Monday morning, the sales team's inbox holds about fifteen purchase orders. Some are clean PDFs, others are crooked scans, and two arrive as WhatsApp photos taken on the bonnet of a car. An assistant opens them one by one, copies the customer, part numbers, quantities and prices into the ERP, then checks that everything matches the quote that was sent.
At the same time, accounting does the same thing with supplier invoices and delivery notes. Nobody really complains: it is the job. But this is also where wrong part numbers, swapped quantities and orders that wait two days to be entered come from. It is one of the problems we cover on our AI and automation for industry page.
Why this typing costs more than it seems
We have not found a reliable figure for the time Tunisian companies spend re-keying documents, and we will not invent one. The cost shows up in the process instead:
- Delay. An order received on Friday does not exist in the ERP until someone enters it. Stock, production and delivery wait.
- Errors. Every copy is a chance to get a part number or a quantity wrong, and the mistake is often found at delivery.
- Dependence on one person. When the assistant who "knows the customers" is away, the pile grows.
Tunisian manufacturers know this. In the EY and Novation City Industry 4.0 barometer (2022, 74 companies), purchasing and procurement rank among the three priority functions to digitise, after production and after marketing and sales. It is a small survey, but it points the same way as what we see in the field.
There is also a regulatory context. With El Fatoora electronic invoicing, run by TTN, backed by penalties since the 2025 finance law and extended to service providers by the 2026 one, invoices must be produced in a structured format. Clean data upstream, without approximate re-keying, makes that compliance easier. Our guide to integrating the TTN / El Fatoora API with your ERP covers the technical side.
What an extraction tool does, step by step
The idea is not to remove the person, but to take the typing away and leave them in control.
- One intake. Documents arrive as they do today (email, scan, WhatsApp) and are gathered in a single queue.
- Reading the document. A document processing model reads the PDF, scan or photo and extracts the useful fields: customer or supplier, document number, date, part numbers, descriptions, quantities, unit prices, totals. Recent vision-language models also read tables and Arabic; we explain how in our article on Document AI.
- Consistency checks. The tool compares what it read with what you already know: does the part number exist in your item master? Does the price match the quote or price list? Is the total right? Does the delivery note match the order?
- Human validation of doubtful cases. Confident documents are prepared for the ERP. The others are shown to a person, with the questionable field highlighted. They correct it in seconds instead of typing everything.
- Sending to the ERP. Validated data goes to your ERP, through its API when there is one, or through an import file otherwise.
What the tool does not do
- It does not replace your ERP. It connects to it. We explain this choice in "The ERP trap: integration, not replacement".
- It does not validate on its own. A person keeps control of what enters the system, especially at the start.
- It does not read everything perfectly. A blurred photo or a handwritten document may resist. We measure the read rate on your real documents before setting a target.
- It does not solve El Fatoora compliance by itself. It improves data quality; transmission to TTN remains a separate integration project.
Start small
There is no need to process every document in the company from day one. A good first scope:
- one document type, for example customer purchase orders;
- one team, the one that types them today;
- one simple indicator set at the start: number of documents re-keyed by hand per week, or the delay between receiving an order and entering it.
After a few weeks of use, look at the indicator. If it moves, extend to delivery notes, then supplier invoices. If it does not, you find out quickly and at small scale.
The same assistant can then serve other uses in this cluster, such as preparing quotes from your catalogues.
If your team spends its mornings copying documents someone else already typed, let's talk.
Do you have this problem? Describe it in a few lines: what happens today, who is involved, what it costs you in time. We get back to you quickly, usually within 48 working hours, for a short discussion: to understand your project and see whether we are the right partner.
Describe your problem · WhatsApp · All our solutions for industry
Sources
- EY / Novation City Industry 4.0 barometer, 2022, Tunisia, 74 companies (purchasing and procurement among the priority functions to digitise), summarised by Managers: managers.tn
- El Fatoora electronic invoicing: noqta.tn guide, TTN API integration