Summary: a WhatsApp chatbot is a flow: trigger, menu, questions, conditions and actions. With a visual builder you assemble that flow by dragging blocks, connect the WhatsApp account, test it in a simulated environment and publish. The most important decision is when the conversation goes to a person.
A large share of the messages that reach a company’s WhatsApp are predictable: opening hours, order status, a copy of an invoice, booking an appointment, talking to sales. A WhatsApp chatbot handles that volume with no queue and no waiting, and passes to your team only what needs a person. This guide shows how to build one from scratch in a visual flow builder, without writing code.
What a flow chatbot is
A flow chatbot is a diagram that runs: each block is a step (send a message, ask a question, query a system, transfer) and each arrow is a transition, with or without a condition. When the customer sends a message, the flow starts at the start block and follows the arrows until it ends or hands the conversation to an agent.
That is different from an AI replying freely. The flow is deterministic: the same answer always leads to the same place. AI can come in at specific points, as we will see, but the backbone of the chatbot is the flow.
Step 1: define what the chatbot solves and what it doesn’t
List the five most frequent reasons for contact. For each one, decide: does the chatbot solve it alone, solve it with a lookup in another system, or only collect the details and transfer? This mapping avoids the most common mistake, which is trying to automate everything and frustrating the people who needed a person.
- Solves alone: hours, address, payment methods, return policy.
- Solves with a lookup: order status, balance, invoice copy, available time slots.
- Collects and transfers: complaints, negotiations, anything off script.
Step 2: sketch the flow before configuring
On paper or in the editor itself, draw the conversation as a diagram: welcome, main menu, one branch per option and, in each branch, the questions, the lookups and the ending or the transfer. A good first flow has few menu levels and a clear path to “talk to an agent” from any point.
Step 3: build the chatbot in the visual builder
In DeskFlow, VipDesk’s flow builder, blocks sit in a palette and are dragged onto the canvas. The ones you will use first:
- Start: the trigger. The flow starts when a message arrives on the WhatsApp account bound to it.
- Text message: the greeting. Use variables such as the contact’s name to personalize it.
- Menu (router): numbered options. Each option becomes an output of the block; the error output covers unexpected replies.
- Buttons and lists: alternatives to the typed menu, within WhatsApp’s limits (buttons allow at most three options).
- Capture data: asks and validates the answer (ID number, order number, email) and stores it in a conversation variable.
- Condition (if/else): picks the path from a variable, such as “registered customer” or “amount greater than”.
- Transfer: hands the conversation to a department, distributing it among available agents.
- End: finishes the flow and clears the conversation context.
Conditions can also live on the arrows themselves (equals, contains, positive answer, digits only, regular expression), which keeps the diagram lean.
Step 4: connect the chatbot to your systems
This is where the chatbot stops being a menu and starts solving things. The HTTP request block calls your system’s API (ERP, e-commerce, calendar, in-house system) with the details the customer gave, and the response becomes a variable for the next messages:
“Your order ${{order.number}} is ${{order.status}}
and should arrive on ${{order.eta}}.”
Store credentials as secrets (never in the flow text), handle the call’s error output (transfer to an agent instead of leaving the customer without a reply) and use the block’s one-off test to check the API response before publishing.
Step 5: use AI only where it helps
Two uses are worth it from day one:
- Understanding free text at the menu. If the customer writes “my delivery is late” instead of typing “2”, the AI picks the matching option. Whoever types the number goes straight through, with no AI.
- Answering questions from your knowledge base. An AI agent with access to your documents answers open questions and transfers when it doesn’t know.
Keep the AI fenced by the flow: deterministic validations before it, controlled actions after it and always a path to your team. The article AI agent for WhatsApp: what it does and when to use it goes into the criteria.
Step 6: decide how the conversation reaches your team
The best chatbot is the one that knows when to stop. Define:
- Which menu options transfer directly, and to which department.
- What happens outside business hours: a condition on the current time says so, logs the request and, if needed, flags the ticket for the next morning.
- How many attempts the customer gets before the conversation goes to a person.
In a customer service system such as VipDesk, the agent receives the conversation in the same queue as WhatsApp with multiple agents, with the history of what the chatbot already asked. Nobody asks for the ID number twice.
Step 7: test, publish and activate
- Test in the playground, with simulated sessions, before any customer sees the flow.
- Publish when you are satisfied. Changes stay in a draft until then, and the version history lets you roll back.
- Bind the WhatsApp account to the flow and activate it. An account belongs to one active flow at a time.
- Watch the execution logs during the first days and adjust error messages and transfer points.
About the WhatsApp connection
The chatbot runs on the WhatsApp account connected to the platform. For business-critical operations, VipDesk’s Terms of Use recommend the official WhatsApp Business API, which is also what allows sending approved templates in scheduled flows such as reminders and billing notices. A QR-code connection exists, but carries a risk of the number being blocked.
Common mistakes
- Too many menu levels. Three taps to reach “talk to an agent” is already too many.
- No error output. Every system lookup and every menu needs a path for unexpected replies.
- Automating complaints. An upset customer wants a person. Collect the minimum and transfer.
- Forgetting what happens after the transfer. The agent needs to see what the chatbot already did; otherwise the customer repeats everything.
In VipDesk, the chatbot is built in DeskFlow, with menus, data capture, API integration, AI agents and hand-off to your team on the same WhatsApp number. In the demo, we build the flow around your scenario.