AI Agent for customer support
I built an app to handle L1 - L2 support with custom AI agents and it's working quite well.
Here's what the agent does now:
Logs in to our livechat (crisp.im) and responds to incoming chats as an operator with <10s response time. It backs off if a human is already replying to that chat.
Handles hosting issues by asking followup questions to customers, restarting services, clearing cache, resetting config, reviewing logs, WordPress password resets and more.
If the issue is not resolved after different troubleshooting attempts, agent sends us a message on slack. When a team member replies to that message, agent proceeds based on that.
For sensitive operations like billing, site migrations, it escalates the chat to team and sends a message in slack todolist channel.
This app has a zapier like workflow builder to build the tools used by AI agents. (Think api calls, stripe, emails, slack, gmail, mysql search)
Would anyone here be interested in seeing the backend of this?
Replies
IXORD
I have seen such systems and in practice I always immediately wanted to call a person to solve my problem. The fact that AI can cover mostly ordinary questions, for example, about a product. But I have never seen an improved version of AI made as an Agent for a product. The main task is to properly teach it all the topics that a client can ask.
@ixord I also looked at existing chatbots and all they could do was to rephrase training material, answer frequently asked questions etc. That's why i built this custom agent to follow same troubleshooting steps as my support staff does.
For example, when a customer says that their website is showing error 504, support employee looks up their domain, goes to their server, restarts services etc. Same thing now happens with the AI workflow, in 10 seconds compared to 10 minutes it takes employee.