Takes orders over WhatsApp, finds the right items in a catalog of thousands of products and leaves the quote loaded in the ERP.
How the data flows
WhatsApp order
Vector search
Quote in the ERP
CRM pipeline
3branches on the same catalog
Client
Electropunto
Industry
Electrical supplies distributor
Where
Corrientes, Paso de los Libres and Posadas, Argentina
Integrations
WhatsApp, Kommo CRM, Cuántico ERP
The context
An electrical supplies distributor with three branches that work on the same ERP: shared catalog, product codes, prices and quotes. The CRM tracks every conversation and how far along it is in the sale.
The problem
The ERP opened an API with thousands of items, but that sales data could not be used as is for an AI to look up quickly and cheaply. On top of that, the system had to understand counter slang and synonyms, voice notes, photos and job site lists, without ever inventing a product or a price.
What we built
01We copied and synced the catalog from the ERP API, cleaned up its data and turned it into vectors.
02On top of that, a hybrid search with embeddings, sales rules and strict filters; an agent coordinated with n8n and MCP tools interprets each order.
03It started with text and later added voice notes and images.
04The CRM was connected so the agent moves the pipeline based on what happens in the conversation. Every real error was turned into a test to prevent regressions.
What changed
The customer sends an order over WhatsApp and gets a quote built with official products.
The quote is loaded in the ERP and the deal moves on its own from “First contact” to “Quoted”.
When the customer arrives at the branch, the follow up is already done: the employee only reviews and charges.
Stack
OpenAIn8nTypeScriptMCPPostgreSQL + pgvectorDocker
Connects with
WhatsAppKommo CRMCuántico ERP
Design something similar for your operation
No system is copied and pasted: what works here came from how this business runs. Tell us about yours and we will see what fits.