AI chatbot vs live chat: when each one wins

Almost nobody who asks this question is choosing between an empty inbox and a robot. They have live chat, it is answering the same forty questions every week, and somebody has asked whether a bot could take some of it.

That is the real question, and it is not about tools. It is about which questions belong to a person, which do not, and what happens at the moment one hands over to the other.

One hour in a support inbox

Eight questions, all real shapes of question. Turn the assistant off and watch where they go.

Five and three is an illustration, not a benchmark. Your own split depends entirely on what you sell and how much of it you have written down. The shape, though, holds everywhere: the questions a person must take have something in common, and it is not difficulty.

What actually separates the two piles

The assistant's five are all questions whose answer already exists in writing. The other three are not harder, they are different in kind.

• A decision nobody wrote down. An exception to a policy is a judgement call. An assistant that invents one has just made policy on your behalf.
• Money already taken, or an emotion. Somebody charged three times wants an apology from a person who can act, not an explanation of the billing cycle.
• A promise about the future. "Yes, it will arrive Friday" is a commitment, and only somebody who can make it should say it.
• Anything with a legal or medical edge. Where being wrong is expensive, the honest answer is a handover.

Notice what is not on that list: complexity. A twelve-part question about compatibility is a great assistant question, because the answer is in your documentation. A one-line question about a refund exception is not.

Where live chat genuinely wins

This is the half most comparisons skip, so here it is plainly. A person reads tone, and can tell an annoyed customer from a curious one. A person can decide, and take the cost of deciding. A person can negotiate, which matters the moment somebody asks for a discount or a custom scope. And a person builds the relationship that makes somebody renew, which no assistant is doing for you.

If your product is sold in conversations rather than bought off a page, live chat is not the thing to automate away. The assistant's job there is to protect those conversations from the forty questions that do not need them.

Where the assistant genuinely wins

Volume and hours, mostly. It answers the same question for the hundredth time in the same voice, at two in the morning, in the language the visitor wrote in. It never gets short with somebody, and it can cite the page each answer came from, which a tired human at the end of a shift cannot.

There is one more thing it does that live chat cannot: it tells you what you have not written. Every question it could not answer is a gap in your content, in a list, with the wording customers actually used.

The handover is the whole product

Both halves only work if the seam between them is good. That means three things.

The failure everybody has experienced as a customer is the opposite: a bot that loops, refuses to escalate, and makes you repeat everything to the human it finally fetches. That is not an argument against assistants. It is an argument against that handover.

What each one costs to keep running

Neither is free after it is installed, and the running costs are different in kind. Live chat costs staffed hours, and the cost grows with the volume of questions. An assistant costs content maintenance: the day your refund policy changes, the answer is wrong until somebody updates the document.

Cheap is not the point. The point is that one cost grows with your traffic and the other grows with how often your business changes.

Questions people ask

Do the A and B exercise on last week's conversations before you buy anything. If your B pile is most of the inbox, hire. If it is a fifth of it, the other four fifths are the argument for an assistant, and the handover is what you should be evaluating.