Customers do not judge an AI mistake the way they judge a human one. The same error costs more trust when it comes from AI. The reason is not the error itself. It is that customers read an AI mistake as a decision the company made, and a human mistake as an accident.
Michael Howlett, who writes the Customer Experience Decoded newsletter, gave this a name: the AI forgiveness gap. It describes a simple, uncomfortable difference. Give a customer a wrong answer through a human agent, and they tend to accept it as a slip. Give them the exact same wrong answer through an AI agent, and they are far less forgiving.
This article explains why the gap exists, why it matters for any team rolling out AI, and what to do about it.
Why the same mistake costs more
The easy assumption is that customers dislike AI mistakes because AI feels cold. That is not quite it. The real reason is about blame.
When a human agent makes a mistake, the customer pictures a person having a bad moment. The agent was tired, or rushed, or new. The mistake belongs to that person, and it feels like an accident. When an AI agent makes a mistake, there is no person to picture. So the customer blames the next thing up: the company that chose to put AI there. The mistake stops feeling like an accident and starts feeling like a decision. The customer's thought is not "the bot slipped." It is "they decided my problem was not worth a person."
Why this matters more than it looks
The forgiveness gap changes the real cost of an AI mistake. A human error costs you one bad interaction. An AI error costs you one bad interaction plus a dent in trust, because the customer now reads it as a sign of what the company thinks of them.
It also means the gap does not close as the AI gets more accurate. A more accurate AI makes fewer mistakes, which is good. Each mistake it does make still lands harder than the human version. Accuracy reduces how often the gap is triggered. It does not shrink the gap itself.
The four kinds of AI failure, and what each one costs
Not every AI mistake is the same. It helps to separate four kinds, because the fix is different for each.
A wrong answer: the AI gave information that was incorrect. This is the most damaging, because the customer may act on it. A no answer: the AI could not help and the customer hit a dead end. This is costly mainly when there was no easy way to reach a human. A wrong tone: the answer was correct but cold, or cheerful at a bad moment. This one is underrated, and it is where the forgiveness gap bites hardest. A wrong escalation: the AI passed the customer to the wrong place, or failed to pass them on when it should have.
Each kind points to a different fix. Wrong answers point to reducing what the AI is allowed to do on its own. Wrong tone points to where the AI was used, not how its words were chosen. Wrong escalation points to the handoff rules.
What to do about the gap
You cannot argue customers out of the forgiveness gap. It is how trust works. You can design around it.
Use AI where a mistake is cheap and easy to forgive, and keep a person in the moments where a mistake would land hardest. Be honest with customers about what the AI can and cannot do, because a customer who expected less is less likely to read a failure as a broken promise. And when the AI does fail, make the recovery fast and human, because the recovery is your chance to turn the decision back into an accident.
The forgiveness gap is not a reason to avoid AI. It is a reason to place it carefully: use AI where a mistake is cheap to forgive, and keep a person where it is not.