Marketing ops · Production

CLICKFORCE reduces advertising industry analysis from weeks to one hour with Amazon Bedrock Agents

The problem

CLICKFORCE's industry analysis process took two to six weeks due to labor-intensive manual stages and disconnected tools, while LLMs produced generic rather than actionable industry-specific insights and lacked integration with internal datasets.

First attempt

Generic LLM outputs lacked actionable, industry-specific intelligence, and foundation models used for Text-to-SQL translation were inflexible and often inaccurate, prompting CLICKFORCE to build a fine-tuned, grounded solution.

Workflow diagram · grounded in source
1
User query via chatbot
trigger
“End-users interact with a chatbot interface that runs on Amazon ECS, developed with Streamlit and fronted by an Application Load Balancer”
2
Route query to Bedrock Agent
routing
“When a user submits a query, it is routed to an AWS Lambda function that invokes an Amazon Bedrock Agent”
3
Knowledge Base retrieval
ai_action
“The agent retrieves relevant information from a Amazon Bedrock Knowledge Bases, which is built from source documents—such as campaign reports, product descriptions, and industry analysis files—hosted in Amazon S3. These documents are aut…”
4
Text-to-SQL query generation
ai_action
“the Bedrock Agent generated JSON schemas via the Agent Actions API Schema. These were passed to Lambda Executor functions that translated requests into Text-to-SQL queries”
5
AWS Glue data sync
integration
“With AWS Glue crawlers continuously updating SQL databases from CSV files in Amazon S3, analysts were able to run precise queries on campaign performance, audience behaviors, and competitive benchmarks”
6
SageMaker pipeline tuning
feedback_loop
“By using SageMaker, the team processed data, evaluated different approaches, and tuned the overall Text-to-SQL pipeline. Once validated, the optimized pipeline was deployed through AWS Lambda functions and integrated back into the agent.…”
7
Automated report generation
output
“Users can generate industry analysis reports through natural language conversations and iteratively refine the content by continuing the dialogue”
Reported outcome

Industry analysis time dropped from two to six weeks to under one hour, and operational costs fell by 47 percent, while brand owners, agencies, analysts, and media partners can now independently generate insights without waiting for centralized analyst teams.

Reported metrics
Industry analysis timeunder one hour (previously two to six weeks)
Operational cost reduction47 percent
Reported stack
Amazon BedrockAmazon Bedrock AgentsAmazon SageMaker AIAmazon OpenSearchAWS GlueAmazon ECSStreamlitAWS LambdaAmazon Bedrock Knowledge BasesAmazon S3MLflow
Source
https://aws.amazon.com/blogs/machine-learning/how-clickforce-accelerates-data-driven-advertising-with-amazon-bedrock-agents?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Industry analysis time dropped from two to six weeks to under one hour, and operational costs fell by 47 percent, while brand owners, agencies, analysts, and media partners can now independently generate insights with…

What tools did this team use?

Amazon Bedrock, Amazon Bedrock Agents, Amazon SageMaker AI, Amazon OpenSearch, AWS Glue, Amazon ECS, Streamlit, AWS Lambda, Amazon Bedrock Knowledge Bases, Amazon S3.

What results were reported?

Industry analysis time: under one hour (previously two to six weeks); Operational cost reduction: 47 percent (source-reported, not independently verified).

What failed first in this deployment?

Generic LLM outputs lacked actionable, industry-specific intelligence, and foundation models used for Text-to-SQL translation were inflexible and often inaccurate, prompting CLICKFORCE to build a fine-tuned, grounded…

How is this marketing ops AI workflow structured?

User query via chatbot → Route query to Bedrock Agent → Knowledge Base retrieval → Text-to-SQL query generation → AWS Glue data sync → SageMaker pipeline tuning → Automated report generation.