Answering the same questions over and over, dealing with enquiries outside opening hours, making a customer wait because nobody is available… These are daily realities for thousands of SMEs. The good news is that artificial intelligence can solve them today, without setting up a call centre or hiring more staff. In this article we explain how AI chatbots for customer service work and what they can do for your business.
Almost everyone has talked to a bad chatbot. That experience (the loop of irrelevant answers, the impossibility of reaching a person) is why many companies dismiss the idea. The problem was not the chatbot: it was how it was built.
What changed from the ones a few years back
The previous generation of chatbots worked on rules and keywords. If the customer typed “opening hours” they gave the opening hours; if they typed “what time do you open tomorrow”, they did not understand. That is where the frustration came from: they forced the user to guess the magic words.
Today's ones interpret intent. They understand that “are you open on Saturday?” and “weekend hours” are asking the same thing. That difference is what makes it worth considering today and not five years ago.
What it actually solves
Most enquiries a company receives repeat. Opening hours, prices, order status, how to book, what paperwork is needed, whether you work with a particular insurer. They are the same fifteen questions every day.
Each one interrupts someone who was doing something else, and none requires judgement. That is exactly where an assistant adds value: it does not replace your team, it takes off their hands what they should not be doing.
There is a secondary benefit often overlooked: it answers at the time people ask. A good share of enquiries arrive outside opening hours, when the customer has a moment. Replying at nine the next morning is already too late for many people.
The four rules of a chatbot that does not annoy
1. Trained on your information, not on generalities
An assistant that answers vaguely is worse than none, because the customer wastes time before writing to you anyway. It must know your hours, your services, your terms and your exceptions. That means feeding it your real documentation.
2. It knows when it does not know
This is the difference between a good one and a bad one. When the question is outside its scope, the right response is not to improvise: it is to hand the conversation to a person with the full history, so the customer does not have to repeat what they have already said.
An assistant that insists on answering what it does not know does more harm than having none.
3. The route to a person is always visible
Hiding the button to talk to someone to “save” enquiries is counterproductive. A customer who wants a person will look for one anyway, and will arrive angrier. Put it in plain sight from the first message.
4. It says what it is
There is no need to pretend it is human. People spot it and it feels bad. An assistant that presents itself as what it is and answers well builds more trust than one that pretends and fails.
Where to put it
WhatsApp is where most enquiries arrive in Spain for local businesses. If you sell to consumers, start here.
The website captures whoever is comparing right then, which is the highest purchase intent.
Email is the least glamorous and often the most profitable in B2B, because it is where the long enquiries arrive that take the most time to answer.
You do not need to be on all three. Look at where people already write to you and start there.
What it must NOT do
- Give information with consequences without review. Nothing resembling medical, legal or financial advice.
- Close complex sales. It can qualify and book; closing is for people.
- Handle angry complaints. Detect the tone and hand over. An upset customer plus an automated reply is an expensive combination.
- Ask for sensitive data over chat. Ni DNI, ni datos bancarios, ni información de salud.
How to know whether it is working
The metric almost everyone looks at (number of conversations handled) says nothing. These three do:
Percentage of conversations resolved without a person. That is the real saving. If it is below half, something is wrong with the training.
How many handovers end in a complaint. If customers arrive at a person already upset, the assistant is holding on too long.
Which questions it could not answer. It is the most valuable list you will have: it tells you exactly what is missing from your documentation, and often what is missing from your site too.
Where to start
Gather the questions you get most. They come from email, from WhatsApp and from asking whoever answers the phone. Twenty or thirty well answered cover most of the volume.
Start with one channel, measure for a month and expand. Launching on three channels at once multiplies the places something can go odd exactly when you have least experience.
If you want to see what else can be automated alongside, here are the processes that work best. And to find out how much time answering the same thing every day is costing you, try the savings calculator.
Want to automate your company with AI?
We help you identify which processes you can automate and how. The first session is free and with no obligation.
Request your free diagnosis