The channel you've already chosen (even if you didn't decide it)
In Spain, France, Venezuela and South Florida, WhatsApp is where your customers prefer to write to you. You didn't choose it; they did. And they write at any hour: price questions at 10 pm, appointment changes on Sunday, "do you have this in stock?" while your team is in a meeting.
Every message left unanswered for more than an hour is a customer who is probably also writing to your competitor. The first one to reply well gets the conversation. It's that simple and that brutal.
Hiring someone to reply 24/7 isn't viable for a small business. An AI assistant is. Let me explain how a well-built one works, because the badly built ones are worse than nothing.
What a WhatsApp AI assistant does (when well designed)
Answers frequent questions instantly. Hours, location, prices, services, availability, terms, payment methods. With your real information, in your tone, in the customer's language.
Qualifies. When someone asks about a service, it asks the two or three questions your team would ask: when, what for, what size. And saves the answers.
Books. Connected to your calendar, it offers real slots, confirms the appointment and sends a reminder. Without anyone stepping in.
Escalates with context. When the question goes beyond what it knows, or when the customer asks for a person, it alerts your team with a summary of the conversation. The person joins already knowing everything.
Logs. Every conversation lands in your CRM or your tracking sheet, with the name, the inquiry and the outcome.
Keeps quiet when it should. If a customer is upset, mentions a complaint, or the conversation gets delicate, the assistant stops and alerts. It doesn't try to be clever.
What it must never do
This is where a serious project differs from a botched one:
- It doesn't make things up. If it lacks the information, it says it will check and alerts a person. An assistant that invents prices or delivery times costs you customers and credibility.
- It doesn't close sales on terms nobody approved. It can quote list prices. It doesn't negotiate discounts.
- It doesn't handle complaints. It detects them and hands them to a person immediately.
- It doesn't pretend to be human. It introduces itself as an assistant. Customers appreciate it and the law in Europe increasingly requires it.
- It doesn't pester. No unsolicited messages. WhatsApp blocks those who do, and rightly so.
Definitive test before launch: feed it the 30 hardest conversations you've had with real customers and watch how it responds. If in any of them it invents something or fails to escalate when it should, it isn't ready.
What you need: WhatsApp Business API
There are three "WhatsApps" for businesses:
- Regular WhatsApp. Your personal number. No good for automating anything.
- WhatsApp Business (the app). Free, with a business profile, catalog and manual quick replies. Fine to start, but doesn't allow AI or real integrations.
- WhatsApp Business API (or Platform). The version for connecting systems. This is what makes the AI assistant, CRM integration and calendar possible.
For the API you need a verified business account with Meta, a dedicated number (can be the one you already use, migrated) and a provider that gives you access. Messaging costs are low: Meta charges per conversation started, at prices ranging from a few cents to about 10 cents depending on country and type. A small business with 500 conversations a month usually pays €20 to €60 in messaging.
What it costs to set up
Implementation: €2,500 to €5,000 for an assistant with FAQs, qualification, booking and escalation to people. Includes API setup, designing the instructions with your information, connecting your calendar and CRM, testing with real cases and training your team.
Monthly: €40 to €150 in usage (AI model plus Meta messaging), depending on volume.
Return: in a clinic with 300 WhatsApp inquiries a month, an assistant resolves 70 percent without intervention and books directly. If that recovers 10 appointments a month previously lost to slow replies, at a ticket of €80, that's €800 a month. Plus the 40 hours your team spends on something else.
How to implement it well (in four weeks)
Week 1: the information. We gather everything the assistant needs to know: services, prices, terms, real frequently asked questions (the actual ones, pulled from your conversations), your brand's tone, what to escalate and to whom.
Week 2: build. API configuration, assistant design, calendar and CRM connections, escalation rules.
Week 3: testing. With your team, with old real conversations, with hard cases. We adjust until it answers the way your best employee would.
Week 4: supervised launch. The assistant replies, but a person reviews the conversations every day. Fine-tuning. From the second month, weekly review.
That's the process I follow in the AI automation service, and it adapts well to restaurants, clinics, real estate, workshops, academies and any business with a volume of inquiries.
Where it works especially well
Bookings and appointments. Clinics, aesthetic centers, workshops, professional practices. The assistant manages the full calendar. It's the basis of the bookings and appointments solution.
Restaurants. Reservations, daily menu, allergens, pickup orders. On weekends, when nobody can answer the phone, the assistant keeps working.
Real estate. Property information, buyer qualification, scheduling viewings. Filters the curious and delivers real contacts.
Retail and distribution. Availability, volume pricing, order status, delivery times.
Professional services. First consultation qualified and booked, without the professional wasting time on calls that lead nowhere.
Mistakes I see constantly
Launching with incomplete information. The assistant answers "I don't have that information" to half the questions. Customers get frustrated. Fix: gather everything properly before building.
Not defining escalation. The assistant tries to solve everything and ends up giving vague answers. Fix: clear rules for when it stops and alerts a person.
Forgetting the tone. An assistant that talks like an instruction manual in a hair salon, or like a buddy in a law firm. Fix: design the tone with examples from your own conversations.
Not reviewing the first weeks. Launch and forget. Fix: daily review in the first month. That's where you catch the cases nobody anticipated.
In short
A well-designed AI assistant on WhatsApp replies instantly, qualifies, books and escalates what it should. It costs less than a month's salary and works every day of the year. Badly designed, it annoys your customers and damages your brand.
The difference lies in the information it's built with, the escalation rules and the supervision in the first month.
If you receive more than 100 WhatsApp inquiries a month, book a call. In 15 minutes I'll tell you whether an assistant makes sense in your case, what percentage of inquiries it could resolve and what return to expect.
