What AI won't answer for you

A founder wrote to me last month to say he didn't need us: he was going to build his French customer support himself, with AI. He was right about a good part of it, and I'd rather say so upfront than defend my own trade.

AI handles tier one very well. Where is my order, how do I change plan, do you have this in stock. Those questions arrive by the hundred, they all look alike, and the answer already sits in your documentation. Refusing to admit that would be dishonest, and you'd see it for yourself in the first week anyway.

The problem isn't what goes away, it's what stays

Here's what nobody tells you when they sell you automation. It doesn't reduce your work evenly. It removes the easy part and leaves the hard part.

Before, out of a hundred requests, you might have had seventy routine ones and thirty delicate ones. After, you're left with the thirty. Your volume dropped by two thirds, but your mental load hasn't moved, because the thirty that remain are exactly the ones that need judgment. The furious customer whose order is lost for the second time. The refund request outside policy that you should perhaps grant anyway, because that account is worth twelve thousand a year. The case that touches compliance. The question that isn't really a question but a cancellation signal.

Three things a system won't do

It won't decide to break your own rule. Sometimes that's the right commercial call, and it takes knowing the customer's value, the history of the relationship, and what your leadership would accept. A system will apply the rule, correctly, and you'll lose the customer.

It won't notice that the problem isn't the one being presented. I've written elsewhere about a customer refunded every week over a billing issue, when the real cause was a wrongly ticked box at signup. Refunding was the right answer to the wrong question, and no automation would have caught it, because it was doing exactly what it was told.

And it won't carry the responsibility. When a reply goes out under your name and it's wrong, someone has to answer the customer, fix it, and make sure it doesn't happen again. That someone can't be a tool.

What we do with it

We don't sell the opposite of AI. We use it every day on the repetitive half: sorting requests, drafting replies from your documentation, data entry, reporting. It would be absurd to go without.

The difference comes down to a simple principle. AI handles the volume, a person makes the decisions, and nothing reaches your customer without a human having read it. If a workflow starts producing work we'd have to correct, we switch it off rather than patch around it.

So if you're building your French support with AI, do it. You'll solve your seventy percent, and that's a real gain. The question that matters will arrive three months later, when you notice the remaining thirty percent still takes your evenings, and that they're now the only ones getting through.

If you're already there, write to me. I'll tell you honestly whether we're the right fit.
Estellina