How Many Support Messages Can AI Phone Control Answer?

How a day’s messages pile up

I know a woman who does support for a cross-border shop. She looks after a dozen or so store accounts. She says the most machine-like stretch of the day is the first five minutes after she opens the computer: signing into each account in turn to see whether anything arrived overnight.

Last Wednesday I watched her work. The busy part starts at half past nine. One message asks “are you there,” one asks “which courier,” one asks “can it be cheaper.” She switches between accounts, finds the matching reply, confirms it went out, and switches back.

She counted once around noon. She had sent about 120 replies that morning. Fewer than 20 of them needed her to think. For the other hundred, she already knew the answer, and each one still had to be sent by hand.

Split them into three groups first

Pour a day’s messages out and there are three kinds.

The first kind has a completely fixed answer. Are you there, what is the price, how much is shipping, how soon does it ship, can it be returned, is it in stock. The standard answer was written long ago. Each time it is typed again.

The second kind is half standard. For example, “which of these two is better,” or “if I buy two, can it be cheaper.” The first half can use a prepared line. The second half needs a look at the specific case.

The third kind has to be a person. Complaints, returns and exchanges, anything that involves a promise, and any conversation where the tone has already gone wrong.

The first kind is the largest volume. The third kind is the hardest. The second sits between them. Most of the time goes to the first kind, because that is where the volume is.

Group one: store the answers

Write the most common questions as one line each:

Yes, in stock
Free shipping, dispatched within 48 hours
7-day no-reason returns accepted
Price is XX, or XX for two together

When you reply, you stop composing the wording on the spot and pick a line to send. The wording is done once. After that, each reply calls it.

One step further, let a machine take these. iEasyRun is free and runs on your own computer. It can be described like this:

Open the support pages for these stores. Find messages asking “are you there,” “what is the price,” “shipping,” or “how soon does it ship,” and reply with the matching line. Read the price and the shipping fee from this sheet.

Three things have to be written down: which messages to look at, which lines to use, and where the data comes from. How to tap stays out of the description.

This kind can be handed over because the repetition is high, the answer is fixed, and the cost of a wrong answer is small. All three apply. A machine is steadier than a person on this kind of question. A person answering it thirty times gets tired, then shortens the reply, then gets it wrong. A machine keeps sending the full line.

Groups two and three

On the half-standard ones, hand over the lookup and keep the send. Have the machine read out the relevant price, stock, and specification and put them in front of you. A person glances and then sends. What you save is the time spent looking things up. In this kind of conversation, looking things up takes longer than typing the reply.

The ones that have to be a person stay on that side of the line. The customer wants to be taken seriously. A correct fact still fails to hold this kind of conversation. Some teams, chasing full automation, answer complaints with a template too. The customer blows up on the spot, then files a complaint with the platform. The few minutes saved are paid for with the account.

There is another boundary. If sending the wrong sentence causes a real loss, leave it with a person. Getting “are you there” wrong is awkward. Getting “can I return this” wrong is a dispute.

How to arrange the day

Fifteen minutes before the day starts, read the overnight messages. Keep this step yourself. You are also seeing what today looks like. A new kind of question in the overnight messages often means the lines have to change that same day.

Use the morning to clear the first kind. Let the machine take the standard questions. People handle the part of the second kind that needs a judgement. People are clearer in the morning, so judgement conversations fit that stretch.

From afternoon into evening, watch the third kind. An overseas customer’s daytime often lands in your afternoon and evening, and complaints and after-sales bunch up then. A person needs to be there for that stretch.

At the end of the day, read the day’s record. Which kind of question suddenly increased, and which line got used over and over: that decides whether tomorrow’s lines need a change.

Put people in the hours that need a judgement. Put the machine in the hours with the most volume and the simplest questions.

Where the accounts sit

Several accounts raise another question: which device each account lives on.

Signing in and out of many accounts on one device, in a short time, is itself unusual behaviour. Spread the accounts across several devices. Each device looks after its own few. When you need to handle them together, work by device group, and leave each device signed in to its own accounts.

With few accounts you will not feel it. Once there are many, this becomes the bottleneck. While nothing has gone wrong, the pace can still feel fine. When something does go wrong, the cause turns out to be the login switching.

How to decide whether to hand a category over

When a category of message comes up, ask three questions.

Is the answer fixed? If it has to be thought through every time, handing it over only moves the thinking somewhere else.

Does a wrong answer cause a real loss? If it does, leave it with a person. Keep that rule.

Does the customer want accuracy, or do they want tone? Accuracy can be handed over. Tone stays with a person.

If all three pass, it can be handed over. If one fails, leave it for now.

Hand over the highest-frequency and simplest ones first, such as “are you there” and “what is the price.” They are the largest volume, and a wrong answer costs the least. Once they run steadily, add harder kinds. Start with the hardest conversations and fail twice, and people put the whole thing down.

For device-side checks, see what to do when a phone keeps going offline. For how to arrange many devices, see one person managing 20 phones.