When a WhatsApp chatbot is the wrong tool (and when it is more than enough)
The right question is not whether it works
It is which query it works for, because the answer changes completely from one to the next.
When a WhatsApp chatbot does not work has a fairly concrete answer: it does not work when the answer is not written down anywhere, when something has to be decided, or when the person on the other side is angry.
And it is more than enough in the opposite case, which happens to be the most frequent one: the question you have already answered two hundred times.
The uncomfortable self-service number
Worth having to hand before deciding how much to automate.
A Gartner survey of 5,728 customers found that only 14% of customer service issues are fully resolved in self-service. Not 80%, not 60%. Fourteen.
That number is the ceiling on "let it all resolve itself", and it explains why so many rollouts leave people worse off than before: the front door got automated without anybody building the exit.
The four queries where it loses you the customer
| Situation | Why it fails |
|---|---|
| An angry complaint | The person wants to be heard, not informed |
| Negotiating price or terms | It is a commercial decision, not a fact |
| A case with no written answer | With no source, it either invents or frustrates |
| An errand that depends on a third party | Coordinating with a supplier is not answering |
The pattern repeats: all four need judgement or physical action. None of them is solved by having more information available.
The fifth, and the most expensive: not letting people out
An AI with no emergency exit annoys more than it helps.
Gartner measured that 87% of 3,566 customers say a company using generative AI in customer service has to provide access to a human agent. It does not say don't use it: it says there has to be a door.
The damage of not having one does not show up in the conversation that fails. It shows up in the customer who calls on the phone next time, or does not come back.
Where it is more than enough
Everything already written down, that somebody today copies and pastes ten times a day.
- Opening hours, addresses, payment methods, delivery times.
- The status of an order that already exists in your system.
- Prices and availability, when they come from your catalogue and not from somebody's memory.
- Requirements and policies: what paperwork you ask for, how returns work, what the warranty covers.
- The same question in different words, which is where the difference shows most.
These five are not the interesting queries in your business. They are the ones that eat your morning.
The rule for telling them apart
One question, answered fast: is the answer already written in a document of yours?
If it is, it can be automated and it will go well, because the answer gets looked up in your text instead of generated. If it is not, you have two honest options: write it, or send it to a person.
What is not an option is expecting the model to fill in the gap. That is where plausible, false answers come from.
How the limit gets set, in practice
You do not have to choose between everything and nothing.
- Define the forbidden topics. Special pricing, complaints, cancellations: whatever you do not want touched.
- Tag what goes straight to a person, without passing through any automated reply.
- Start gradually, answering only the conversations you mark.
- Say so when you hand off, with the reason written into the conversation.
What a cheap chatbot will not tell you
That the problem is almost never the model. It is that nobody gave it anything of yours to answer with.
A service without your documents answers what anybody in your sector would answer. It sounds fine and it is useless, because your returns policy, your delivery zones and your volume discount are not on the internet.
Key takeaways
- A WhatsApp chatbot does not work for angry complaints, negotiations, cases with no written answer, or errands that depend on a third party.
- It is more than enough for opening hours, catalogue prices, order status, policies and repeated questions.
- Gartner measured that only 14% of issues are fully resolved in self-service.
- 87% of customers demand being able to reach a person when AI is involved.
- The practical rule is whether the answer already exists written in a document of yours.
- Limits are set by topic, by tag and by adoption mode, not with a single switch.
Frequently asked questions
- Is automating all of customer service a good idea?
- No, and the numbers back it up: Gartner measured that only 14% of issues are fully resolved in self-service. What is worth automating is the repetitive part that has a written answer, leaving the rest to your team.
- How do I stop it answering certain topics?
- By setting handoff rules per topic. Whatever you mark as out of scope is not answered automatically, it goes to a person with the reason written down.
- What happens when a customer asks to talk to somebody?
- It should go to a person, without an argument. 87% of the customers Gartner surveyed say a company using AI in customer service has to provide access to a human.
- Why did my previous chatbot answer nonsense?
- Most likely it did not have your documents. With no source of its own, the model fills in with what it knows about the sector, which is not what it knows about your business.
- Is it worth it for a business with few queries a day?
- It depends on how many of them repeat. If ten of the fifteen daily queries are the same question, the saving is real even if the volume is small.
Related posts
Get twice the output from your team without asking more of them
It answers WhatsApp, Instagram and web. It learns your business, takes the orders, and hands off to a person when it matters. You see every action noted in the chat.
Free Pro trial. No credit card required.