How to choose a WhatsApp chatbot: what to test before you buy
By Ignacio Damián González
A demo may start well: you ask about a service, the chatbot replies quickly, and the conversation sounds natural. To decide whether it fits your business, you need to see what happens when someone changes their request, asks about a policy you have not defined, or needs to speak to a person.
Choose a solution based on the work it can complete and how it hands over anything left unresolved. This guide offers five practical checks to compare options on the same terms before buying WhatsApp automation.
Start with the work you need to handle
Write down one specific task that takes your team time today. It might be answering questions about haircuts, collecting an appointment request, or recording a booking in your calendar.
Each task requires different information and capabilities. An up-to-date description may be enough to explain a service. Offering a time requires checking availability. Confirming an appointment requires a booking recorded in the agreed calendar.
Bring these three definitions to the demo:
- What it should handle: the service or inquiry you want to automate.
- What it needs to consult: prices, policies, a calendar, or another business information source.
- When your team should step in: exceptions, missing information, or a request to speak to a person.
This helps you ask for a demonstration of the workflow you will actually use. If your business takes bookings, our guide to automating appointments on WhatsApp explains what to prepare and how to verify a confirmation.
Which options should you compare?
Compare these options against the work you need to handle. One solution can combine several approaches:
| Option | What to evaluate |
|---|---|
| Basic WhatsApp Business tools | Include automatic greeting and away messages, plus quick replies sent by a person. Check whether they remove enough manual work. |
| Meta Business Agent | Answers using business information and supports human intervention. Available to eligible businesses in selected markets: check the features and terms enabled for your account. |
| A workflow with predefined options or steps | Whether customers can complete the task and handle changes without getting stuck. |
| A solution with AI and integrations | Which information it consults, which actions it can complete, and how your team intervenes. |
Sources checked on October 20, 2026.
Ask the provider to identify what the proposal includes, what requires an integration, and what a person still handles. The labels “chatbot” and “AI agent” do not establish those capabilities on their own.
Five checks to run during a demo
We will use a hair salon that wants to answer questions about haircuts and manage appointments. This is a fictional example for evaluating a workflow, not a customer case study.
Agree on a test calendar and fictional customer details with the provider. Anything simulated during the demo should be identified, along with the connections still needed to make it work in your business.
1. Ask about a service, its price, and a condition
Ask: “How much is a haircut, and how long does it take? Can I come without an appointment?”
Compare the answer with the information you supplied for the demo. Check whether it answers all three questions and includes the relevant conditions. If a price depends on an assessment, the response should explain that.
Then ask the provider how this information is updated and who can update it. A price change needs to reach the source the assistant consults; correcting one response may leave other conversations using outdated information.
2. Ask something you have not defined
Try a promotion that is missing from the supplied material: “Is there a discount if I book for three people?”
The response should acknowledge that someone needs to check the policy and explain what happens next. Watch for an unsupported discount, a promise of an immediate reply that nobody will handle, or a handoff that lacks the necessary context.
Agree with the provider on which topics need human review. A friendly response is useful when it also explains the next step.
3. Request an action and check the record
Request a haircut at an available time and supply the test details. When the confirmation appears, ask to see the booking in the connected calendar: service, date, time, and required details.
If the proposal only covers collecting requests, the correct result is a request pending confirmation, clearly described that way. To evaluate automatic booking, you need to verify the calendar entry.
Include a situation where the selected time is no longer available. The workflow should explain what happened and offer a verified alternative or hand the case to someone who can resolve it.
4. Change your mind during the conversation
Before confirming, write: “Actually, Thursday afternoon would be better.” Check whether the assistant understands that your preference changed, checks availability again, and retains the selected service.
If you need to change confirmed appointments, also try that workflow within the agreed scope. Check what happened to the previous booking and the new one. When a change requires human intervention, it should be clear which booking remains valid while you wait.
This check helps reveal workflows that follow a fixed script successfully but lose information when a customer corrects their request.
5. Ask to speak to a person
Write: “I would like to speak to someone at the salon.”
Ask to see what the team receives: the conversation, the service discussed, and anything still unresolved. Check how a person takes over and what happens to automatic replies while they handle the conversation.
Also review the message outside staffed hours. Customers need to know that their request is pending and when they can expect someone to follow up, based on the business's actual hours and commitments.
How to record what you observed
Use the same scenarios for each option. This table helps distinguish a demonstrated feature from an integration that is still pending:
| Situation | Expected result | Evidence to request |
|---|---|---|
| Service, price, and condition | Information consistent with the approved material. | The response and the source used to keep it current. |
| Question without a defined answer | Acknowledged limitation and a clear next step. | The conversation and where the pending request is sent. |
| Booking request | A recorded appointment or a request clearly marked as pending. | The calendar entry or the record in the system receiving the request. |
| Changed preference | Updated details and an understandable booking status. | The conversation and the final calendar state, where applicable. |
| Human assistance | A person can continue with the available context. | The team's view and how automation behaves during human handling. |
For each scenario, note whether it worked, needed correction, or requires an integration. Several correct answers to secondary questions should not outweigh a failure in an essential capability.
Compare the ongoing work and costs, too
Ask who updates services and prices, who reviews conversations, and which adjustments support includes. Clear responsibilities help you estimate the work your team will continue to do.
Also ask what data is stored, who can access conversation history, and what happens to that information when the service ends.
Compare quotes using the same scope and inquiry volume. Include setup, subscription fees, usage, and the integrations needed for the checks above. Our guide to WhatsApp automation costs breaks down these items and shows how to calculate the first month.
Choose one workflow to evaluate in a pilot
If an option handles the tasks you need, agree on a pilot with one service, one channel, and clear responsibilities. Record how you handle those inquiries today and what should improve: team time, complete requests, confirmed bookings, or unresolved cases.
During the pilot, review full conversations, corrections, and human intervention alongside the volume handled. This gives you evidence from your own operation to support the decision.
With Agents Platform, we define the first workflow, its information sources, the necessary integrations, and the cases that require a person. The public demo is informational; a pilot lets you evaluate the agreed scope for your business. The proposal includes paid setup and a monthly subscription.
Share the situations you need to handle, and we will define what a pilot should demonstrate.