Every messaging play on this site assumes a customer might reply, because that is the point of the channel. Almost nobody decides in advance who reads those replies, and that omission is the most common reason a promising messaging programme quietly stops.
Answer four questions first
- Who sees a reply? A named person, not 'the team'.
- Within how long, during business hours? Pick a number you can actually hit.
- What happens outside those hours? Silence is a valid answer if you say so.
- Who covers when that person is off? This is where most systems fail.
Set the expectation in the message
If you cannot cover evenings, say so. A customer told that replies are answered from 9 to 5 is not disappointed at 9pm. A customer left guessing is.
Ridge Dental
iMessage
Hi Dana — Ridge Dental. You're booked tomorrow at 2pm. Reply C to confirm. We check messages 8am–6pm weekdays; for anything urgent please call {number}.
C
The one exception
If you have advertised a coverage window, honour it. A business that says 8am–6pm and answers at 10pm has trained customers to expect 10pm. Under-promise here — it is the only promise in messaging you make explicitly.
Where the replies should land
| Setup | Works when |
|---|---|
| A shared inbox in the provider | Nobody writes code. Miss Blue ships one. |
| Webhook into your CRM | You have a CRM people already live in |
| Webhook into Slack or Teams | Small team, fast culture, low volume |
| Forwarded to one person's phone | Very small volume, and a real single point of failure |
The last row is how most businesses start and it works until that person takes a holiday. Build the second option before you need it, not during the week you discover you needed it.
How much work is this actually?
Less than people fear. A clinic sending 400 reminders a month might see 100 replies, and 80 of them are the single letter C. The remaining 20 are real questions taking a couple of minutes each — well under an hour a month of genuine work.
The load is not volume. It is the interruption pattern: replies arrive unpredictably and each one feels urgent. Batching them — twice a morning, twice an afternoon — is almost always better than reacting to each notification, and customers cannot tell the difference.
Where an agent helps and where it does not
An AI agent is genuinely good at the 80 replies that say C. It should not touch the ones that do not. The two-message rule is the practical boundary: an agent gets the opener and one structured response, then a human owns the thread.
Before your first send
- A named person owns replies, with a named backup
- A response-time target you can actually hit
- Coverage hours stated in the message itself
- An escalation path for anger — see when a customer replies angry
- Replies land somewhere that survives one person being away