AI for customer service

The jobs inside this work

The same twenty questions

Hours, availability, delivery, returns, what happens next.

Order and job status

Where is it, when is it coming, who is turning up.

First response out of hours

Acknowledging a message at 9pm so the customer is not left hanging.

Finding the answer

The information exists — in an old email, a PDF, or someone's head.

Handling complaints

The part that decides whether you keep the customer.

Recording what happened

So the next person is not starting from nothing.

What AI can realistically help with

• Answer routine questions from a set of approved answers you wrote.
• Acknowledge a message immediately and set an honest expectation.
• Route unusual questions to the right person with a summary attached.
• Draft a reply for you to adjust rather than write from scratch.
• Spot the questions being asked often enough to deserve a proper answer.

What should stay with a person

• Complaints, refunds, and anything that has already gone wrong.
• Anything about money owed or a legal obligation.
• Anything the approved answers do not clearly cover.
• Situations where a customer is upset. A person is the whole point there.

Risk and review

• A wrong answer given confidently to a customer is worse than a slow answer.
• Support conversations contain personal data — check what the provider retains.
• Never let a service bot promise a refund, a date, or a discount on its own.