Skip to content
BlogAI

AI7 min read

Reading invoices in Arabic and French with AI

We trained a model on 20,000 local invoices. Here is what worked and what did not.

Amine KHDev, Tradix Technology
00:00-01:44

The problem: five tools

Before the project, every branch kept its own Excel file. Orders came in through WhatsApp, invoices were typed twice, and nobody trusted the stock numbers. The finance team spent ten days each month reconciling what should have taken an afternoon.
  1. Every order is entered once, at the source
  2. Stock updates live across the three warehouses
  3. Invoices and accounting entries are generated automatically

Choosing one ERP

We compared three options with the team that would use them every day: the sales reps, the warehouse and finance. The winner was not the one with the most features, but the one people understood in the first demo.
We stopped arguing about which number was right. There is only one number now.
— Finance manager, Oran distributor

Migration in 3 weeks

We moved five years of data in three steps: clients and products first, then open orders, then history. Each step was tested with real users before the next one started.
The new stock dashboard, shared by the three warehouses.
The new stock dashboard, shared by the three warehouses.

Stock, sales and accounting in one place

Every order now updates stock and accounting at the same time. The warehouses see transfers live, and managers get one dashboard instead of five files.
sync_stock.py
def sync_stock(warehouse):
    for item in warehouse.items():
        erp.stock.update(item.sku, item.qty)
    return erp.stock.total()

Month-end in 2 days

The first closing on the new system took four days. The third took two. Finance now spends its time on analysis instead of copy and paste.

What we learned

Start small, train people early, and measure one number that matters. For this client, that number was the time to close the month.
Share

Written by

Amine KH

Keep reading