All questions

How to automate WhatsApp customer service with AI without losing customers

Automating WhatsApp with AI saves time, but a poorly designed agent can also drive customers away: answers that miss the point, repeated questions, or a conversation that never reaches a person. The goal is not to replace your team. It is to make sure no message goes unanswered and every lead lands in the right place.

In my projects I follow a few rules that make the difference between a helpful agent and an annoying one.

Start with the process, not the bot

Before writing any code, I map the current process: where messages come in, which questions repeat, who answers, and where follow-up gets lost. The same pattern shows up almost every time: after-hours inquiries go cold before anyone sees them, and there is no consistent way to decide who to call first.

With that map we define what the agent does and what it does not do. For example: greet, understand what the person needs, ask for the key details and qualify the contact. Anything that needs human judgment, like a negotiation or a sensitive case, stays with the team.

The rules that keep customers

The first is a way out to a person. The agent has to recognize when it cannot answer or when the customer asks for someone, and hand over the conversation with its full context, so nobody has to repeat their story.

The second is to ask only for what you need. Every extra question is a chance for the person to give up. If a detail can be collected later, it waits until later.

The third is that the information does not stay in the chat. A qualified lead that never reaches the CRM is a lost lead. So the agent creates or updates the contact, the stage and the follow-up task in the CRM, with no one copying data by hand.

The fourth is to measure and adjust. The first weeks show which questions nobody planned for and where the agent gets confused. That is what you use to tune the rules and the messages.

A real case: G&M Capital

For G&M Capital, a credit institution, I built a lead capture and evaluation system with a WhatsApp AI agent. The agent answers interested people and qualifies them, and qualified leads land in the Kommo CRM with no manual data entry.

The Kommo connection runs through my own MCP server, kommo-mcp, which I published as open source. It lets the agent create and update leads, pipelines and tasks through its own tools. The system also includes a credit calculator and a PIN-protected portal for the team, and the workflows that tie the pieces together run on n8n.

What matters in this case is not the technology but the effect on daily work: the team works on leads that already arrive organized in the CRM, instead of chasing scattered messages.

Where to start

If you answer WhatsApp by hand today, spend one week writing down the questions that repeat most and the details you need from each customer. With that list you can design a first, narrow agent, test it on real conversations and grow from there, instead of trying to automate everything on day one.

Do you have a similar case?

Tell me how you work today and what you want to solve. Native Spanish; English (B2).

Automate WhatsApp Support with AI Without Losing Customers