Automation and AI for a Small E-commerce Support Team
UrbanGoods is a small e-commerce store selling home goods, with 5 support agents using Service Cloud. They handle 500 cases per month, mostly order status and return requests. They want to use Agentforce (AI) to automate responses and reduce agent workload. They have Service Cloud Enterprise and are evaluating Agentforce. They also have a Flow that creates a case from email. They are concerned about AI accuracy and cost. They want to start with a pilot for order status inquiries, which make up 40% of cases. They have a knowledge base with 50 articles, but it's outdated.
1. What is the best way to prepare for an Agentforce pilot for order status inquiries?
Read the worked answer
Reasoning: Agentforce uses AI to automate responses. For order status, the AI needs access to real-time order data. They should first ensure that order data is integrated with Salesforce (e.g., via B2C Commerce or custom object). They need to update the knowledge base because AI relies on accurate articles. They should also define the scope: only order status, not returns yet. They should set up a Flow or Apex to fetch order status from the e-commerce platform and expose it to Agentforce via a custom action. Agentforce can be configured with topics and actions. They should create a topic for order status and an action that calls the order status API. They need to train the model with sample utterances. They should also set up a pilot group of users or a percentage of traffic. They must monitor accuracy and have a fallback to human agent. Since they have Service Cloud Enterprise, they may need to purchase Agentforce add-on. They should start with a small pilot, measure deflection rate and CSAT, and iterate. Key steps: data integration, knowledge update, Agentforce configuration, pilot, measure.
2. How should they address the outdated knowledge base and ensure AI responses are accurate?
Read the worked answer
Reasoning: The knowledge base is outdated, which will lead to inaccurate AI responses. They must update it before or during the pilot. They should audit the 50 articles, identify outdated ones, and update them. They can use Salesforce Knowledge (available in Service Cloud) to manage articles. They should also set up a process for regular review. For AI, they can use Einstein Article Recommendations or Agentforce's knowledge grounding. Agentforce can be grounded on knowledge articles. They should ensure articles are written in a customer-friendly tone and cover common order status questions. They should also consider using Flow to fetch real-time order status instead of relying solely on knowledge, as order status is dynamic. For static info like return policy, knowledge is fine. They should also implement a feedback loop: when AI gives an answer, allow customers to rate it, and use that to improve. They should also set up a dashboard to track AI accuracy and escalation rate. Since they have a small team, they can assign one agent to own knowledge updates. They should also consider using Agentforce's testing tools to simulate conversations before go-live.