Chatbot or person: what is worth automating in customer service and what is not

Chatbot or person: what is worth automating in customer service and what is not

A badly placed bot drives away more customers than it serves. The rule for deciding what to automate, what must always reach a person and what the law requires when a machine answers.

· Customer support

Everybody wants a chatbot until they have to be a customer of one. The experience of going round a menu that does not cover your case, with no way of reaching anybody, is what makes people leave for a competitor without saying a word.

And even so, automating support is a good idea. What has to be got right is what.

The rule that decides what gets automated

Automate whatever meets all three conditions at once:

  1. The answer is always the same, and it is written down somewhere.
  2. Getting it wrong is not expensive. Giving the wrong opening hours can be corrected; giving the wrong legal deadline cannot.
  3. The customer prefers the immediate answer to the good answer. Opening hours at eleven at night: yes. A complaint: absolutely not.

If a single one fails, it goes to a person.

What works

  • Opening hours, address, payment methods, delivery times. The bulk of the messages many businesses get.
  • The status of an order or a file, if the bot can genuinely look it up. A bot answering «check your private area» has served nobody.
  • Collecting details before handing over. So that when it reaches the person, it arrives with the subject, the customer number and the problem already written down.
  • Appointments and bookings. It is a closed task, with clear rules, and it works well outside opening hours.
  • Warning proactively. «Your order ships today», «your instalment is due on the 20th». This cuts incoming messages far more than any bot.

What must never be automated

  • A complaint. Somebody complaining is deciding whether to stay with you. A bot there confirms that you do not care.
  • Anything with legal, tax or medical consequences. A deadline, an amount, an obligation. The plausible but false answer is these systems' speciality.
  • Cancellation. Putting a bot in the way of leaving is regulated as an unfair practice in several countries, and it always ends in a one-star review.
  • A customer who has already said they want to speak to somebody. When they ask, they get passed over. Without insisting.

The three things that stop a bot being a nuisance

A visible exit from the first message. «Type person and I will put you through.» A bot with a clear exit is tolerated; one without it is hated.

Saying what it does not know. «I do not know this, I will put you through» is infinitely better than inventing an answer. And with generative AI that is a design decision: it has to be tied to the real documentation and forbidden from improvising.

Not repeating the question. If the customer already gave their order number, the person picking up the conversation has it in front of them. If not, you have added a step instead of removing one.

What the law requires when a machine answers

Since 2 August 2026, art. 50 of the European artificial intelligence regulation applies: if the customer is interacting with an AI system, they have to be told. A friendly name hinting that it is a robot is not enough: you inform them.

And as soon as the bot collects data — name, phone, order number — there is personal data processing: legal basis, information to the user and a contract with the tool provider. In Spain under the GDPR; in Chile, under Ley 21.719. We develop it in using AI with customer data within the law.

Start by measuring, not by installing

Before buying anything, take last month's messages and classify them. It nearly always turns out that four or five questions account for the majority. With that you already know what to automate, and often the conclusion is that you do not need a bot: you need to answer those five questions on your website, where the customer looks for them before writing to you.

Saying that takes work away from us, but it is how it is.

If this sounds like you

We build automated support that knows when to keep quiet and hand over to a person: customer service. If what you want to automate is not the support but the repetitive work behind it, that is AI and automation, and it starts with knowing what can genuinely be automated through the automation audit.

We are Mindset & Code: automation, data and development for small businesses. We deal with customers every day, so we know what happens when the person writing is angry. You can see what we do and what it costs.

Sources: Regulation (EU) 2024/1689 on artificial intelligence, art. 50, applicable from 2 August 2026; the GDPR and Ley 21.719 of Chile for data processing. General guidance.