Choose the outcome first
Create Actions for repeated next steps: lead capture, appointment booking, ticket handoff, Shopify assistance, forms, buttons, suggested replies, web search, or selected OpenAPI operations. Each Action should have a clear purpose and approval boundary.
Collect clean fields
Ask for the minimum information needed to complete the Action. Use forms, guided questions, and buttons when structured fields matter, especially for contact details, order identifiers, booking preferences, and ticket summaries.
Explain what happened
After an Action runs, the agent should summarize the result in plain language and make the next step obvious. If the tool fails or the request is risky, the agent should escalate instead of pretending the workflow succeeded.