Digitising delivery notes and contracts without retyping them
When a small company decides to "digitise its paperwork", it almost always starts with invoices. That is logical: they are what the accountant handles and what hurts if they go missing. But the papers that quietly lose the most money are different ones: delivery notes and contracts.
The delivery note is the document that says what was actually delivered. The contract is the one that says what was agreed. If both live in a folder nobody opens, the company is invoicing blind.
The hole nobody sees
An example of how it happens. A building-materials yard serves construction sites every day. The driver leaves the signed delivery note with the customer, the crumpled copy goes back in the van and, with luck, ends up in a tray at the office on Friday. At the end of the month, someone cross-checks delivery notes against orders to raise the invoices.
What happens in practice: 40 bags get invoiced when 44 were delivered, and nobody claims the difference because nobody knows. A delivery note gets lost and that delivery is never invoiced. The customer says a different price had been agreed, and nobody can find the contract to check. And when a discrepancy surfaces three months later, looking into it costs half an afternoon.
It is not a problem of careless people. It is a problem of information that exists on paper but does not exist in any system.
What digitising them actually means
Scanning is not digitising. A PDF in a folder is the same piece of paper, only lit up: you still cannot search it, compare it or set an alert on it.
Digitising properly means turning the document into data. From a delivery note: supplier or customer, date, number, references, quantities, who signed. From a contract: parties, start and end dates, agreed prices, renewal terms, notice periods.
Once you have that in fields, things become possible that were not before:
- Automatically matching order ↔ delivery note ↔ invoice, with an alert only when there is a difference.
- Searching by content: "all deliveries to this customer in May", in seconds.
- Expiry alerts: a contract that renews itself in 60 days that nobody had looked at.
That third point is the one that surprises people most. Many maintenance, equipment-hire or service contracts renew silently year after year because the notice period passed without anyone seeing it.
One piece of advice before starting
If you do it, do not start with everything. Choose one document type and one circuit. For example: only incoming supplier delivery notes, for one month.
Two reasons. The first is that you will learn things about your own process you did not know — such as the fact that there are three different delivery-note layouts because three suppliers each do it their own way. The second is that if the system fails somewhere small, it takes nothing down with it.
And one essential check: in the first weeks, review a sample by hand. Ten random documents a week. If the fields match, expand. If not, fix it before you have three months of badly recorded data. No extraction system gets 100% right from day one, and what you need to know is what it gets wrong and how often.
The point you cannot skip: privacy
Delivery notes carry customer data and addresses. Contracts carry financial terms, names and, often, personal data of signatories. That is not material to upload to any free tool on the internet just to try it out.
Three concrete questions before signing up to anything:
- Where is the document processed and stored? It must be inside the EU, and you must be able to get that in writing.
- Does the vendor use your documents to train its models? The correct answer is no. If the answer is "in anonymised form", ask exactly what that means.
- Who has access within your company? A contract with financial terms should not be openable by everyone. Access is assigned by role.
If a vendor gets uncomfortable with these three questions, you already have the answer you needed.
An example of how it can be set up
ParsIA, the document-extraction tool I built, installs on your own server or cloud, extracts the fields from delivery notes and contracts, and keeps human review on anything involving amounts or deadlines. The design decisions behind it are written up on the project page.
This article is general information, not legal advice for a specific case.
About this article
iaetica.cat is a space for writing in the open: what I am learning about artificial intelligence, its ethics and its law.
— Laura Graells