AI agents
AI agents and chatbots that answer, qualify and book
Leads message at 11 pm and expect an answer before they message your competitor. An AI agent can give that answer, ask the right questions and offer a slot. Then a person takes over.
What an AI agent does well, and where it stops
An agent is good at the first ten minutes of a conversation. It replies at once, in the customer’s language, from facts you approved. It collects the name and phone number, asks the qualifying questions your team would ask, and offers the next free slots. When the lead is ready, it books.
It stops at judgement. A diagnosis, a discount, a complaint, a question outside its knowledge: these go to a person in the unified inbox, with the whole thread on the lead record, so nobody asks the customer to repeat themselves.
Built this way, an agent keeps leads from waiting until morning, which is most of what speed to lead means. Your team starts each conversation already knowing who they are talking to.
Where we deploy agents
On the WhatsApp Business Platform. The agent can reply freely for 24 hours after the customer’s last message; after that, only pre-approved templates go out. For campaigns, see WhatsApp marketing.
Instagram and Messenger
Direct messages answered and qualified, with the source of each conversation kept for reporting.
Website chat
A light chat window that loads when the visitor opens it, so it does not slow the page.
Booking
The agent offers real slots from the team calendar your booking forms use, and books. The lead lands in Napexy CRM with its source.
Handover to a person
A clear rule for when a human takes over, and an alert to the right person in the inbox.
Reminders
Confirmation and reminder messages before the appointment, using approved templates on WhatsApp.
How we build an agent
Collect the real questions
We read past conversations from your inbox and list what people really ask, in the words they use.
Write the knowledge base
Approved answers on services, prices, process, location and policies. The agent answers from this and nothing else.
Set qualification and handover rules
What to ask, what makes a lead qualified, and what always goes to a person.
Build and connect
Channels, calendar, CRM and inbox connected, with every message logged.
Try to break it
Your team and ours test it with wrong prices, off-topic requests, angry messages, mixed languages and dialect.
Launch and review
We read a sample of conversations every week, fix wrong or weak answers and update the knowledge base.
Rule-based bot or AI agent?
| Rule-based bot | AI agent | |
|---|---|---|
| How it answers | Buttons and fixed menus | Free text, in the customer’s words and language |
| Good for | Opening hours, simple menus, order status | Open questions, mixed languages, qualification |
| Main risk | Frustrates anyone whose question is not on the menu | Can answer wrongly if not held to approved facts |
| Upkeep | Edit the flows | Update the knowledge base and review conversations |
What we measure
First reply time
How long a customer waits for the first answer, day and night.
Handover rate
Conversations the agent completes alone and those it passes to a person, with the reasons.
Qualified leads
Leads that met your qualification rules, by channel.
Booked appointments
Appointments booked by the agent, and how many of those people came.
Answer quality
Wrong, weak or unsure answers found in the weekly review, and what we changed.
Frequently asked questions
Can the agent speak Arabic and French?
Yes, and English, and it switches when the customer does. A local dialect of Arabic is understood less reliably than standard Arabic or French, so we test with real messages from your market before launch.
Is an AI agent allowed on WhatsApp?
Yes, through the WhatsApp Business Platform and within Meta’s policies. The main practical rule is the 24-hour window: once it closes, the business can send only pre-approved templates.
What happens when the agent does not know the answer?
It is instructed to say so, tell the customer a colleague will reply, and alert your team in the inbox. We review conversations every week to catch any answer that slipped past that rule.
Will customers know they are talking to a bot?
Yes. The agent introduces itself as an assistant. It is fairer to customers, and rules such as the EU AI Act include a duty to tell people when they are talking to an AI system.
Is our customer data used to train the AI model?
We choose business APIs whose terms exclude training on your data by default, and conversations stay in your own tools. Sensitive health questions are handed to your team.
How is this different from marketing automation?
Automation sends planned messages on a schedule or a trigger. An agent holds a conversation. The two work together: see AI workflow automation.
Last updated · By the Napexy team
Keep reading
Tell us where you want to grow.
A 30-minute call with a strategist. You get a plan in plain words and a fixed quote, whether or not you work with us.