Skip to content
MANASA Studios
All work

From 40 hours a week to 4 a month.

Invoice processing, automated end to end for a Dubai business that handles about 1,000 vendor invoices a month: from the moment an invoice arrives to the finished record in their tracking platform.

Client
A Dubai business
Volume
About 1,000 vendor invoices a month
Services
AI integration · Automation · Web app · Machine learning
Built with
Google Cloud Document AI · Tracking platform API · Custom ML model
Al Noor Supplies LLCTAX INVOICEInvoice no.INV-20418Date12 MarTotal dueAED 4,725.00ExtractedVendorAl Noor Supplies LLCInvoice no.INV-20418Date12 MarTotalAED 4,725.00CategoryFacilitiesAssigned by modelSent to tracking platform

The problem

Every invoice was handled by hand. Paper copies were scanned, then each field was read off the page and typed into a form on the company's tracking platform.

At around 1,000 invoices a month, that came to about 40 hours of data entry every week: a full-time job spent copying numbers from one place to another, with a chance of a typo on every field.

What we built

Five parts, each feeding the next, on top of the tools the team already used.

  1. PDFs at the source

    Vendors were asked to send PDF invoices instead of paper, so nothing needs scanning and every document arrives ready to be read by software.

  2. Extraction tuned to their invoices

    We fine-tuned Google Cloud Document AI on the company's own invoices so it pulls out exactly the fields the team needs, whichever vendor sent them.

  3. Straight into the tracking platform

    Extracted data goes to the company's existing tracking platform through its API, filling the same records the team used to type by hand. No new system to learn.

  4. A web app for the team

    The team works from one web app: every invoice tracked from arrival to entry, a full history to search, and analytics across all of it.

  5. Automatic categorisation

    A custom machine-learning model analyses each invoice and assigns it a category, so nobody has to sort them by hand.

The result

  • Time on invoices

    Before About 40 hours a week

    After About 4 hours a month

  • How data gets in

    Before Scanned, then typed field by field

    After Read from the PDF and sent by API

  • Sorting

    Before By hand

    After Categorised by the model

And it keeps improving. Extraction has kept getting more accurate the longer the system has run.

Have a process like this? Let's automate it.

Tell us what your team does by hand every week. The first call is free and takes 15 minutes.

Book a free 15-min call