AI Agents · WhatsApp AI agents
WhatsApp AI agents that work with your systems.
Your customers already message you on WhatsApp. A WhatsApp agent answers the routine questions, looks things up in your own systems, and passes anything else to a person with the conversation attached.
The problem
Where the time goes today.
- A phone that never stops: the same “where is my order?” and “what are your hours?” all day.
- Replies that wait until someone is free, so customers go elsewhere.
- Conversations that live on one person’s phone, with no record and no handover.
- Reminders and follow-ups sent by hand, or not at all.
How it works
What the agent does, step by step.
- A customer messages your numberThrough the WhatsApp Business Platform, not a personal phone.
- The agent identifies the intentStatus, booking, price enquiry, complaint or something it cannot handle.
- It looks up the answerFrom your approved data: orders, appointments, policies. It does not guess.
- It replies, or proposes an actionRoutine answers go straight back. Anything with consequences is queued for a person.
- It escalates cleanlyA human takes over with the whole conversation and the data already found.
Integrations
What it connects to.
“Used in our projects” means we have built and run that kind of connection. Anything else needs a feasibility review before we commit.
| System | Status | Note |
|---|---|---|
| WhatsApp Business Platform (Cloud API) | In our product plans | Reminders and recall are a planned feature of Odontofusion PMS. |
| Your order, booking or case system | Used in our projects | Through approved read-only queries. |
| CRM | Proposed: feasibility review | To log conversations against the customer. |
| Email and SMS fallback | Proposed: feasibility review | For customers not reachable on WhatsApp. |
In practice
Examples of the work.
Illustrative exchanges with fictional data. They show intended behaviour, not recorded client conversations.
Status enquiry
Customer: “Is my order ready?”
Agent: looks up the order for the verified phone number and replies with the stage and expected date.
Appointment reminder
Approved template sent the day before. Customer replies “reschedule”.
Agent offers two free slots and proposes the change; staff confirm.
Security and control
What keeps it safe.
- Customers are identified by their registered number before any personal detail is shared.
- Only approved message templates are used for business-initiated messages.
- Opt-in is recorded, and customers can stop messages at any time.
- No payment or card details are collected in chat.
- Conversations are logged and retained according to your policy.
Implementation
How we would approach it.
- Platform readinessBusiness verification, number registration, template approval. Meta’s rules and timelines apply.
- Conversation designThe intents, the answers, the tone, the languages and the escalation rules.
- Integration and testingConnect read-only lookups, test in each language you need, then add actions behind approvals.
- Supervised launchStart with a subset of enquiry types and a person reviewing, then widen.
Commercial considerations
What it costs, and what it saves.
- Meta charges per-message fees for many business conversations. Pricing and rules change, so we check the current price list when scoping.
- Meta’s policies restrict general-purpose assistants on the platform. A business-specific agent with a defined job fits; we confirm against the current policy at the start.
- Model usage, hosting and monitoring are the other recurring costs, all itemised in the proposal.
- Language quality varies. We test each language you need, including Arabic or Amharic where relevant, before launch.
Related
Keep exploring.
Build this agent with Odontofusion.
Tell us the job. We will tell you honestly whether an agent is the right tool for it.