Agentforce can triage cases, answer customers and help reps, but only when it has good data and clear limits. These are the six things we check before building any agent.
1. One clear job
Start with a single, narrow task, like classifying incoming cases or answering order-status questions. Agents that try to do everything tend to do nothing well. Write down what success looks like in numbers, such as time to first response.
2. Clean records
If your case, account or product data has duplicates, empty fields or out-of-date values, the agent will repeat those mistakes confidently. Fix the fields the agent will read first.
3. Knowledge that's current
For service agents, answers usually come from knowledge articles. Retire outdated articles, merge duplicates, and make sure each article answers one question clearly.
4. The right permissions
An agent should see only the data its job requires. Review the agent user's permission sets and sharing, and be strict with sensitive fields like health or financial data.
5. Guardrails and handoff
Decide what the agent must never do, and when it should hand off to a person. Write these as explicit instructions, and give customers an easy way to reach a human.
6. Real test cases
Before launch, run the agent against a batch of real historical cases or conversations and review the results. This catches wrong answers far better than a few demo questions.