Martha, the electrical supplies store and the Christmas lunch
Martha runs the electrical supplies store her father opened in Cartagena back in 1987. Thirty-seven employees, a catalog of twelve thousand part numbers, customers calling to ask whether the 40-amp ABB breaker comes with screws or without them. That kind of thing all day long.
One Thursday in November, at the local business association's Christmas lunch, the guy who runs the repair shop next door told her he had put in artificial intelligence.
—What does it do?
—It answers the customers.
—Right. What does it tell them?
—Well, whatever needs to be said to them.
Martha didn't ask anything else. When someone says "whatever needs to be said" without getting specific, it's because they don't know, or it doesn't work.
Three months later, the repair shop guy had taken the chatbot out. He said customers were complaining. One lady asked about the price of a shut-off valve and it spat out something about the privacy policy.
The line between what works and what's smoke
What Martha didn't know is that there's a huge difference between dropping in a generic chatbot and building a system that understands your business. The difference isn't technology. It's knowledge.
The repair shop's chatbot didn't know anything about shut-off valves. It didn't know that when a plumber asks "do you have the three-quarter-inch in brass?" he wants to know price, stock and lead time, in that order. It didn't know that a regular consumer asking the same thing needs an explanation of what "three-quarter-inch" means and why brass costs more.
The AI that works starts from groundwork: someone sat down with the business owner, understood how their customers talk, what they ask, what they need to hear and what wears out their patience. Then that information gets translated into precise instructions for the system. It's not magic. It's translation work.
The one that's smoke installs in an afternoon. Someone sells you a generic solution, connects it to your site and says you're done. It works on the first easy question. On the second, it starts making things up. On the third, the customer closes the window and calls on the phone.
Classifying documents: the part that doesn't make the news
There's another area where AI delivers concrete results with no need for science fiction: document classification. Martha gets orders, technical questions, complaints and supplier invoices arriving every day. All to the same inbox. Alice, the office manager, spends the first hour of every morning reading each email and forwarding it to the right person. She's been doing it for nine years.
A smart classification system doesn't replace Alice. It reads each email, identifies whether it's an order, a question, a complaint or an invoice, and routes it to the right department. Alice is still there, but now she spends that first hour solving problems instead of sorting mail.
It doesn't sound spectacular. It doesn't make the news. But it saves four hours a week of skilled work and brings routing mistakes down to practically zero.
Reality is more boring and more useful
The problem with the public conversation about artificial intelligence is that it swings between two extremes. On one side, people promising that AI will transform your business top to bottom. On the other, people saying it's all a bubble. Neither of them is right.
Reality is more boring and more useful. There are specific tasks where a well-built system saves time and money from the first month:
- Answering frequently asked questions with precision.
- Classifying documents.
- Prioritizing tickets by urgency.
- Detecting patterns in customer complaints.
None of this takes a six-month project or a multinational-sized budget. What it does take is someone understanding your business before writing a single line of code.
Two mornings listening
A developer spent two mornings with Martha understanding how the store worked before proposing anything. Two mornings. Not a market study, not a digital audit. Two mornings listening to how customers talk, reading the emails coming in, watching how the team responds.
Martha now has a system that answers eighty percent of the technical questions about her catalog. It doesn't answer everything. When it doesn't know, it says so and hands the question to a person. Customers don't complain. Some of them haven't even realized a human isn't replying. Others have, and don't care, because the answer is correct and arrives in ten seconds.
The guy at the repair shop is still answering the phone by hand.
The difference, summed up
- The AI that works is born from understanding the business. Someone sits down, listens and translates that knowledge into instructions for the machine.
- The AI that's smoke is born from an afternoon of installation. Generic, no context, making up answers when it doesn't know.
- Spectacular is a waste of time and money. Useful is boring. And it works.
Can AI help you with something concrete?
Probably yes. But the honest answer starts with a conversation about your business, not about technology. No commitment, no smoke.
Let's talk