# Does AI make customer service less human?

> Does AI take the human touch out of customer service? What my years in customer service taught me, and four questions to check your own AI.

Author: Fabien Monnier, Gem CX. Published 2026-11-06.
URL: https://gem-cx.com/en/blog/does-ai-make-customer-service-less-human

## Key points

- A live chat handled without AI mixes investigation and writing: order, email history and stock are checked across several tabs while the customer waits.
- An AI agent can gather the request, check the order and stock in seconds, then hand over to a person with a summary and a link to the order.
- AI makes service less human when it closes the door: a loop with no way out, scripted empathy, a handover where the customer has to start over.
- Read by an AI, a passing hint about a phone call in a long email can trigger an immediate callback from the team, without waiting in the email queue.
- Four questions tell you which way your AI leans: does the customer repeat themselves, can they reach someone, does the agent get the context, what happens when the customer gets annoyed?

Not when it's in the right place. For a customer, a human exchange means being understood quickly, not having to repeat yourself and feeling that someone is actually dealing with the problem. In my experience, without AI, a support agent spends much of every conversation looking up the order, the history and the stock while typing. An AI that does that investigation in seconds gives the person time to reassure and act.

![Two rows. Without AI: the support agent looks up the order, emails and stock, writes and asks the customer to wait, then reassures and acts. With AI: the AI gathers the need and checks, hands over a summary and a link to the order, the support agent reassures and acts](https://gem-cx.com/blog/ai-human-touch/time-en.svg)

_Illustration: where a support agent's time goes, with and without AI._

## Why does AI feel like it makes customer service less human?

Because many deployments use it to close the door. Customers notice straight away, and so do support teams.

You've seen the cases:

- the bot that goes in circles and never offers a person,
- the phone number that vanishes from the website the day the chatbot goes live,
- scripted empathy, "I understand your frustration" on the third identical reply,
- the cold handover: the customer finally gets an agent, who asks for the order number again.

These complaints are fair, and customers say so themselves. In [a Gartner survey](https://www.gartner.com/en/newsroom/press-releases/2024-07-09-gartner-survey-finds-64-percent-of-customers-would-prefer-that-companies-didnt-use-ai-for-customer-service) of 5,728 customers run in December 2023, 64% said they'd prefer companies didn't use AI in customer service, and their top concern was that it would get harder to reach a person.

As I see it, that number mostly judges how AI gets set up. In the deployments customers complain about, I almost always find three gaps:

- **architecture**: a bot placed in front of customer service with no access to orders, stock or history, so it can only answer in general terms,
- **automation**: nothing is planned for what comes next, and the handover to a person carries no summary and no link to the order,
- **team organisation**: nobody knows who picks up the conversation, when, or with what priority.

With those three gaps, AI can only cut the number of contacts with the team, even if it leaves the customer alone with the problem. On the team side, the worry follows: if AI is there to avoid conversations, agents wonder what's left for them.

I argue for a narrower use of AI. It does the investigation and sets things up, and a person takes over as soon as the situation calls for it. To see why that makes service more human, look at what an agent actually does during a live chat.

## Fifty chats a day

About ten years ago, in my first customer service job, I handled around fifty chats a day, often two or three at once. A customer came in, I asked for the order number and tried to understand what they really needed. Meanwhile I was checking other tabs: their email history, the order in the back office, whether the product was in stock on the website. And I was writing, looking for the clearest, most reassuring way to put things.

Looking back, I did fine. Customers got a reply in under 30 seconds on average, which is reasonable. A conversation lasted 5 to 25 minutes, for the customer and for me. When you do fifty a day, the reflexes come on their own.

A few years later I was running a customer service team. Now and then I'd pick up a chat myself. I hadn't handled conversations for a long time, and the reflexes were gone. I felt how heavy the whole thing was: work out the need, keep the customer waiting, reassure them at the same time, open the right tabs, find the right words. The person who was supposed to have the most experience on the team was struggling to give good service.

Honestly, picking up those chats is when I realised live chat has its limits. It's a good channel, but it carries a tension customers feel: the silences while you search, the "let me check and get back to you", the replies typed too fast. In those moments the person has all the goodwill in the world and no time. Investigation and writing crowd out what I call the human part: listening, reassuring, deciding.

## How can AI make live chat more human?

By doing the investigation before the agent arrives. An [AI agent](https://gem-cx.com/en/ai-agent) connected to your tools gathers the customer's request smoothly, reads the order, the email and conversation history, checks product availability, and replies in seconds. It also picks up frustration or worry in the messages and adjusts: reassurance first, explanation second.

When the situation needs a person, it offers the handover. The agent then gets the conversation with a summary of the context and a direct link to the order and the product. No investigation to redo, no need to keep the customer waiting. They can focus on the most human part: reassuring and setting the fix in motion, for example swapping an out-of-stock product for an alternative.

An example, rewritten:

**Customer**: Hi, I ordered a jacket ten days ago and still haven't received anything. I'm travelling on Saturday.
**AI agent**: Hi, I'm looking into it now. Could you give me your order number?
**Customer**: 48213
**AI agent**: Thanks. Your order is on hold: the jacket in size M is out of stock and the restock date isn't confirmed. With your trip on Saturday, I understand that's worrying.
Two options are in stock: the same jacket in navy in size M, or the original colour in size L. I'm passing you to a colleague who can make the swap and speed up delivery.
_Handed over to the team with a summary: order 48213, jacket size M out of stock, travelling Saturday, two alternatives in stock_
**Support advisor**: Hi, it's Julie. I've got your order in front of me. I can swap it for the navy in M if that works for you, and upgrade it to express delivery. You'll have it Thursday.

For the customer, the request took two messages, with no waiting. Then a person arrives who already knows the case and acts straight away. Julie spent her time on the one thing the AI couldn't do for her: deciding on the swap and the express delivery.

## The human touch keeps changing channels

Today, I think an email answered by a person gets put on a pedestal, as proof of human service. It wasn't always that way. For a long time most of it went through the phone, and as I see it email first felt less human: no voice, a delayed exchange, a reply after several hours, sometimes more than a day, with no idea who was behind it. In the end the customer gets text on a screen, written by a person, after a wait.

So the useful question, as I see it, is how AI can make the exchange more human, whatever the channel.

At one of the brands we work with, AI reads every new email as soon as it arrives. When a customer hints that they'd rather be called, an alert goes out on the team's internal chat: this person wants a call, call them now, top priority. The first person free takes the alert, flags it to the others and calls.

The detail that matters: customers rarely spell it out. Picture a fifteen-line email about a delivery that went wrong, with the expected date, the carrier, a damaged box, a photo. And in the second-to-last sentence: "I'm home all tomorrow morning if it's easier to talk." A rule on the word "call" would miss it, while an AI that reads the whole email, the way an attentive agent would, gets the request.

The result: instead of waiting hours in the email queue, the customer gets a call back from someone on the team.

## What AI doesn't do for the team

Decisions stay with the team. A goodwill gesture, a case the rules don't cover, a voice on the phone for a worried customer: all of that stays with the agents.

For the team, the job changes. Fewer tabs to open, less copy and paste, and more conversations where you listen, reassure and decide. I think that's the part of the job most agents prefer, and the part the investigation used to eat.

Two conditions still matter. First, a handover to a person that's easy to get, at any point in the conversation. Second, an AI that reads accurate information. An AI connected to an outdated knowledge base repeats rules the team no longer applies, and the customer ends up repeating themselves to an agent. We've written about [building that knowledge base from what you already have](https://gem-cx.com/en/blog/knowledge-base-before-ai-agent), without rewriting everything.

## How do you know if your AI makes service more or less human?

Look at the conversations where AI hands the customer over to the team. Four questions are enough:

1. After the handover, does the customer have to give their order number again or re-explain the problem?
2. Can the customer reach a person easily when they ask, or does the AI keep them going in circles?
3. Does the agent get a summary and a direct link to the order, or do they have to redo the investigation?
4. When the customer gets annoyed, does the AI adjust its reply and offer a person, or repeat the same answer with an empathy line?

To start, take twenty or so conversations from last week where AI passed the customer to the team. Write down every piece of information the agent had to ask for again. That list is the first thing your AI should check before the handover.

## FAQ

### Will AI replace customer service agents?

That's not what I see in practice. AI takes on the investigation and the simple questions, and agents keep the situations that need a decision, a gesture or a voice. Their job changes: they search less and spend more time listening and acting.

### Do customers prefer talking to a human?

Not necessarily, but they want to be able to reach one. In Gartner's December 2023 survey, customers' top concern about AI was that it would get harder to reach a person. As I see it, they mostly want a correct, fast answer, and a person when their situation calls for one. An AI that replies in seconds about their actual order, then hands over with the context, gives them both. What they don't put up with is a closed door: a bot that goes in circles without ever offering someone.

### Can AI show empathy to a customer?

It can pick up frustration or worry in the messages and adjust its reply: reassure first, acknowledge the customer's specific constraint (a trip, a gift, a date), offer a person sooner. A stock phrase repeated on every message just annoys the customer more.

### Which channel is the most human: live chat, email or phone?

No channel is human in itself. The phone was the benchmark for a long time. As I see it, email first felt cold before becoming a sign of a personal reply. An exchange feels human when it's fast, the customer doesn't have to repeat themselves, and a person acts at the right moment.