Transforming Query Indexing with AI Integration

This document outlines an innovative AI-driven process to enhance query indexing in the banking industry. The goal is to replace manual tasks with an AI assistant, significantly improving turnaround time, accuracy, and customer experience. The AI assistant will read, classify, and process customer queries, integrating seamlessly with existing CRM and workflow systems.

Step 1

In the banking sector, query indexing serves as the initial step in back-office operations. Each request, whether it involves account closure, mandate changes, address updates, or applications, arrives as an email or form. These requests must be read, comprehended, and validated before processing. Currently, this high-volume task affects turnaround time, compliance accuracy, and customer experience.

Step 2

Currently, around 30,000 cases are processed monthly. Approximately 70% are routed to bots; however, about 25% of these fail and require rework. Human agents manually interpret emails, switch to CRM for sender and product validation, then revert to Pega to select and submit the correct workflow.

Step 3

This manual process leads to an error rate above 20%, with structured forms still experiencing close to 15% errors. Dependencies on VDA sessions add friction, and human capacity limits throughput. In the desired state, an AI employee serves as the primary indexer, automatically reading and classifying emails and forms, validating details via CRM integration, and orchestrating the appropriate workflow within Pega.

Step 4

With this AI setup, 95% of cases are managed without human intervention, halving error rates and significantly improving turnaround times. Human intervention is needed only for exceptions, creating a scalable, reliable, and consistent back-office operation using AI. Let's examine the process flow shift from the current to the desired state. Currently, a human agent manually reviews emails, checks in Pega and CRM for validity, and updates the case manually.

Step 5

Currently, customer communication is handled manually, leading to fragmentation, delays, and errors. In the desired state, the process is transformed end-to-end using AI, where the AI assistant immediately takes over upon email receipt. An entity extraction agent identifies key details such as account number, sender name, request type, and amount from the email body and attachments.

Step 6

The external tool caller agent, integrated with CRM and Pega, validates and cross-checks contact and product information. The case handling agent selects the correct workflow and creates a case once validation is complete. Finally, the escalation and communication agent manages exceptions or sends responses back to the customer, shifting the process from manual and reactive to AI-driven, automated, and focused on exceptions, ensuring speed, accuracy, and compliance.

Step 7

Having explored the current challenges and envisioned the future, let's demonstrate how the AI employee operates in the indexing use case. Observe the sample emails drafted as requests for business account closure, mandate amendments, and payment queries. These emails are sent to a designated bank email, simulating customer queries.

Step 8

The AI employee continuously monitors the mailbox. When an email arrives, it retrieves the email body and attachments. The email will now be sent to another address.

Step 9

Upon email arrival, the AI employee triggers a workflow within its persona, executing backend agents. The AI retrieves email content, including the body and attachments. As the email arrives, the AI prepares to activate its agentic mesh.

Step 10

With the content retrieved, the agentic mesh, comprising multiple collaborative agents, extracts the request type, sender name, account details, and requested actions. An external tool caller agent, integrated with Salesforce, creates cases and contacts based on customer requests.

Step 11

Once the content is uploaded, the agentic mesh executes seamlessly.

Step 12

In this example, a business account closure request is processed by the AI. The AI identifies the request type, sender, account details, and actions needed, creating a contact in Salesforce accordingly.

Step 13

The final agent fetches detailed information from Salesforce. This demo illustrates how vocations can be tailored to specific requests and applications, enabling comprehensive agentic AI transformation for query indexing.

Step 14

Emma's AI employee transforms processes end-to-end, minimizing manual intervention while maximizing speed, efficiency, and accuracy through advanced AI workflows.