The wizard is orchestrated on n8n (self-hosted, on client infrastructure) with a RAG architecture based on the company's knowledge base:
For an international B2B e-commerce platform with clients in multiple countries and languages, we set up a support agent based on their knowledge base and CRM. The operational result: 78% of queries are resolved without team intervention, with an average response time of 38 seconds, allowing the human team to focus on sales conversations and issues requiring expert judgment. The patterns identified in the queries also helped improve the product pages that generated the most questions.
It says so and passes the conversation to someone on your team, with all the context of what was discussed. It's configured to respond only with information from your documentation: if it doesn't have it, it doesn't make it up.
The assistant only responds based on your own documentation: catalogs, terms and conditions, policies, and FAQs. Before being put in front of real customers, it's tested with your team and fine-tuned until the responses are accurate and sound like your company's.
The assistant presents itself as what it is: it doesn't deceive anyone. What your customers value is the right answer in seconds, at any time, in their language. And when the situation calls for it, there's a real person on the other side of the conversation.
The system is built on infrastructure that you control, and your customer data is not used to train any model. We define with you what information the assistant can see and what is excluded.
WhatsApp Business, website chat, and email—separately or in combination. The conversation is recorded in your CRM regardless of the channel it originates from.
Between two and five weeks, depending on the channels and available documentation. We start with a single channel and the most frequent type of inquiry, measure the results, and expand from there.
Free assessment: we identify processes, flows, and custom AI opportunities for your operation.
