← Insights

Agentforce for Service: How to Handle 1M Support Requests

A use-case spotlight on Agentforce for Service — the three layers that make it work in production: setup, measurement, and MCP-powered orchestration.

You've got 1,000 support requests landing today. How many will your agents handle — and I don't just mean the human ones?

Here's how Salesforce is reshaping the future of service, distilled into three layers worth watching:

1. Real-world use case — by Jim Roth

Jim walks through how Agentforce handled over 1 million support requests in a single deployment. The piece is worth reading specifically because it explains the failure modes that show up at that scale — the things you can't see in a 50-ticket pilot. Read it here.

2. Operational playbook — by Bernard Slowey

The 7 support metrics that actually matter when AI is in the loop. Most teams are still measuring AHT and CSAT. Bernard's argument: those metrics tell you about the human flow, not the agent flow. Read it here.

3. The technical layer — MCP

If your agents don't have memory or planning, they're just guessing faster. MCP changes that — by giving agents a structured way to call external tools, hold state, and plan multi-step actions. Watch the explainer.

Where this is heading

Agents talking to other agents. Your support agent connecting via API to your billing system, your shipping vendor, your fraud detection layer — without a human relay in between. The orgs that lay this plumbing in 2025 will run support operations in 2026 that the rest of the industry can't match.

Agentforce for Service: How to Handle 1M Support Requests — Aventary