Home/Blog/AI chatbots

The same ten questions every month

It wasn't the customers asking. It was the people inside: "how many holiday days do I have left?", "how do I request an advance?". Always the same, always to the same person.

Charo was the HR person at a haulage company in Zaragoza, north-eastern Spain. Eighty employees, half of them drivers who rarely came by the office. Charo did payroll, contracts, hirings, leavers, holidays and, above all, with no mention of it anywhere in her contract, answered questions. All day long.

The questions were almost always the same. "Charo, how many holiday days do I have left?" "Charo, how do I request an advance?" "Charo, what do I do if I lose a fuel receipt?" "Charo, is Father's Day a public holiday?" "Charo, until when can I swap a shift?" Ten questions, twelve, fifteen. The same every week, the same every month, asked by different people. A driver calling from France to ask about the advance. Another dropping by the office to ask about holidays. The inbox full. The phone ringing.

Charo answered patiently, because she was good at her job and because the questions, to the person asking, were important. But every answer pulled her out of what she was doing. And what she was doing — the payroll, the contracts, the real tangles — was what only she could do. The questions, by contrast, could be answered by anyone with the information in front of them. The trouble was that only Charo had the information, in her head and in some folders.

A repeated question is an unsolved question

There's something worth understanding, which is that a question that keeps coming up isn't bad luck: it's a sign that something isn't solved. If fifteen people ask the same thing over a month, it isn't that they're a nuisance. It's that the information they need isn't where they look for it. They look for it in Charo because they can't find it anywhere else.

The company had the answers. They were in the collective agreement, in the employee handbook, in a couple of memos, in Charo's common sense built over so many years. But they were scattered, in documents nobody read because they were long and dull, and written in a language — that of collective agreements — that doesn't answer "how many days do I have left?" but talks about "proportional accrual of the holiday period". So people, instead of reading the agreement, did the logical thing: they asked Charo.

A chatbot that looks inward

When people talk about chatbots, they almost always think of the customer-service one, the website one, the one that answers the people outside. But there's another, talked about less and often worth more: the one that answers the people inside. The internal assistant.

We built one for Charo. Not connected to the website, but to the company's tools. On one side, we gave it all the information that lived loose — the agreement, the handbook, the memos, the rules on holidays and advances — translated into clear answers, not legal paragraphs. On the other, we connected it to the system that held each employee's data, for the questions that depended on the person, like how many holiday days each one had left.

So when a driver wrote from their phone "how many holiday days do I have left?", the assistant looked at their data and answered: "You have eleven days left for this year. Remember you can take them until 31 January." And when they asked "how do I request an advance?", it answered with the exact steps and the link to the form. The same answers Charo gave, but at any hour, to several people at once, and without pulling Charo away from the payroll.

What an internal assistant must have (and almost never does)

Here you have to be honest about two things, because a badly made internal assistant is worse than having none.

The first: it has to say "I don't know" when it doesn't know. An assistant that makes up how many holiday days you have left is a disaster, because people make decisions on it. Charo's, when a question went beyond what it knew for certain, didn't improvise: it answered "better get Charo to confirm this" and passed her the flag. Better an assistant that refers ten per cent of questions than one that answers a hundred per cent and makes part of it up.

The second: the information has to be up to date. If the agreement changes and the assistant keeps answering with the old one, it does harm. That's why the important thing wasn't building it, but making clear who maintained it and how. An internal assistant isn't a painting you hang and forget; it's a plant you have to water.

What Charo got back

The first month, the drivers were wary and kept calling Charo "just in case". Charo, instead of answering, told them: "ask the assistant and, if it doesn't convince you, call me." It nearly always convinced them. Within a few weeks, Charo's phone rang half as much. The inbox dropped. And Charo, for the first time in years, finished the payroll without staying until eight.

Charo wasn't surplus, it's worth saying. What was surplus was Charo acting as a human search engine for answers that were already written down. Her real work — the kind that needs judgement, dealing with people, a head on your shoulders — was still hers and hers alone. What the machine took was the repetitive, the mechanical, the thing that interrupted her thirty times a day.

She put it to me in a sentence that stayed with me.

—I thought my job was answering people. And it turns out my job was everything I wasn't allowed to do because I was busy answering people.

Because that's almost always what a good internal assistant is. It doesn't remove the human touch where it's needed. It removes the thirty daily interruptions that stop the human touch from happening where it really matters.

Inward, not just outward

The chatbot everyone talks about is the customer-service one. But the one that looks inward — the one that answers your own employees' usual questions — often gives more, because it frees valuable people from acting as a human search engine for answers that are already written down.

Two conditions for it to work: it must know how to say "I don't know" and refer to a person instead of inventing, and its information must be kept current. An internal assistant isn't a painting you hang: it's a plant you have to water.

Does your team answer the same internal questions over and over?

If someone in your company spends the day answering the same usual queries, you can almost always build an assistant that answers them on its own and refers what really needs a person. We can look at it together.

Let's talk