Coca-Cola Bottling Shqipëria

Agentic AI for Purchase Order and Fiscal Invoice Matching at Coca-Cola Bottling Shqipëria

An AI agent on Azure matches incoming Albanian fiscal invoices to purchase orders, saving at least one FTE of manual work every month.

Coca-Cola Bottling Shqipëria case study

Overview

Coca-Cola Bottling Shqipëria (CCBS) produces and distributes beverages of The Coca-Cola Company across Albania. Its finance team processes a steady stream of supplier invoices, each of which has to be matched to the right purchase order before it can be booked and paid. DataMax built an agentic AI solution on Azure that takes over this matching work and leaves the team with only the cases that genuinely need a human decision.

Challenge

Supplier invoices in Albania arrive as fiscalized e-invoices, but they rarely line up one-to-one with the purchase orders behind them. Descriptions differ, deliveries are split, and quantities or amounts can be slightly off. Working out which invoice belongs to which PO was done by hand: slow, repetitive work that took a large share of the team’s time every month and created real pressure whenever volumes peaked.

Approach

We started from how the finance team actually matches invoices: which fields they check first, which differences they accept, and when they escalate. Those rules became the backbone of the agent. The AI handles what rules alone cannot, such as reading free-text line descriptions and reconciling differently worded items, while every decision stays explainable and reviewable.

Solution

The agent runs in containers on Azure and uses models from Azure AI Foundry. For each incoming fiscal invoice it finds candidate purchase orders, compares supplier, line items, quantities, amounts and VAT, and either confirms the match or flags the specific discrepancy, for example a quantity that differs from what was ordered. Match results and their reasoning are stored in Azure Table Storage, so the finance team reviews a short list of exceptions instead of searching through every invoice.

Results

Capacity freed
At least 1 FTE of manual work saved every month
Exceptions, not searches
The team reviews flagged discrepancies instead of matching every invoice by hand
Less pressure
A tedious, stressful task taken off the finance team’s plate

Ready to accelerate your AI journey?

Let's talk about your data and AI challenges. We'll help you build the right strategy and execute with speed.

Get in Touch