Case studyRAK Ceramics

Invoice Booking

rakceramics.com
AzurePythonOffice (outlook integration)

The challenge

RAK Ceramics booked supplier invoices into SAP by hand. AP staff keyed line items, ran compliance checks, and 3-way matched invoice against PO and goods receipt, across local vendor, imported material, and freight documents, each with its own layout and its own rules. Automating it was the obvious move, but accuracy was the blocker: an extraction pipeline sitting in the 60% is fine for a demo and unusable in production, where every miss is a payment error someone has to chase down.

The outcome

Designed and shipped an AI workflow that automates invoice validation, compliance checks, and 3-way matching, now running in production. Accuracy moved from just over 60% to 98.9%, not by swapping in a bigger model but by diagnosing the recurring error patterns behind the failures and engineering targeted filters against them. Results were validated on a 2,000-invoice set deliberately weighted with 400+ negative samples (~20% of cases), so the number reflects behaviour on the hard documents, not just the clean ones.

Have a programme like this one?

Tell us where it is stuck. We will come back with the shortest credible path from where you are to a system in production.

Start a conversationsupport@multiversemena.com