Auto-replies that help, and the four that make people angry

An auto-reply earns its place only when it answers the question or sets an accurate expectation. Everything else - acknowledgements, menus, chatbots that cannot hand over, and cheerful messages sent to somebody complaining - makes the experience worse than silence.

The two that work

The answer

Where are you, are you open, do you deliver, how much is delivery. These have one correct answer and it does not change. Answering them in a second is better service than a person answering in twenty minutes, and it frees the person for the questions that need them.

The accurate expectation

"Thanks - we are closed now and will reply at 8am." The value is entirely in the accuracy. If you reply at 11am, you have taught them your messages are not true, and the next one gets less attention than it deserves.

The four that make people angry

1. The empty acknowledgement. "Thank you for your message, we will get back to you." It contains no information. Its only effect is to make the wait feel deliberate.
2. The menu with no exit. Reply 1 for sales, 2 for support - and no option that reaches a person. Somebody with a problem that is not on the list is now trapped in your phone system, in a chat app.
3. The bot that will not hand over. Three rounds of "I did not understand that" is worse than nothing, because it consumed the customer's patience and produced no answer.
4. The cheerful reply to an angry message. An auto-reply has no idea it is answering a complaint. If you automate anything, exclude conversations that are already open, and consider excluding messages with the obvious anger words - or accept that you will occasionally send a smiley face to somebody whose order never arrived.

Rules that keep them useful

• Once per conversation, not per message. Somebody sending three messages in a row should not get three identical replies.
• Never inside an open conversation. If a person is already replying, the automation stays out.
• Say it is automatic. Not in those words necessarily, but do not let somebody believe they are talking to a person and then discover otherwise.
• Match the language they wrote in where you can, and pick the most common one where you cannot.
• Review what triggers them monthly. Keyword rules rot: a product name changes and the rule now fires on something unrelated.

A worked example

Bad:

Thank you for contacting us! An agent will be with you shortly.

Better:

Thanks for your message. We are open 8am-6pm Mon-Sat and
usually reply within an hour.

Right now: we are closed, and we will answer at 8am.

Deliveries: tracking is at example.com/track
Directions: 4th floor, River House, off Ngong Road

The second one has a chance of ending the conversation to the customer's satisfaction before anybody arrives at work, which is the only thing an auto-reply can do that a person cannot.