Finance ops · Production

BGL democratizes business intelligence with Claude Agent SDK and Amazon Bedrock AgentCore

The problem

BGL's business users without technical knowledge had to rely on the data team for queries, creating a bottleneck, while traditional text-to-SQL solutions failed to provide consistent or accurate results across their complex financial compliance data.

First attempt

Text-to-SQL agents that try to handle everything — schema understanding, joins, aggregations, and business logic — produce inconsistent results by joining tables incorrectly, missing edge cases, or generating wrong aggregations.

Workflow diagram · grounded in source
1
User submits question via Slack
trigger
“A user asks a business question using Slack (for example, Which products had the most negative feedback last quarter?).”
2
Schema discovery and SQL generation
ai_action
“The agent identifies relevant tables using skills and writes SQL queries.”
3
SQL security validation
validation
“a security layer allows only SELECT queries and blocks DELETE, UPDATE, and DROP operations”
4
Query execution in Athena
integration
“Athena executes the query and stores results into Amazon Simple Storage Service (Amazon S3)”
5
Result download to file system
integration
“The agent downloads the resulting CSV file to the file system on AgentCore, completely bypassing the context window to avoid token limits.”
6
Analysis and visualization via Python
ai_action
“The agent writes Python code to analyze the CSV file and generate visualizations or refined datasets depending on the business question.”
7
Response delivered in Slack
output
“Final insights and visualizations are formatted and returned to the user in Slack.”
8
Iterative skill refinement
feedback_loop
“the team will gather feedback from users, identify the gaps, and add new knowledge to existing skills using a human-in-the-loop process so skills are updated and refined iteratively”
Reported outcome

For BGL's more than 200 employees, the AI agent represents a significant shift in how they extract business intelligence — product managers can validate hypotheses instantly, compliance teams can spot risk trends without SQL, and the data team is freed to focus on strategic initiatives.

Reported metrics
employees benefiting from AI agentmore than 200 employees
Business intelligence accesssignificant shift in how they extract business intelligence
Data team query workloadfreeing the data team to focus on strategic initiatives rather than one-time query requests
Hypothesis validation speedvalidate hypotheses instantly without waiting for the data team
Reported stack
Claude Agent SDKAmazon Bedrock AgentCoreAmazon Athenadbt LabsAmazon S3PythonAmazon BedrockClaude CodeSlack
Source
https://aws.amazon.com/blogs/machine-learning/democratizing-business-intelligence-bgls-journey-with-claude-agent-sdk-and-amazon-bedrock-agentcore?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

For BGL's more than 200 employees, the AI agent represents a significant shift in how they extract business intelligence — product managers can validate hypotheses instantly, compliance teams can spot risk trends with…

What tools did this team use?

Claude Agent SDK, Amazon Bedrock AgentCore, Amazon Athena, dbt Labs, Amazon S3, Python, Amazon Bedrock, Claude Code, Slack.

What results were reported?

employees benefiting from AI agent: more than 200 employees; Business intelligence access: significant shift in how they extract business intelligence; Data team query workload: freeing the data team to focus on strategic initiatives rather than one-time query requests; Hypothesis validation speed: validate hypotheses instantly without waiting for the data team (source-reported, not independently verified).

What failed first in this deployment?

Text-to-SQL agents that try to handle everything — schema understanding, joins, aggregations, and business logic — produce inconsistent results by joining tables incorrectly, missing edge cases, or generating wrong ag…

How is this finance ops AI workflow structured?

User submits question via Slack → Schema discovery and SQL generation → SQL security validation → Query execution in Athena → Result download to file system → Analysis and visualization via Python → Response delivered in Slack → Iterative skill refinement.

Related finance ops cases