The percentage is a slice
Published claims that a support agent “handles 80 percent” of conversations are easy to misread. The honest version of that sentence names the slice: billing questions that match a dozen articles, password resets, or order-status mail. It does not mean 80 percent of everything a customer might say.
Orvia House is not publishing a support case study. The figure in the address of this page is the claim people repeat. It is not a result from our work.
Define the slice before the pilot
List the intents in scope. List the intents that are always out of scope: refunds above a threshold, legal threats, anything that needs an account change you would not let a new hire make alone. A ticket outside the list is a person-ticket, and it does not count against the agent.
If you cannot write the list, you cannot interpret a percentage later. Collect the list from real tickets, not from a workshop guess, and say how many weeks of mail you read.
Design the handoff
A support workflow we will actually propose has four moves. Classify the ticket against the list. Retrieve the approved answer or the order state. Draft a reply. Stop. A person sends it.
Containment is not resolution. A chat that closes and returns the next day is a failure of the draft, even if the first session never reached an agent on your team. Count reopen rates inside the slice.
- Share of in-slice tickets approved with no edit
- Share edited before send
- Share rejected and rewritten
- Share that reopen within a window you set
- Median time a person spends on the approval
Do not average the hard tickets away
Report the slice and the rest of the inbox side by side. A strong number on password resets next to an untouched pile of billing disputes is a useful finding. Averaging them into one percentage hides the project you still have to do.
Related service: Agentic AI and automation