What AI support should not do

Most writing about AI support is about capability. The more useful question for a small team is the boundary: what should the assistant never do without a human?

Four things to keep human

  1. Deciding that a ticket is closed. An assistant can propose it; a person should confirm, or the desk teaches customers that raising a problem gets them ignored.
  2. Promising dates, discounts or exceptions. Anything that binds the company should come from someone who can be held to it.
  3. Handling an angry customer. Tone is the whole message in a complaint; a generated reply reads as indifference.
  4. Anything involving personal data it was not given for this purpose. The safe default is to escalate.

What AI should do

Draft answers to the questions that repeat, summarise a long thread before handover, and tag incoming conversations so triage is faster. All three make people faster without changing what the company promises.

That is the design constraint behind Onsites AI: replies live inside the conversation, any agent can take over mid-thread, and the assistant is billed per action rather than per seat.

AI inside the inbox · cost per month · the complete guide

Free for the first 3 seats. onsitesai.com

Comments

Popular posts from this blog

The small-team support desk: a complete guide

Handling refunds without a second system

A support policy that fits on one page