A-CPO & A2-C2 (pronounced Atoo-Ceetoo)

Pronouns: It/ Them Functions: A2-C2 & A-CPO simplifies the claims intake process with AI, automation, fraud detection, and unparalleled efficiency. Key Traits: Follows Marc Benioff's 3 Laws of Trust.

Comment 1

The Curious Stewardship of A2- C2 & A-CPO

🎬 The Stellar Cast:

  1. Agentforce in dual roles A2-C2 and A-CPO
  2. AWS Textract
  3. AWS Comprehend
  4. AWS Simple Notification Services
  5. AWS Lambda
  6. AWS Document DB (Storing the output)
  7. Sagemaker
  8. AWS Fraud Detector
  9. AWS bedrock Agent
  10. Sharpoint online for knowledge

💡 Foreward

Reimagining the intricate world of data and claims processing into a seamless, efficient operation with agents. A2 - C2 automates data intake, streamlines customer onboarding, intelligently triages and prioritises claims, and employs advanced fraud detection to mitigate risk—all in one unified solution. This means a simplified experience that slashes operational costs, boosts processing speed and capacity, minimises errors through superior data accuracy, and provides unparalleled fraud detection and loss prevention. By integrating these capabilities, teams can focus on strategic decision-making rather than being bogged down by manual tasks. It’s not just an agent—it’s a smart partner that optimises workflows, minimises risk, and sets a new standard in operational excellence.

🤖 Prelude - The Inspiration

The initial claims reporting process remains a significant challenge for both policyholders and insurers. Traditional methods heavily rely on manual data entry, which is both time-intensive and error-prone. Additionally, document processing continues to impose substantial operational burdens. The insurance industry faces considerable financial losses due to fraud, costing billions annually and ultimately resulting in poor customer experiences.

Unverified claims lead to an average delay of 14 days for resolution, as repetitive back-and-forth data requests create dissatisfaction. Furthermore, these claims incur 23% higher processing costs because of the manual rework required. it is estimated that insurers save approximately $18 per claim by implementing verified policies, thereby reducing fraud payouts.

Insurers can analyse claim details in real time by introducing agents to extract structured and unstructured data from claim-related documents. This enables accurate identification of appropriate policies, driving the process toward verified claims. Automation also facilitates prioritisation of high-urgency claims, anomaly detection, and pattern recognition, as well as the analysis of large datasets to identify suspicious activities. These automated systems, operating 24/7, provide continuous support, enhancing efficiency and reducing overall costs.

🧱Plot Chemistry- How it was Built

The workflow includes the following steps: Episode 1: Claim Initiation

  1. The User starts initiates the claim by starting chat with Agentforce agent[A2- C2] available on the experience site.
  2. Agent [A2-C2] helps clarify any open questions by performing a semantic search on the documents maintained in sharepoint by integrating with AWS bedrock agent. This guides the user to start the claim process with confidence through a flow available within the experience cloud. This also solves the problem of maintaining a copy of the document in the data cloud.
  3. Agent[A2- C2 ] will escalate to a human agent if it not able to provide satisfactory answer. Episode 2: Data collection
  4. User starts the process and is able to provide all the required data through form and upload necessary documents and opt in for proactive updates on the channel.
  5. Create a new draft claim and add attachments Episode 3: Triage and Assignment
  6. The intelligent document reader [IDR]module (headless agent) is triggered when the user completes the process. It streams the uploaded documents to the S3 bucket for virus scanning. Infected documents are quarantined and removed from claims.
  7. Upload triggers a AWS lambda function (Invoke document analysis) that calls Amazon textract API for text extraction.
  8. On completion of text extraction, a notification is sent via Amazon SNS(simple notification service) triggering a lambda function which invokes amazon comprehend for custom document classification. This process also extract all the PII data and stores them back in the document DB indexed by claim number
  9. At the end of the document extraction process Lambda invokes a step function that calls theSageMaker inference endpoint is called for a fraud prediction mask of the input documents.
  10. Amazon Fraud Detector is called for a fraud prediction score using the data extracted from the mortgage documents.
  11. The results from Amazon Fraud Detector and the SageMaker inference endpoint are aggregated into the document DB available for retrieval for the agent.
  12. The document DB updates triggers a event which is published into platform events using lambda function.
  13. The PE event updates the case with the summarized results of the output. Final Episode: The Grind.
  14. Processor can invoke the Agentforce[A-CPO] which will invoke the AWS bedrock agent to summarize the results from document DB to retrieve the information and present the results back .
  15. Agentforce is able to pull PII data extracted and make intelligent recommendations to convert a unverified Policy into identiying where it should be linked
  16. Aid the backprocessor in decision making process and recommending next steps by leveraging the vector DB based semantic search. A- CPO help accelerate the processing

🌪️The Twist - Challenges

  1. Training data sets for Sagemaker
  2. Integrating with Amazon Comprehend using Agentforce for real-time.
  3. Financial service cloud would have simplified the data model problem. Ended up recreating lot of fields and objects

🦸Action Block(Adrenaline pumping music) - Accomplishements

  1. Built a Modular reusable agent that can be used/
  2. Integrate sagemaker with Data cloud Model builder.
  3. Leveraging semantic search using data in sharepoint.
  4. Leveraging agent to agent communication to retrieve answers
  5. Integration with Amazon Comprehend for Intelligent Document processing.
  6. Integrating with Amazon fraud detection models using headless agents.

💭Post Credits - What is next

Upgrade A2-C2 to act as a personalised omnichannel Insurance Advisor(including integration through Service Cloud Voice). It can simplify the process of understanding insurance needs and coverage by analyzing customer profiles and preferences. It also strengthens customer relationships by identifying relevant opportunities for tailored cross-selling. This will also help drive the operational cost by deflecting rudimentary questions from the call centre and giving back the time to focus on critical customer needs

Box office Numbers:

~$21 per claim in operational savings ~$25 per claim from fraud prevention ~$15 per claim in error reduction.

🙏Credits - Inspirations

https://developer.salesforce.com/blogs/2025/04/build-headless-agents-with-the-agent-api https://www.postman.com/salesforce-developers/salesforce-developers/collection/gwv9bjy/agent-api https://aws.amazon.com/blogs/machine-learning/automate-document-validation-and-fraud-detection-in-the-mortgage-underwriting-process-using-aws-ai-services-part-1/

PS: A2- C2 = Agentforce 2.0 based ClaimWhiz 2.0 | A-CPO = Agenforce Claim processing officer

Built With

Share this project:

Updates