Nobody asks us "can AI answer our tickets?" any more. They arrive with a number.
In recent conversations with support leaders, we heard four different answers to the same question: how much of your ticket volume should AI automate?
- One said a maximum of 30%.
- One is at 40% today and sees room for 20 percentage points more.
- One is aiming for somewhere between 50% and 90%.
- One would only replace their helpdesk if AI handled around 80%.
Same question, wildly different answers. And when you listen to the reasons, almost none of them are about what the AI can do.
The number is low when the data is messy
A head of customer service at a travel deals portal put her ceiling at roughly 30% of total volume. Her support is outsourced and billed by the minute, so she has every reason to want that number higher. Her explanation for why it isn't:
"The topics are the same, but the answers aren't."
For one very simple customer question, her team has 35 different answer variants, and no central database that says which one applies. An AI agent can only automate what can be answered consistently. Until the 35 variants become one rule, the automation rate stops well below what the technology could handle.
That is the pattern behind most low numbers we hear. The ceiling looks like an AI limitation. It is a data problem wearing an AI costume.
Sometimes the number is low on purpose
The customer service lead at an online travel agency already automates 40% of tickets. He sees about 20 percentage points more as realistic in the medium term. The remaining 40%, complex after-sales cases and sales conversations, he would rather not automate at all.
Not because AI couldn't. Because he wants to keep a human in the sales conversation while, in his words, everyone else jumps on AI. He sees that as something that can set them apart in the market.
This is the part of the automation debate that rarely gets written down. Some of the ceiling is not a constraint. It is a decision about where a person is the product.
The number means nothing without the cost line
The same customer service lead gave us the figure the whole conversation actually turns on: a simple ticket costs him about €5 all-in today. His target is three to four euros.
That is why buyers show up with a percentage. The automation rate is not a goal in itself. It is what has to be true to bring a €5 ticket down to his target. Any conversation about how much AI can automate is really a conversation about how much it needs to, and at what cost per ticket the answer stops mattering.
What this means if you're deciding on your own number
The buyers we talk to are not arguing about the floor. Whether AI can handle simple tickets is settled. They are arguing about the ceiling, and the ceiling is set by three things they control: how consistent their answers are, which conversations they deliberately keep human, and what a ticket costs them today.
To see where you sit, we've laid out the four levels of customer service automation and how to work out your real cost per ticket.
Quotes are translated from German and anonymised. Companies are described by industry and role only.
