WhatsApp-based sales CRM with RFV segmentation

Challenge

The sales team at an apparel manufacturer works in the field and doesn’t open the ERP. Account portfolio, target and history information stayed locked in screens designed for the office, and visit reports arrived as a stray message, when they arrived at all. There was also no objective criterion for prioritising who to visit: the portfolio was worked through habit.

The company wanted to measure customer and employee satisfaction, but email surveys had a response rate too low to serve as an indicator.

Solution

I built the backend for the CRM app: a 360º customer account view, with orders, receivables, returns, contact history and credit limit in the same response; targets and rep rankings; and an in-house RFV segmentation engine (recency, frequency and value) that classifies the portfolio into quadrants and points, with criteria, to who needs contact now.

Visit logging moved to WhatsApp. The rep sends text or audio; the audio goes to S3, is transcribed by AWS Transcribe, and the text enters the customer history already linked to the visit. On the same infrastructure I built the satisfaction survey, for customers and employees, with a resend state machine: whoever hasn’t answered gets it again, up to a limit, and whoever answered drops out of the queue.

The messaging layer started on a commercial WhatsApp gateway and gained a second gateway on top of Evolution API in 2026, with a connection monitor and QR-code reading from the ERP’s own screen. The two coexist, because the sending layer is abstract and the gateway is configuration.

Outcome

The sales team started operating through the channel it already used all day, with no new app to learn, and portfolio prioritisation gained a quantitative criterion. The satisfaction survey reached a response volume you can actually use as a management indicator. The company also stopped depending on a single WhatsApp provider for an operation that had become critical.

What this case shows

You can build an app backend on top of a legacy ERP without breaking the legacy, and transcription through a managed service fits inside a real business flow. I kept the RFV analytical modelling inside the ERP, rather than exporting it to BI. The WhatsApp gateway abstraction was a supplier-risk decision.