Auto-Resolving 65% of Support Tickets with an AI Agent
A SaaS support team was spending its time on repetitive tier-1 tickets, which pushed response times out and put SLAs at risk. I led the build of a custom AI chatbot with knowledge base search, intent detection, and human escalation workflows. It now auto-resolves 65% of tickets, response time went from two hours to thirty seconds, and customer satisfaction rose from 3.2 to 4.6 out of 5.
Results
The problem
Repetitive tier-1 tickets were consuming agent time, hurting response SLAs.
The approach
I led the build of a custom AI chatbot with knowledge base search, intent detection, and human escalation workflows.
-
Grounded answers in the knowledge base
The agent answers from the client's own documented material rather than from model memory, which is what keeps support answers correct and current.
-
Detected intent before answering
Intent detection decides what kind of request this is, so the system knows when it is out of its depth instead of confidently guessing.
-
Designed escalation as a first-class path
Anything the agent should not handle goes to a human with the conversation attached. Escalation is a feature, not a failure state.
-
Tracked satisfaction, not deflection alone
Auto-resolution only counts if customers are happy with the resolution, so CSAT was reported alongside the deflection rate.
The outcome
65% of tickets auto-resolved. Response time from 2 hours to 30 seconds. Customer satisfaction from 3.2 to 4.6 out of 5.
Stack
Questions
What stops the chatbot from inventing answers?
It answers from the client's knowledge base rather than model memory, and intent detection routes anything outside that scope to a human instead of guessing.
How does escalation to a human work?
When the agent detects a request it should not resolve, the conversation is handed to a support agent with the full context attached, so the customer does not repeat themselves.
Does deflection hurt customer satisfaction?
It did not here — satisfaction rose from 3.2 to 4.6 out of 5, because straightforward questions were answered in seconds and complex ones still reached a person.
Building something similar?
If you are shipping AI features or scaling a remote engineering team, I work with founders across the UK, EU, US, and MENA.
Get in touch →