An AI agent on WhatsApp: what it actually takes to make it work

For many small businesses, WhatsApp is where the real conversation with customers happens — more than email, sometimes more than the phone. That is why a well-built AI agent there has a direct impact, and a poorly built one shows immediately.

First: the right infrastructure

An AI agent on WhatsApp does not connect to the company’s regular mobile app; it needs the WhatsApp Business API, managed through a provider authorized by Meta. This is not a minor technical detail: without it, you cannot reliably automate responses or scale to multiple simultaneous agents, and the number risks getting blocked if you automate on top of the regular app.

Tone matters more than it seems

An agent that replies with robotic, generic phrases stands out, and not in a good way. The goal is not to hide that it is an automated assistant (that breeds distrust if the customer finds out later), but to make sure its way of speaking is recognizably your brand: the same warmth, the same level of formality, the same expressions someone on your team would use.

Watch out: an agent that sounds like a generic AI template loses trust faster than one that takes a bit longer to reply.

Hours: the most common complaint

A frequent mistake is having the agent reply at any hour with the same tone as during business hours, setting the expectation that someone is “there” at 3 a.m. The fix is clear from the first message: outside business hours, the agent can keep resolving simple queries (hours, prices, availability), but it must be explicit about when a real person will follow up on anything that needs a human.

What to automate and what not to

• Yes: frequently asked questions, availability, order tracking, appointment booking, initial lead qualification.
• Not directly: complaints, off-policy price negotiations, any message where the customer explicitly asks to speak to a person.

How to measure whether it is working

“How many messages did the agent answer” is not enough. The metrics that matter are how many conversations were resolved without human intervention, how fast it escalated when it should have, and whether the customer asked the same thing again (a sign the first answer was not enough).

How we do it at Bitora

We set up the WhatsApp Business API correctly from the start, design the agent’s tone from your team’s real conversations (not a generic one), and define the hours and escalation rules before the first automated message, not after the first complaint.

If WhatsApp is an important channel for your business, see the full approach on AI agents and automation, or check our chatbots for business page directly.