What an AI-answered WhatsApp line can (and can't) do

Most small businesses we talk to already run half their customer conversations through WhatsApp — a booking question here, a "do you have this in stock" there, a reschedule request at 9pm. It's rarely written down as "our support channel." It just is one, by default, because that's where customers already are.

That makes it a good candidate for automation, and also the channel where it's easiest to overpromise. Here's how we actually scope it.

Start from the questions, not the technology

Before any bot logic gets written, we pull a few weeks of real conversation history and sort it. Almost every business finds the same shape: a small number of questions (hours, pricing, availability, "where's my order") make up the majority of volume, and a long tail of one-off, judgment-heavy requests make up the rest. The first group is exactly what an AI agent should answer directly. The second group is where a bot that tries to sound confident does more damage than a slow human reply.

Answer from your own material, not a guess

The agent should only ever draw from things you actually publish and stand behind — your FAQ, your price list, your policies, your calendar. If it doesn't have a real answer in that material, the right behavior is to say so and hand off, not to improvise something plausible. Getting this wrong is the single most common way these projects lose trust after launch.

Decide the handoff before the pricing

The question we ask first isn't "how do we automate this" — it's "when does this stop being something the bot should touch." Refunds, complaints, anything with money attached past a small threshold, anything where the customer sounds frustrated — those route to a human, with the full conversation history attached so nobody has to repeat themselves. Getting the handoff right matters more than getting the automation clever.

Measure it like a channel, not a toy

Once it's live, the numbers that matter are ordinary support numbers: percentage of conversations resolved without a human, time to first response, and how often a customer has to repeat a question the bot should have already known. A WhatsApp agent that's fun to demo but doesn't move those numbers isn't worth running.

None of this is specific to WhatsApp — it's the same discipline any support automation needs. WhatsApp just happens to be where the conversation is already happening, which makes it the highest-leverage place to start.