Train a WhatsApp AI Agent With FAQs

Outcome Summary

Turn messy WhatsApp questions into a clean FAQ library your AI Agent can answer consistently.
Add “business rules” and do/don’t instructions so replies stay on-policy (and don’t over-promise).
Use Inquiry + Escalation rules so the Agent hands off safely when confidence is low.
Set up Lead capture prompts that collect the right details without turning every chat into a form.

What Clarivo Actually Does (Truth Block)

✅ Clarivo does

Deploys an AI Agent on your WhatsApp Business number to respond to inbound customer messages.
Learns from your business description, FAQ Q/A pairs, services/products list, and additional instructions (tone, do/don’t, sales behavior).
Detects intent and performs Lead capture (when enabled), then shows captured Leads in a filterable table.
Creates an Inquiry (when enabled) when it can’t answer confidently, supporting safe Escalation and clean Human handoff.
Understands WhatsApp voice notes (with transcription visible in the dashboard) and replies in the customer’s language.

❌ Clarivo does not

Run outbound broadcast campaigns or mass messaging.
Guarantee appointment booking (it can collect preferred details; a human typically confirms).
Provide built-in payments processing.
Sync real-time inventory or external systems automatically (it responds based on what you provide/configure).
Act as an omnichannel inbox (it’s WhatsApp-focused).

The Core Problem

Your “FAQ” isn’t really an FAQ—it’s scattered across WhatsApp history, staff memory, and inconsistent replies.
Staff answers vary by person, which creates avoidable disputes (“But you told me…”).
Customers ask in mixed languages, use slang/dialects, and send voice notes—simple keyword rules break.
Some questions require nuance; without clear boundaries, an Agent might try to answer when it shouldn’t.
Lead details come in incomplete (“How much?” without city/service/urgency), making follow-up slow.

Framework

Pull your real questions from WhatsApp: skim recent chats and copy the repeated questions exactly as customers ask them (including informal phrasing).
Group questions by intent (not wording): many variations map to one intent (e.g., “price?”, “how much?”, “cost?”). Write one canonical answer.
Write answers in “policy-first” format: state what you can confirm, what depends, and what you need to ask next.
Add required follow-ups directly into the answer: if an accurate quote depends on details (city, service type, urgency), ask for them inside the FAQ answer.
Define Lead capture triggers: decide which intents should create a Lead (e.g., “book”, “quote”, “availability”, “location-based service”).
Define Inquiry/Escalation triggers: decide when the Agent must create an Inquiry instead of guessing (exceptions, edge cases, non-standard requests, policy conflicts).
Add do/don’t instructions: capture tone, what to never promise, and how to handle sensitive topics.
Handle greetings and routing: decide what should happen on generic greetings, and when to offer quick options (ask a clarifying question vs. wait).
Test like a customer (and like a troublemaker): try ambiguous questions, incomplete requests, and “special case” wording; if answers get risky, tighten rules or move to Inquiry.

Copy/paste templates you can use in Clarivo training inputs

FAQ entry template
- Customer intent: (what they mean)
- Question examples: (the ways customers ask)
- Answer: (short + accurate)
- Ask next: (what you must collect to proceed)
- Boundaries: (what you won’t claim / what varies)
- Escalate when: (conditions that must become an Inquiry)
Business rule template (do/don’t)
- Do: (tone, steps, what to confirm)
- Don’t: (claims to avoid, prohibited promises)
- If customer asks for an exception: (create Inquiry / ask for details / human handoff)
Escalation rule template
- Create an Inquiry when: (low confidence, missing critical info, policy conflict, edge-case request)
- What to capture in the Inquiry: (customer request + missing fields + context)
- What to tell the customer: (acknowledge + set expectation + confirm next step)
Lead confirmation message template
- “Got it — I can help with that. To confirm, you’re asking about {service} in {city}. What’s the best way to reach you, and how urgent is this?”

Use Cases

Use case: Service pricing requests that depend on details

Scenario: A customer sends “How much for cleaning?” with no location or scope.
Recommended approach: Answer with what you can state confidently, then ask for the missing fields you need; enable Lead capture on “quote/pricing” intent.
Common mistake: Putting a single “price” answer with no boundaries, which creates conflict when the real quote depends on details.

Use case: Policy questions (refunds, rescheduling, cancellations)

Scenario: A customer asks for an exception to your policy.
Recommended approach: Provide the standard policy; if the request is an exception, create an Inquiry for safe Escalation and Human handoff.
Common mistake: Letting the Agent “negotiate” policies without a clear rule for exceptions.

Use case: Voice notes in mixed language/dialect

Scenario: A customer sends a short voice note with slang and incomplete info.
Recommended approach: Rely on Clarivo’s voice-note understanding and reply in the customer’s language; ask a single clarifying question to move the conversation forward.
Common mistake: Asking multiple follow-ups at once, which increases drop-off and creates messy lead data.

Decision Checklist

Do we have a single “source of truth” for policies (hours, coverage areas, refunds, booking rules) that we can copy into the Agent instructions?
Which intents should trigger Lead capture (quote, booking, availability, consultation) versus simple FAQ answers?
What are the non-negotiables the Agent must never promise (discounts, guaranteed availability, exact outcomes)?
For each FAQ intent, what is the minimum info needed before a human can act (city, service, urgency, preferred time, contact)?
Which topics must always become an Inquiry (edge cases, complaints, legal/medical nuance, exceptions, high-value deals)?
Do we want the Agent to ask one clarifying question first, or present a short menu of options?
Who owns the Human handoff queue (how often it’s checked, response tone, and closing the loop)?

Constraints

The Agent is inbound-focused: it responds when customers message first.
Your answers are only as accurate as the content you provide; outdated policies/prices should be handled with boundaries and safe escalation.
Not every conversation should become a Lead; over-capturing can annoy customers and reduce replies.
Some requests are inherently “human-only” (exceptions, nuanced negotiations, complex complaints) and should route to Inquiry.
WhatsApp conversations are messy (voice notes, slang, missing context); your FAQ structure needs clear “ask next” prompts.

Practical Example (Illustrative)

Goal: Train a “Pricing” FAQ that avoids risky promises and still moves the chat forward.

Intent: Customer wants a price.
Draft answer (safe + useful):
- Confirm you can help.
- State what pricing depends on (scope/location/urgency) without inventing details.
- Ask for the minimum fields.
- If customer insists on an exception or a complex bundle, create an Inquiry.

How this looks in a real WhatsApp exchange

Customer: “How much is it?”
Agent: “I can help. Pricing depends on the service type and your location. What city are you in, and what exactly do you need? If you share that, I’ll confirm the options and next steps.”
Escalation note (internal behavior): If the customer asks for a custom package or exception, create an Inquiry for a human to review.

FAQ

How many FAQs should I add before going live? Start with the questions you see repeatedly in real WhatsApp chats. It’s better to have fewer, high-confidence FAQs with clear escalation rules than a huge library that includes edge cases.

Should I include pricing in my FAQ answers? Only include pricing if you can keep it accurate and scoped. If pricing varies, state what it depends on and collect the missing details; route exceptions to an Inquiry.

What’s the difference between Lead capture and an Inquiry? Lead capture is for sales-ready intent where you want structured details for follow-up. An Inquiry is a safety mechanism for low-confidence or exception cases where the Agent should escalate instead of guessing.

How do I keep the Agent’s tone consistent across languages? Add tone rules in your additional instructions (formal vs friendly, emoji rules, how to greet). Clarivo is designed to reply in the customer’s language, so your tone rules should be written clearly and simply.

Can the Agent handle voice notes? Clarivo is designed to understand WhatsApp voice notes and show the transcription in the dashboard, which helps with review and handoff.