How do you balance the human side of customer support with the AI side? That’s the last question at an AI lunch-and-learn for a tactical gear maker with a few dozen employees, and it’s the one underneath every conversation about AI in customer service.
The person who asks it tells me why it matters to them.
“We get a lot of feedback saying, I’m so glad I talked to somebody. And they’ll leave that as a five star.”
For this company, a real person on the phone is not a cost to be cut. It’s the reason people leave five stars. “That’s my selling point,” they say. And everyone in the room knows what AI in customer service usually looks like: a chat bubble whose whole job is to keep you away from a person.
The worry is well founded. In a 2024 Gartner survey of 5,728 customers, 64% said they would prefer companies didn’t use AI for customer service, and 53% said they would consider switching to a competitor if they found out a company was going to. Their top fear was that it would get harder to reach a person.
Earlier in the session, I had answered the team’s first question, which was this one in a different form: how can customer service use AI while keeping the human connection?
I gave them the mechanics. I had connected my AI to my email. Before I replied to someone, I had it search every conversation I’d ever had with that person, then draft a reply with all of it in mind. If a customer once got a refund they weren’t happy with, the draft knew that, so it didn’t cheerfully pitch them something new. And a person read every draft before it went out. “That’s where the human connection is,” I told them.
All of that was true. It was also a list of features.
The person with the five-star reviews says it better than I did.
“That call used to be: okay, who are you, where are you from, what do you do?”
With the history already pulled, none of that is needed.
The phone rings.
The rep already knows who is calling. What they bought. When it shipped. What went wrong last time.
And the call opens with, “How can I help you?”
That was the best answer anyone gave all day, including me. I had been describing AI as a way to answer better and faster. They described it as a way to start the conversation in the right place. The AI never talks to the customer. It takes the interrogation off the front of the call, and the person who picks up gets to spend the whole call being a person.
Put AI on the prep, never on the conversation
The newest data lands in the same place. In a 2026 Gartner survey of 3,566 business and consumer customers, half said their interactions are easier when companies use generative AI. And 87% said it’s essential to offer a way to reach a human. Customers are not against AI. They’re against AI standing between them and a person. As Gartner analyst Eric Keller put it, “Service leaders should not use GenAI as a mandatory first step for every issue.”
So put it behind the rep instead of in front of the customer. Four places it earns its keep:
Pull the history before the reply. What they bought, when it shipped, whether it shipped late, what they wrote last time. By hand, that means digging through several systems, which is exactly the step that gets skipped on a busy day. AI does it every time.
Draft, never send. AI writes the first draft. A person reads it, fixes it, and sends it. The win isn’t speed on the easy tickets. It’s that your best people spend their attention on the hard ones instead of typing out a tracking number for the hundredth time.
Reuse what you’ve already answered. Most support answers disappear into a sent folder. The next customer asks something similar, and someone researches it from scratch. Ask the AI how you answered questions like this before, and have it show you the original.
Run a wording pass. Picture a tired rep at 4:45 on a Friday, writing a short reply that reads as curt. AI flags it before it goes out. That’s a save, not a replacement.
Your AI can only use what it can see
All of this depends on one thing. An AI model only works with what’s in front of it, its context window. If it can’t see the customer’s history, it can’t prep the call.
The way it sees that history is usually an MCP connection. MCP stands for Model Context Protocol, and the plain version is that it’s a bridge between the AI’s window and wherever your data lives: Gmail, Jira, Trello. Connect your email, and “search every conversation with this customer” becomes a question you can ask instead of an afternoon you have to spend.
So before you put AI anywhere near customer service, ask two questions. Can it see the history? And does a person still own the conversation?
If the answer to both is yes, every minute the AI takes off the mechanical half goes back into the human half.
You don’t get fewer conversations. You get better ones.
The kind that start with “How can I help you?”
