AI chatbots that sell: the guide for small businesses that want to stop replying and start closing

Most chatbots reply. Very few sell. And the difference between the two is, literally, money: an AI chatbot built to sell doesn’t stop at “we’re open from 10 to 7” — it finds the product in your catalogue, quotes real prices, builds the order and sends the payment link, all in the same conversation. This guide explains what a chatbot needs in order to genuinely sell, how to hand it your sales script, and which metrics tell you whether it’s working.

Replying is not selling

Let’s draw the line clearly, because the market lumps everything together under the word “chatbot”:

• A bot that replies answers frequently asked questions. Useful, but when the customer says “I want to buy it”, it answers “a representative will contact you”. The sale goes back to depending on someone being free — and the customer was already holding their card.
• A bot that sells understands purchase intent and acts on it: it checks stock, gives the real price, handles objections using your business information, quotes and takes payment. The conversation ends in a confirmed payment, not in “we’ll be in touch”.

The cost of staying at the first level is concrete. In WITHMIA’s pilot with real businesses, going from replying in hours to replying (and selling) in under a minute translated into +30% conversion with 95% customer satisfaction. Not because the bot is charming: because it catches the buyer at the exact moment they want to buy.

The 4 pieces of a chatbot that sells

For AI to sell, it needs the same things a good human salesperson needs: product, knowledge, a script and a till.

1. A connected catalogue (with real stock and prices)

The golden rule: the bot must never invent a price or promise stock that doesn’t exist. You get there by connecting it to the source of truth — your store. In WITHMIA the assistant reads directly from Shopify, WooCommerce, MercadoLibre, your own API or a catalogue managed inside the platform itself. When the customer asks “do you have the blue parka in a medium?”, the bot searches the live catalogue and answers with what’s actually there, at today’s price.

No out-of-date spreadsheets, no “let me check and get back to you”.

2. Knowledge of the business

Price and stock don’t close a sale on their own. Customers ask about delivery, exchanges, warranties, lead times, payment methods. All of that lives in the assistant’s training: you load your policies and your business information, and it answers from there. If it doesn’t know something, it should say so and escalate — not fill the gap with imagination.

3. Sales playbook: your script, running 24/7

This is the step up in quality that almost nobody takes advantage of. A good salesperson has a script: what to ask first, what to recommend, how to handle “I’ll think about it”, when to offer the cheaper alternative. Your chatbot can have one too.

In WITHMIA you define this in the bot instructions field (Training section): you write, in plain language, how you want it to sell. For example:

“Always ask who the product is for and what they need it for before recommending anything. Suggest the mid-range model first. If they ask about discounts, offer free delivery over $50.000. If the customer hesitates, offer to hold the order for them and mention that only a few units are left — only if that’s true. If there’s a complaint, hand over to a human immediately.”

That’s your commercial strategy running at 2 in the morning, on a Sunday, with the same discipline as Monday at 10.

4. Taking payment inside the chat

Friction kills sales. Every extra step — “transfer to this account”, “go to the website and find the product” — loses buyers along the way. A selling chatbot generates the payment link inside the conversation itself (in Chile: Flow, Webpay or Mercado Pago), the customer pays in pesos without leaving the chat, and the payment is confirmed and recorded.

The full flow: from “do you have…?” to confirmed payment

Here’s what an end-to-end automated sale looks like, based on a real case:

1. 11:04 pm — A customer asks on WhatsApp whether you stock a certain product. The shop closed at 8 pm.
2. 11:04 pm — The assistant searches the catalogue, confirms stock and replies with price, photo and description.
3. 11:05 pm — The customer asks whether you deliver to their area. The bot answers with the real delivery policy.
4. 11:06 pm — “Right, I’ll take it.” The assistant builds the order and sends the payment link.
5. 11:07 pm — Payment confirmed. The order is recorded and the lead lands in the CRM with the whole conversation attached.
6. 8:00 am — The owner wakes up, sees a closed sale and only has to ship it.

This isn’t theory: it’s exactly the kind of operation Motos GYV, one of WITHMIA’s first customers, moved to when they cut their response time from 24 hours to under 1 minute, with a real catalogue and a payment link in the chat.

And one detail that keeps the house in order: every conversation with purchase intent is recorded as a lead in a pipeline — with its stage, its value and its history — so no quote ever gets left forgotten on somebody’s phone.

The metrics that matter (and the ones that don’t)

Automating without measuring is just decorating. These are the metrics that really tell you whether your chatbot sells:

• Time to first response. The mother of all speed-to-lead indicators. Target: under 1 minute, at any hour. The research shows that replying within the first 5 minutes multiplies conversion compared with answering hours later.
• Conversations handled outside opening hours. That’s revenue that didn’t exist before. If 30% of your enquiries arrive with the shop closed, that’s your hidden growth.
• Conversation → sale conversion rate. Out of every 100 conversations with purchase intent, how many end in a payment? Compare before and after automating.
• Quotes sent and their close rate. A bot that quotes on the spot stops the lead going cold — the quote you send “tomorrow” is the one you lose.
• Sales closed with a payment link. Cash you can attribute directly to the channel.
• Human handover rate. Neither too high (the bot isn’t resolving anything) nor zero (careful: it may be improvising where it should be escalating).

And the vanity metrics? Messages sent, users “reached”, conversation length. None of that pays wages.

Mistakes that kill the sale

• A generic bot with no catalogue. “We have a wide range of products, visit our website” is a goodbye, not a sale.
• Letting the AI improvise prices or stock. A false promise costs more than a lost sale. That’s why a connected catalogue isn’t optional.
• An aggressive playbook. Pushing in every message scares people off. The script should sell the way your best salesperson does: useful first, closing second.
• Not escalating in time. Complaints, big negotiations and unusual cases go to a human — with the full context of the conversation, not from scratch.
• Not reading the conversations. The bot’s chats are a goldmine: what people ask, where they hesitate, which product they want and you don’t stock. Read them every week.

What it costs and how to start

You can test the concept without investing: WITHMIA starts with a 14-day free trial, no card required, and the Inicial plan ($9.990 CLP/month) brings 500 AI messages a month, WhatsApp, Instagram and web chat — enough to train the assistant on your business and watch it really hold a conversation.

To sell with a connected catalogue, quotes and payment links, paid plans start at $24.990 CLP/month (VAT included): Pro $24.990, Team $44.990 and Max $149.990, with 15% off if you pay annually. Paid plans run on Anthropic’s Claude models — the same AI family behind the most demanding products on the market — and every subscription charge automatically issues its electronic receipt or invoice with the Chilean tax authority (SII). Full details at /en/precios.

To get off to a good start:

1. Create the account and train the assistant with your information.
2. Connect the catalogue (or load your main products).
3. Write your playbook: 5 to 10 clear instructions on how to sell.
4. Play the difficult customer yourself, and adjust.
5. Connect your WhatsApp and let it work. Read the first week’s conversations.

Frequently asked questions

Is a sales chatbot worth it if I only have a few products?

Yes — better, even: with 10 or 20 products the assistant knows every one of them inside out. If you don’t have an e-commerce store, you can manage the catalogue inside the platform itself, no online shop required.

What happens when a customer asks something the bot doesn’t know?

A well-configured assistant recognises it, answers with what it does know and passes the conversation to your team, who see it in the inbox with the full history. The one thing it must never do is make things up — which is why it matters that it answers from your catalogue and your information, not from its imagination.

How does the chatbot take payment? Is it safe?

It generates a payment link from established Chilean processors (Flow, Webpay, Mercado Pago) and the customer pays on that gateway, not by “reading out their card in the chat”. You get the payment confirmation and the order is recorded.

Can I jump into a conversation whenever I want?

Yes. Every conversation arrives in one unified inbox; when you or someone on your team starts replying, the assistant steps aside and won’t talk over you. It picks up again afterwards.

---

The best way to judge it is live: create your free account, load your products and ask it the question your customers ask you every single day. If you like how it sells, you already know where to grow.