The automation you should not build
Automate a message when the answer is the same every time, cheap to be wrong about, faster than a person, and does not need judgement. If any of the four is false, a person should send it.
The four tests
| Test | Passes | Fails |
|---|---|---|
| Same every time | Opening hours | Whether we can deliver by Friday |
| Cheap to be wrong | Directions | A quoted price |
| Faster than a person | Instant, at 2am | A reply a person would send in a minute anyway |
| No judgement | Order shipped | Whether to refund |
The category that always fails
Anything responding to bad news. A complaint, a delay, a damaged item, a bereavement mentioned in passing. Automation cannot tell it is happening, and the reply it sends will be cheerful.
This is the failure that gets screenshotted and shared. The engineering fix is to exclude conversations that are already open from every automated send - a rule worth having even if you never build anything cleverer.
The one that is tempting and wrong
Automating a reply a person would have sent in under a minute. It saves almost nothing, and it costs the small human contact that makes people forgive the next mistake. "Yes, we have it" typed by somebody is worth more than the same words generated - not because the words differ, but because the next message can build on it.
What to automate instead
- The facts. Hours, address, delivery cost, tracking.
- The status. Order events, appointment reminders - the three worth sending.
- The expectation. When somebody will reply.
- The routing. Getting a conversation to the right person is invisible automation and the best kind.
Say when it is automated
Not with a disclaimer on every message. With a tone that does not impersonate: "This is an automatic reply - we open at 8 and will answer then" is honest, brief, and nobody minds. Discovering afterwards that the warm, personal reply was a template is what people mind.
Questions people actually ask
What should never be automated in customer messaging?
Anything responding to bad news, anything involving money the customer disputes, and anything where being wrong is expensive for them. Those need a person who can be held to the answer.
Is it dishonest to automate a reply?
Not if it is accurate and does not pretend to be a person. The dishonesty is in the pretending, not in the automation.
Njiwa
One inbox for WhatsApp, Facebook and Instagram, and an API that sends over either WhatsApp rail. 7 days free, no card.