentation cheap (Accelerating M&A due diligence with Amazon Bedrock AgentCore).
Deploy into one of four supported Regions: us-east-1, us-west-2, ap-southeast-2, or eu-central-1 (Accelerating M&A due diligence with Amazon Bedrock AgentCore). Region and model access are verified automatically by the deploy script before any resource is created.
Step 1: Clone and deploy the stack
The complete implementation ships in the M&A Due Diligence Multi-Agent Sample repository, seeded with synthetic companies so no real financial data is involved. Clone it and run the deploy script for your platform:
# macOS / Linux
git clone https://github.com/aws-samples/sample-ma-due-diligence-agentcore.git
cd sample-ma-due-diligence-agentcore
./deploy.sh
# Windows
git clone https://github.com/aws-samples/sample-ma-due-diligence-agentcore.git
cd sample-ma-due-diligence-agentcore
.\deploy.ps1
The AWS team documents that first-time deployment takes 20 to 25 minutes, because the script verifies your Region and model access, creates a virtual environment, runs cdk deploy --all, seeds synthetic data, and executes a post-deploy smoke test (Accelerating M&A due diligence with Amazon Bedrock AgentCore). When it finishes, the project is installed in editable mode and registers the mna command-line tool you use in the next step.
Source: AWS Machine Learning — Accelerating M&A due diligence with Amazon Bedrock AgentCore
Step 2: Understand the agent topology
The reference architecture orchestrates a supervisor agent that coordinates four specialists through the Strands Agents SDK, with Amazon Bedrock AgentCore Runtime managing execution (Accelerating M&A due diligence with Amazon Bedrock AgentCore). The Target Screening Agent converts natural-language queries into SQL and runs them against Amazon Aurora PostgreSQL Serverless v2 through the RDS Data API. The Financial Analysis Agent applies discounted cash flow and comparable-company methods, pulling market multiples through an AgentCore Gateway-backed tool. The Strategic Fit Agent compares the current target against completed acquisitions stored in AgentCore memory, and the Compliance Validation Agent runs a citation-check evaluator on every response.
The user interacts through a Jupyter notebook or terminal interface, while AgentCore Runtime manages agent execution. AgentCore memory keeps session continuity and a long-term prior_deals namespace that stores lessons from completed acquisitions (Accelerating M&A due diligence with Amazon Bedrock AgentCore).
Source: AWS Machine Learning — Accelerating M&A due diligence with Amazon Bedrock AgentCore
The data layer combines structured and unstructured sources. Amazon documents that Aurora PostgreSQL Serverless v2, the engine the architecture provisions for target financials, delivers up to 6x the throughput of stock PostgreSQL on comparable hardware (Amazon Aurora overview), and the same engine is what the reference design queries through the RDS Data API (Accelerating M&A due diligence with Amazon Bedrock AgentCore). For unstructured documents, Amazon Bedrock Knowledge Bases index confidential information memoranda, financial statements, and governance checklists, then return passages the agents cite inline (Amazon Bedrock Knowledge Bases; Accelerating M&A due diligence with Amazon Bedrock AgentCore).
Step 3: Invoke the agents
The mna CLI sends prompts to individual specialists or lets the supervisor route across them. Screening uses text-to-SQL against Aurora plus narrative enrichment from the knowledge base. The post shows a query for transportation companies with revenue between 100 million and 500 million USD, EBITDA margin above 12 percent, and fleet size above 200, surfacing the top three with the leader’s growth trajectory from its CIM (Accelerating M&A due diligence with Amazon Bedrock AgentCore).
mna invoke target_screening "Screen the target pipeline for transportation companies with revenue between 100M and 500M USD and fleet size above 200. Surface the top three and tell me what the CIM says about the leader's growth trajectory." --session-id walkthrough-session-00000000-0001
Financial analysis runs a DCF on a target using its CIM, flags any management projection that diverges from historical performance by more than 20 percent, and pulls comparable multiples for the mid-market transportation and logistics segment (Accelerating M&A due diligence with Amazon Bedrock AgentCore). Strategic fit reads prior deals from the long-term prior_deals memory namespace, while the compliance agent invokes a Lambda-based citation-check evaluator that returns a structured pass or fail with counts of supported and unsupported claims.
The sample is a starting point, not a finished product. The post notes that teams with proprietary methodologies can extend the custom architecture with specialized internal systems or their own model selection, because AgentCore gives fine-grained control over agent behavior and coordination (Accelerating M&A due diligence with Amazon Bedrock AgentCore). Begin by editing the screening SQL templates and the valuation assumptions the Financial Analysis Agent documents inline.
Step 4: Inspect the audit trail
Every invocation produces an Amazon CloudWatch log stream and an AWS X-Ray trace that captures the full supervisor-to-specialist-to-tool call hierarchy (Accelerating M&A due diligence with Amazon Bedrock AgentCore). The trace lets a compliance reviewer see exactly which specialist answered, which tool it called, and which document backed each cited claim. Pair the trace with the citation-check evaluator’s pass or fail report to close the loop on governance.
How the citation guardrail works
Every assertion the agents emit is checked against a source. The Compliance Validation Agent calls a custom citation-check evaluator implemented as an AWS Lambda function, which examines each factual claim in a response and flags assertions that lack a supporting citation (Accelerating M&A due diligence with Amazon Bedrock AgentCore). Amazon Bedrock Guardrails enforce safety and response controls at the supervisor level, so specialist output still passes organizational policy before it reaches a user. Each agent invocation also writes an Amazon CloudWatch log stream and an AWS X-Ray trace capturing the full supervisor-to-specialist-to-tool call hierarchy for end-to-end observability (Accelerating M&A due diligence with Amazon Bedrock AgentCore).
Security model for confidential deal data
Deal data is among the most sensitive a company handles, so the architecture treats security as a first-class concern. Every IAM permission the agents can invoke is scoped to a specific Amazon Resource Name rather than a wildcard, and the AgentCore Gateway market-data tool enforces default-deny, attribute-based authorization through a Cedar policy engine that only permits the call when the requested industry code matches an approved vertical (Accelerating M&A due diligence with Amazon Bedrock AgentCore). The database tier lives in private, internet-isolated subnets behind VPC endpoints and is encrypted at rest with KMS, while Guardrails screen every supervisor response for harmful content and personalized financial advice (Accelerating M&A due diligence with Amazon Bedrock AgentCore). This layered model of IAM, Cedar, network isolation, and encryption covers different failure modes for systems that handle confidential deal data (Accelerating M&A due diligence with Amazon Bedrock AgentCore).
Where to take the architecture next
The same AgentCore building blocks appear in other production patterns. Our coverage of Amazon Bedrock AgentCore’s new temporal policies and rate limits shows how to bound agent behavior and cost beyond a single action, and running production AI agents in n8n with AgentCore demonstrates wiring the runtime into a low-code orchestration layer. Both build on the supervisor-and-specialist topology described here.
Bottom line
The AWS sample is a working, citation-checked template rather than a vendor pitch. Standing it up gives M&A teams a supervisor that routes to four specialist agents, a data layer on Aurora and Bedrock Knowledge Bases, and a compliance evaluator that refuses unsupported claims. Deploy the stack, invoke the mna CLI against synthetic targets, and adapt the specialist prompts to your own screening and valuation criteria.
