Home › How-to guides › Automate customer support replies
How to automate customer support replies with an AI agent
I looked through 575 real cases about customer support replies and pulled out what people actually did, the prompts they used and the places it went sideways. No theory, no invented success stories.
What people actually do
- Draft replies and promo codes for L'Oréal support advisors
- deinePflege makes in-home healthcare more accessible with Agentforce.
- bunq builds intelligent banking assistant Finn with Claude
- Zota’s investment in digital labor pays off with 30% YoY growth.
- Zapia powers local commerce for millions across Latin America with Claude and Google Cloud
- Young Drivers gears up to expand globally with Agentforce.
- Workato connects enterprise AI agents to business systems with Claude
- Wiley sees 213% return on investment with Salesforce.
Browse all 2754 cases to see the rest.
Prompts to start from
These are the reusable prompts from real cases. Swap in your own details, run it once on something low-stakes, and read what the agent did before you trust it.
For each new support case, summarise the conversation so far, confirm what the customer actually wants, and draft a reply in our brand voice for me to review. Before I close the case, check that the case category matches the conversation.
I want an agent to handle routine customer questions end to end, and pass anything unclear to a person. The tool in the source case was Agentforce.
Tools that show up most
Agentforce (84), Claude (39), Amazon Bedrock (34), Claude Code (31), ChatGPT (29), Cognigy (28).
Guardrails
Start small: run it once on a low-stakes example, read what the agent did, and only then let it run on real work.
More mess: the AI agent fails page. Related: the prompt library.