Use this scenario to understand how to build a car loan AI Agent using the features within the Automation 360 platform to automate complex business tasks, enhance decision-making, and improve efficiency in the loan approval process.

The following scenario highlights the power of AI in building the car loan process, demonstrating how Salesforce, Automation Co-Pilot, AI Agent Studio, and Microsoft Teams work together to create a seamless and customer-centric experience:

Users in this scenario

  1. Paul: Wants to buy a car
  2. Vincent: Salesperson in the dealership
  3. Natalia: The loan underwriter
  4. Marcus: A professional developer

Scenario summary

Visiting the car showroom
Paul is greeted by Vincent, a knowledgeable salesperson who is experienced in guiding through the car-buying journey. They discuss Paul's dream car and financial details. Vincent creates a profile for Paul in Salesforce, the dealership's CRM system.
Sharing the essentials: W2 and financial documents
To start the car loan process, Vincent requests Paul's W2 to verify his income. Paul provides his W2, which outlines his annual earnings. Instead of manually entering the data from Paul's W2, Vincent leverages Automation Co-Pilot, an AI-powered tool integrated with Salesforce. Vincent uploads the W2 via Automation Co-Pilot. This uses Document Automation and generative AI to extract the key information from the W2, streamlining the process and reducing the risk of manual errors.
Document processing through Document Automation
While Paul explores his potential new car, Automation Co-Pilot scans his W2. Document Automation, using generative AI, extracts key data such as employer name, income, and tax deductions. This automation eliminates manual data entry and reduces errors.
Behind the scenes: Data flows to the AI recommendation agent
The extracted W2 data is securely transferred to the AI recommendation agent, a specialized AI Agent developed in AI Agent Studio. This AI Agent along with other AI Agents are orchestrated within the Process Composer to get the desired output.
Process Composer is used to orchestrate AI Agents across complex enterprise processes with scheduling, and agent decisioning. This orchestration helps achieve the desired output from multiple AI Agents.

The combined power of orchestration using Process Composer and human validation using Automation Co-Pilot capabilities ensures that AI-driven processes are not only efficient but also maintain the necessary level of human oversight for important business decisions.

Notification sent to underwriter
Natalia, the loan underwriter, receives a notification on Microsoft Teams about Paul's application. She benefits from the AI recommendation agent's analysis. The AI recommendation agent has taken the information it gathered from Paul's application materials and, using an AI Skill, has compiled that information and used it as input for a prompt to an LLM, asking the LLM to generate a recommendation.
Intelligent automation role
Earlier in the process, Marcus, the developer, played a crucial role in building and configuring the AI recommendation agent. He uses AI Agent Studio to connect the agent to various large language models (LLMs) and fine-tunes the agent's prompts for accurate recommendations.
AI analyzes loan recommended to buyer
The AI Recommendation Agent, connected to a grounded model, analyzes Paul's financial profile based on the W2 data. It generates a personalized loan offer tailored to his specific needs, considering income, desired loan amount, and preferred terms.
Swift loan approval
Natalia reviews the AI-generated recommendation and quickly approves Paul's loan, benefiting from a clear and concise presentation of the data.
Happy buyer with new electric vehicle
Thanks to the efficient AI-powered process, Paul drives off in his new electric vehicle, satisfied with the efficient and personalized experience.

Detailed scenario

The following image, table and procedure summarizes how Marcus creates a car loan AI Agent to streamline the car loan process. It leverages the capabilities of Process Composer to orchestrate various AI Agents and actions, creating a seamless workflow. By automating tasks such as data extraction, document processing, and decision-making, the AI Agent significantly reduces processing time by 50 to 60 percent and improves accuracy.


Car loan AI Agent
ProcessDescription
oneData collection and initial processing
  • Sales interaction: The process commences with a salesperson engaging with a potential customer interested in securing a car loan.
  • Document collection: The salesperson collects essential information from the customer such as W2 details and financial data.
  • Document upload: The collected documents are uploaded to Salesforce CRM.
twoAutomation Co-Pilot and data extraction: Automation Co-Pilot, embedded in Salesforce, uses Document Automation driven by generative AI to extract key data points from the uploaded documents. Automation Co-Pilot enables real-time communication between users and automations. Using Automation Co-Pilot, you can automate repetitive and time-consuming tasks, navigate controls, and input data when interacting with a customer.
threeReview and approval process
  • Information routing: The extracted information passes through a series of steps and ultimately reaches a car loan approver.
  • Notification: In this example, the loan approver, who typically works within Microsoft Teams, receives an automatic notification when a new loan application is ready for review. The approver is shown information collected from multiple sources both structured and unstructured data. A recommendation that includes vital information such as the applicant's credit score, loan amount, suggested term, interest rate and monthly payments is displayed.
AI Agent 1 - Credit checkerThis API Task based AI Agent uses data extracted from documents such as W2s to assess the applicant's creditworthiness and overall financial standing.
AI Agent 2 - AI recommendation agentThis API Task based AI Agent employs an AI Skill to generate car loan recommendations specifically to the loan underwriter, drawing insights from customer data. The AI Skill is powered by an LLM, potentially grounded in relevant data like income brackets, loan amounts, terms, and interest rates to ensure the accuracy and relevancy of the recommendations. The output of the AI Skill, often structured in JSON format for seamless integration, forms the basis of the recommendation presented to the underwriter within their workflow, such as Microsoft Teams. A detailed example of building this AI Agent can be found in Building an AI Recommendation Agent.
AI Agent 3 - Approve/Reject email notificationThis API Task based AI Agent automates communication by automatically sending an email to the customer upon the loan underwriter's decision (approval or rejection). This agent ensures timely updates to the customer, keeping them informed throughout the process.
AI Agent 4 - Loan InitializationThis API Task based AI Agent initiates the loan process based on the customer's chosen loan term. This sets the loan's framework in motion, incorporating the customer's preferences from the outset.

Automation development

  1. Identifying needs and objectives: Marcus uses Process Composer an environment that leverages multiple products across the Automation 360 platform to deliver an Intelligent Automation experience.
    1. It provides a drag-and-drop interface that professional developers can use to create process flows and configure when to execute an automation. Additionally, it displays correct data to users and exchanges data across multiple teams.
    2. Marcus begins by understanding the specific requirements and objectives of the car loan system. This involves collaborating with stakeholders such as sales representatives (Vincent), loan underwriters (Natalia), and potentially even customers (Paul) to gather insights into their needs and pain points. For instance, Marcus needs to determine what data points are crucial for loan applications, the criteria for loan approval, and the desired level of automation.
  2. Selecting appropriate AI technologies: Based on the requirements, Marcus chooses the most suitable AI technologies and tools.
    1. AI Agent Studio: AI Agent Studio empowers Marcus to create, train, and deploy intelligent agents that automate tasks and make decisions.
    2. Large language models (LLMs): Marcus connects to pre-built connectors for LLMs such as Azure OpenAI, Bedrock, and Google Vertex through Model connections within AI Agent Studio. An alternative option for Marcus is to explore the option of connecting to custom-built models hosted in private clouds.
    3. Document Automation: Leveraging Document Automation with generative AI is essential for extracting relevant data from customer documents, such as W2 forms.
    4. API Tasks: Incorporating API Tasks within the agents allows faster executions, seamless interaction with external systems, such as credit bureaus, for pulling financial data. For more information, see API Task.
  3. Building the AI recommendation agent: This agent is crucial for analyzing applicant data and generating personalized loan recommendations. Marcus uses AI Agent Studio to define the agent's workflow and decision-making logic. He connects the agent to an AI Skill that leverages an LLM grounded in a relevant dataset. Marcus carefully crafts the prompt to guide the LLM's analysis and ensure accurate recommendations.
  4. Integrating with existing systems: To expedite the workflow, Marcus integrates the AI Agents with existing systems, including:
    1. Salesforce: This integration allows seamless data flow from the customer interaction stage managed by sales representatives such as Vincent to the loan processing stage.
    2. Microsoft Teams: Connecting to Teams enables real-time notifications and facilitates communication between different stakeholders, such as Natalia the loan underwriter.
  5. Ensuring data security and governance: Marcus leverages the AI Governance capabilities built into AI Agent Studio to ensure responsible and compliant use of AI. This involves monitoring model usage, tracking token consumption, and potentially implementing data masking techniques to protect sensitive customer information. For more information, see AI Governance.
  6. Testing and refining: Before deployment, Marcus rigorously tests the AI Agents to ensure their accuracy, efficiency, and adherence to compliance standards. This involves running simulations with sample data and gathering feedback from stakeholders to fine-tune the agent's performance.