Last week OpenAI published a case study on Ringg, a company that runs AI customer service agents over phone calls, chat, WhatsApp and the web. The headline number: agents resolve up to 65% of routine customer requests without a human. If you run a spa, a travel agency or a rental business, and your staff spend half the day answering the same WhatsApp questions, that number is tempting. I want to look at what sits behind it, and whether the idea works at a much smaller scale.

One note before I start. I have not built a WhatsApp or voice AI agent for a client. I have built the parts these agents depend on: a ticket system that emails the customer when a ticket arrives and every time its status changes, and a reservation and scheduling system for a spa. So this is an analysis, not a case study, but it comes from someone who has wired up the plumbing.

What the numbers actually say

According to the OpenAI case study, Ringg's agents handle more than 7 million connected calls a month. A few customer results stand out:

Policybazaar, a large Indian insurance platform, has 67% of calls handled without a human, and average response time fell from 8–12 minutes to under 60 seconds.
Practo, a healthcare booking platform, reports operating costs 70% lower than its previous human-led workflow, with more than 1,000 appointments booked by the agents each day.

Two things to keep in mind. First, this is a vendor story published on a vendor's website. The numbers may be accurate, but they were chosen to impress. Second, these are large consumer companies with enormous volume. At millions of calls, even a small improvement pays for a lot of engineering. A business that receives forty messages a day has very different math.

What AI customer service agents really do

The interesting part of the case study is not the conversation. It is the list of things the agent does to finish a request: check a policy, pull up a customer record, book an appointment, update the CRM (the system where you keep customer details and history), or hand the conversation to a person with the context kept intact.

In other words, the chatbot is the easy part. The hard part is the systems it talks to. An agent that can only chat can answer "what time do you open?" It cannot book a 3pm slot, because there is no slot to book unless your schedule lives somewhere a computer can read.

Automated ticket emails taught me the same lesson on a smaller scale: an automated message is only as correct as the record behind it. If the status in the system is wrong, the customer gets a confident, wrong update. An AI agent makes that problem bigger, not smaller.

The small-business version: one spa, one WhatsApp number

Imagine a day spa in Seminyak with one WhatsApp number. A normal day brings messages like these:

"Are you open today?"
"How much is the 90-minute massage?"
"Can I book 3pm for two people?"
"Do you have a female therapist available?"
"I was charged twice for my booking yesterday."

Here is how an agent would handle each one, honestly.

Hours and prices

An agent can answer these well, if you give it one clean, current price list. If your prices live in three places (an Instagram highlight, a printed menu and the front desk's memory), it will quote the wrong one with complete confidence.

Bookings

This only works if availability sits in a real system: which therapists work which shifts, which rooms are free. If bookings are written in a notebook, the agent cannot know that 3pm is already taken. This is the step most small businesses skip, and it is the one that matters most. When I built the reservation and scheduling system for Sakura Spa, the booking calendar was the foundation everything else sat on.

Special requests and complaints

These go to a person. The value an agent adds here is a short summary, so your staff do not have to scroll through thirty messages to understand the problem.

So for this spa, a realistic agent answers hours and prices, handles some bookings, and routes the rest. Whether that adds up to 65% or 20% depends entirely on the mix of messages you receive. The good news is that you can measure that mix yourself.

What it costs, beyond the AI bill

The AI part keeps getting cheaper. Ringg says moving some of its workloads to a newer OpenAI model cut model costs by about 90%. What does not get cheaper is everything around the model: getting your prices and availability into one place, connecting the agent to your booking system, testing it on real conversations, and having someone check its work.

Ringg tests new setups on past conversations first, then sends only a small share of real traffic to them before rolling out further. A small business should copy that habit. Start with the agent drafting replies that a staff member approves with one tap. Once the drafts are consistently right, let it send the simple ones on its own. I wrote more about weighing these costs in my guide to calculating AI ROI.

When you should not do this yet

Most of your messages are complaints, negotiations or custom requests. An agent will frustrate customers who need a person.
Your prices and availability are not in one place. Fix that first. It saves time on its own, with or without AI.
Your volume is low. If you get ten messages a day, the saved quick replies in WhatsApp Business may be all you need.

One test to run this week

Before you talk to any vendor, find out what your customers actually ask.

Export two weeks of chats from your business WhatsApp. WhatsApp lets you export a chat as a text file.
Go through the incoming messages and tag each one: price, hours, booking, complaint or other.
Count them.

If most messages are price, hours or booking, you are a real candidate. Your next step is not the AI. It is making sure your price list and availability live in one system a computer can read. Tools like n8n can then connect WhatsApp, your calendar and your database, as I explained in my post on n8n.

If most messages are complaints or "other", an agent is the wrong tool for now. The better investment is fixing whatever is causing those messages in the first place.

If you have done the count and want help deciding what to automate first, or want to get your bookings and prices into a system an agent could use, get in touch. I am happy to look at your numbers with you.