Using AI to answer customer texts, and when a human steps in

What an AI assistant can actually handle over text

Most of the replies a small business gets by SMS are the same handful of questions asked in different words. What time do you open. Can I move my appointment. Have you got my invoice. Where are you. These are not hard questions, but they arrive at all hours and each one interrupts whatever you were doing.

An AI assistant sitting on your SMS number can take most of these on directly. Because it is connected to your systems rather than making things up, it can do real work, not just chat. In practice that means it can:

• Answer questions about your hours, location, services and pricing, drawn from information you have given it
• Book, confirm or cancel an appointment, and offer alternative times when someone wants to move
• Look up an invoice and tell a customer what they owe and how to pay it
• Record a promise to pay, so when a customer says "I will pay Friday" that commitment is captured rather than lost in a thread

The key word is grounded. The assistant answers from your own knowledge base, not from general internet guesses, so it says what is true about your business and nothing more. If you have not told it your Saturday hours, it does not invent them. Getting that knowledge base right is its own job, and we cover it in building a knowledge base your AI assistant can use.

A real conversation, start to finish

Here is the kind of exchange an assistant handles without anyone lifting a finger:

Customer: hi do you have anything for a haircut this week
Assistant: Hi Marco, yes. We have Thursday 4pm or Friday 10:30am with Sam this week. Which would you like?
Customer: friday morning
Assistant: Booked you in for Friday 18 Sep at 10:30am with Sam. We will text a reminder the day before. Anything else?
Customer: nope thanks

No one at the salon touched the phone. The booking landed in the calendar, the reminder is already scheduled, and Marco got an answer in seconds at whatever hour he happened to text. That is the ordinary case, and it is most cases.

The line where a human steps in

The point of automation is not to remove people. It is to save them for the moments that need a person. A good setup is clear about where that line sits, and hands over cleanly.

A human should take over when:

• The customer is upset or complaining. An unhappy customer needs a person who can apologise properly and make a decision, not a smooth automated reply that makes them angrier.
• The request is unusual or high stakes. A quote for a big job, a dispute over a bill, anything involving money or a promise the business has to keep, belongs with a human.
• The assistant is not sure. A well-built assistant knows the edge of its own knowledge. When it cannot answer confidently from what it has been given, it should say a team member will follow up rather than guess.
• The customer asks for a person. If someone says "can I talk to a human", that request is honoured immediately. Full stop.

The handover matters as much as the trigger. When the assistant steps back, the whole conversation should land in front of your team with its history intact, so the person picking it up can see what has already been said. That is where a shared team SMS inbox earns its place: the assistant handles the routine, and the exceptions arrive in one place with full context, ready for whoever is free to answer.

Why texting suits AI better than a chatbot on your website

Website chatbots have a reputation problem, and mostly they have earned it. They pop up uninvited, they trap you in menus, and they vanish when you close the tab. SMS is different in ways that happen to suit an assistant well.

A text conversation is patient. The customer can reply now or in three hours, and the thread is still there. It is personal, arriving on the same device where they message friends and family. And because it runs on a dedicated Australian number that both sends and receives, the customer can save you as a contact and come back to the same thread next time. This only works as a two-way channel, which is the whole idea behind letting customers reply in the first place.

Being straight about the limits

An assistant is a genuine help, not magic, and it is worth being honest about the edges. It answers from what you give it, so a thin or out-of-date knowledge base produces thin or out-of-date answers. It should never make firm promises it cannot keep, which is exactly why the handover rules above exist.

It also cannot change the physics of messaging. When a reply leaves the platform it is sent, meaning handed to the carrier, and it only shows as delivered when a real delivery receipt returns. No assistant, and no platform, can guarantee a message reaches a switched-off phone. What it can do is make sure that when a customer does text, they get a fast, accurate, useful answer, and that the tricky ones reach a human quickly.

Set up well, the split is simple: the assistant carries the volume, your team carries the judgement, and the customer gets a quick reply either way. To see how this fits a trade, clinic or shopfront, have a look at our use cases.