Chatbots that actually sell: from "hi" to a real customer

A while ago a client came to me with a WhatsApp chatbot that "didn't work". When I asked to see the conversations, I understood the problem in five minutes: the bot replied "I didn't understand your query" to 60% of messages. The customer arrived willing to buy, and the bot slammed the door in their face. Literally.

The problem was not the technology. The problem was that they had built the bot in one afternoon with predefined responses, without thinking about how people actually talk. When someone writes "hi, I wanted to know if you ship to Córdoba", the bot looked for the word "ship" in its list of three possible answers and couldn't find it. Result: customer frustration and lost money.

Chatbots that actually convert share four design decisions, and none of them have to do with having the most expensive AI model.

They truly know the business. I don't mean they have a PDF with the price list. I mean the bot accesses real data: current stock, delivery times by area, current promotions, business hours. When a customer asks "do you have model X in black?", the bot doesn't say "we'll connect you with an advisor": it says "yes, we have 3 units, arrives in Córdoba in 48 hours, shall we reserve it?". The difference in conversion rate is enormous.

They know when to shut up. This is the one nobody thinks about. A good bot detects when a conversation needs a human and makes the transition seamlessly. But not only that: it passes the conversation with all the context summarized. The customer never, never has to repeat their story. "Hi, here's what we discussed: Juan asked about model X in black, shipping to Córdoba, wants to reserve. Can you handle this?". That is respect for the customer's time.

They don't give up after the first message. Most sales don't close in the first conversation. The customer asks, leaves, thinks, comes back to ask something else. Bots that sell have automatic follow-up: if a question was left open, after 24 hours they send a gentle message. "Hi Juan, we were pending on confirming shipping. Did you get a chance to look at the options?". That is where most sales happen. The bot that follows the thread converts; the one that forgets, doesn't.

They measure everything and adjust. Every conversation is logged with clear metrics: what was asked, at what moment people dropped off, what closed and what didn't. That gives you an exact map of where the friction is. If 40% abandon when the bot asks for ID number, maybe that step is unnecessary or should be optional. Without data, you are automating blind.

AI doesn't close sales alone. What it does is remove the friction between someone showing interest and a team member responding. That friction —the wait time, the generic response, having to repeat your problem— is more business than it seems. Every minute that passes between the question and the relevant response, you lose a percentage of the customer.

What I find most interesting is that these principles are not exclusive to large companies. A law firm with three people, an online clothing store, a mechanic shop: all can have a converting bot if they design it from the customer's perspective and not from the company's convenience.

The question is not "can I afford a chatbot?". The question is "how much is it costing me today not to have one that works well?".