AI Agent on company mailboxes: quotes, info and support for Kerberos Energy

The context

Kerberos Energy needed to manage incoming requests on company sales and support mailboxes in a structured, scalable way. The goal: consistent responses, no request lost, and a clear flow between quote, order, information and support requests.

The agent that listens to the mailboxes

We developed an AI agent that monitors company sales and support mailboxes in real time. Each incoming email is not processed at random: the system first checks whether it is a duplicate. Only if the request is not a duplicate is a conversation history initialised in the database (built with Supabase), so that if the customer continues the exchange, the conversation remains tracked.

When there is a back-and-forth within the same mailbox (customer replies, company responds, etc.), the system tracks it and uses it to keep the exchange as contextually coherent as possible. If the email is new, a new conversation is created on first contact; if it is a reply to an existing thread, the record is updated. In addition, spam is filtered: only legitimate requests are processed by the agent.

Enrichment and contact management

The agent has a tool structured around 5 main categories for enrichment of customers and contacts. Kerberos Energy can query the agent and ask it to create useful contacts for the company, keeping a reference database aligned with commercial and support needs.

The core flow: VAT number and request type

The system mainly handles support for quote, order, info and support requests. When an email arrives, the first step is to look up the sender's VAT number. If the VAT number is found, the agent continues with the operations; if it is not found, the agent politely asks the sender to provide their VAT number. This step is necessary because, via subsequent APIs, the system retrieves the VAT subject and can correctly apply company rates and procedures.

Quote request

If the request is a quote, once the party is identified via VAT number, the agent creates a quote for the intervention based on the company database (rates, types of intervention, etc.), ensuring consistency and traceability.

Information request

If the request is for information, the agent responds by querying a RAG (Retrieval-Augmented Generation) on company documentation and knowledge, providing precise answers aligned with policies and products.

Support request

If the request is for support, the system correctly identifies urgency and actual support request; once these fields are identified, a structured ticket is created and the support team is informed of the ticket creation, so they can act with all details already organised.

Closing the loop

Each call is unique and the agent receives feedback on whether execution succeeded or not. When one of these processes (quote, info response or ticket creation) is completed, the system sends a reply email to the customer, closing the loop in a professional and tracked way.