How to Build Your AI Sales Agent on WABA.FLOTA

WABA.FLOTA's AI sales agent is what actually replies to your customers on WhatsApp — it answers questions, quotes prices, takes orders, and can even pull live data from your own website mid-conversation. This guide covers how to configure it properly, from the basic personality prompt to giving it real tools to work with.

Where to configure it

Go to Dashboard → Bot Prompt. This is the single place that controls how your AI agent behaves — no code, just a text box that becomes the agent's "system prompt" (the instructions it follows on every single customer message).

Writing a good system prompt

A weak prompt ("You are a helpful assistant for my shop") produces generic, unconvincing replies. A strong one is specific. Structure yours around these four things:

1. Who you are — business name, what you sell, city/region if relevant.
2. Tone — formal or casual, how long replies should be, whether to use emojis.
3. Boundaries — what it should never promise (exact delivery dates, discounts it can't verify, medical/legal claims) and what it should hand off to a human for.
4. Concrete facts — pricing logic, delivery areas, payment methods, return policy. Anything a real salesperson would know by heart.

Example structure:

You are the WhatsApp sales assistant for [Business Name], a [what you sell]
based in [city]. Be warm but concise — 2-4 sentences per reply, no walls of
text. Use the customer's name if they give it.

Never invent a price, stock status, or delivery date you don't actually
have — say you'll confirm instead of guessing.

We deliver within [city] in 1-2 days, nationwide in 3-5 days. We accept
[payment methods]. Returns accepted within 7 days if unused.

Save it, then send yourself a test WhatsApp message to see how it responds — iterate a few times until the tone feels right.

Giving your agent live data (not just a script)

A prompt alone only gives the agent knowledge you typed in — it can't see real inventory, real order status, or anything that changes day to day. For that, connect a data source:

• Dashboard → Menu — for a simple product catalog (name, price, photo) the agent can quote directly and use to build orders.
• Dashboard → Website API — for anything more dynamic: live stock counts, order status lookups, or even letting the agent update a record (like marking an order dispatched) on your own website's backend. This is the difference between an agent that says "let me check and get back to you" and one that actually checks, in the same reply. Full walkthrough: Connecting your website to WABA.FLOTA.

What the agent can already do without extra setup

Beyond replying with text, the agent can — mid-conversation, when it judges it's appropriate — send images, documents, videos, location pins, contact cards, quick-reply buttons/lists, and reactions. It also silently tracks lead temperature (cold/warm/hot) and order status changes for your own dashboard, without ever mentioning that tracking to the customer.

Testing before you go live

Before pointing real customers at it:

1. Message it yourself with the 5-10 questions customers ask you most often in real life.
2. Try a deliberately vague question ("what do you have?") — a well-written prompt should ask a clarifying question rather than dump everything at once.
3. Try to get it to promise something it shouldn't (a discount, a guaranteed same-day delivery) — confirm your boundaries section holds.
4. If you connected Website API tools, ask something that requires a real lookup ("do you have the blue one in stock?") and confirm it actually checks rather than guesses.

Common questions

Can I turn the AI agent off for specific conversations and reply myself?
Yes — Dashboard → Messages gives you the shared inbox view where a human agent can take over any conversation at any time.

Does the AI agent work in languages other than English?
Yes — set your preferred language in the tenant's language preference, and reasonable multilingual prompts work well; the agent replies in whatever language the customer writes in when that's supported by your plan's translation features.

How do I stop the agent from making up information?
The single biggest lever is your prompt's boundaries section — explicitly tell it what it doesn't know and should never guess. The second lever is connecting real data (Menu or Website API) so it has facts to draw on instead of filling gaps itself.