Self-Service Deflection with an Agentforce Service Agent

The scenario

Falcon Outdoors runs a busy help site. Look at a week of cases and a pattern jumps out: "Where's my order?", "How do I start a return?", "How do I reset my password?" — the same dozen questions, over and over, each one a human agent typing an answer they've typed a hundred times. We'll stand up an Agentforce service agent on the web and messaging channels that answers those from your Knowledge base, and transfers to a human through Omni-Channel the moment it's out of its depth.

For beginners: an Agentforce Service Agent is an autonomous, customer-facing AI agent for support. It reasons over your instructions and grounds its answers in your business data through a Data Library (retrieval-augmented generation over Knowledge articles and files), and every response runs through the Einstein Trust Layer. It is not a keyword chatbot with scripted flows — it plans, retrieves, and responds in natural language, within the guardrails you give it.

Step 1 — Ground it on Knowledge

An ungrounded agent guesses; a grounded one cites. In Agent Builder, attach a Data Library and point it at your Knowledge articles (and any support PDFs). Salesforce indexes and chunks that content and builds the retriever for you — you don't manage embeddings by hand.

• Publish the articles that actually answer your top questions first — deflection is only as good as the content behind it.
• Keep articles single-topic and current. A stale "returns policy" article becomes a confidently wrong answer.
• Set the agent's tone and scope in its instructions: what it may answer, and what it must escalate.

Step 2 — Deploy to web and messaging channels

The same agent can serve multiple front doors. Connect it to a Messaging for In-App and Web deployment so it appears as a chat widget on the help site, and to whichever messaging channels your customers already use.

1. Create the Embedded Service / Messaging deployment and connect your agent to it.
2. Configure the pre-chat and branding so it feels native to your site.
3. Test the real journeys — order status, returns, password reset — as a customer would type them, including messy phrasing.

Step 3 — Design the escalation handoff

Deflection without a good exit door frustrates customers. Agentforce ships an Escalation subagent (topic) for exactly this: when the agent can't resolve an issue — or the customer asks for a person — it transfers the conversation to a human queue using an outbound Omni-Channel flow.

• Add the default Escalation topic and tailor its instructions to capture the context you want carried over (order number, what was already tried).
• Route the transfer through Omni-Channel so it lands with the right skilled queue, not a black hole.
• Define clear escalation triggers: repeated failure to answer, explicit request for a human, or sensitive intents (billing disputes, cancellations).

The handoff should feel seamless — the human agent picks up with the full transcript, so the customer never repeats themselves.

Make it trustworthy — and prove the ROI

• Measure deflection rate: the share of conversations fully resolved without a human. That single number is your business case.
• Watch escalation quality: when the agent hands off, did it capture enough context? Review transcripts weekly at first.
• Close the content gap: every question the agent couldn't answer is a missing or unclear Knowledge article. Feed that back into your content.
• Stay honest with customers: make it clear they're talking to an AI agent and that a human is one request away.

Sources: Trailhead: Get Started with Agentforce Service Agent · Trailhead: Use Knowledge in Agentforce for Service · Transfer Conversations with an Omni-Channel Flow (Salesforce Help)