AI and ERP: automating goods receipt from delivery notes, PDFs, CSV and Excel files

After exploring how artificial intelligence can automate the receipt and entry of sales orders, we continue our series dedicated to the practical use of AI in ERP processes.

In this second article, we address another everyday activity for many companies: receiving goods and recording documents received from suppliers.

Once again, we are not talking about future features or simple experiments.

The features described in this series are already available and can be used with GERP.

The goal is simple: to transform documents received in different formats — PDF, CSV, Excel and other structured or unstructured documents — into information that can be used directly by the ERP, reducing the manual work required to record incoming materials.

From supplier documents to goods receipt

When goods arrive at a company, they are normally accompanied by a document describing what has been delivered.

This may be a delivery note in PDF format received by email, a paper document, a CSV file, an Excel spreadsheet or another format generated by the supplier's information system.

The problem is that each supplier may use a different format.

The position of the information, item descriptions, codes, units of measure and even the way quantities, batches or order references are indicated can vary.

Traditionally, an operator therefore needs to read the document and transfer the information into the ERP.

Artificial intelligence makes it possible to approach the problem differently.

PDF, CSV and Excel: different formats, same process

A traditional import system normally requires a predefined format.

To import a CSV file, we need to know which column contains the item code, which contains the quantity, which contains the batch number, and so on.

With PDF documents, the situation becomes even more complex because the document is generally designed to be read by a person rather than by software.

AI makes it possible to add a layer of interpretation.

GERP can use the received document to identify information such as:

• supplier;


• document number and date;


• related purchase order or orders;


• item codes;


• descriptions;


• delivered quantities;


• units of measure;


• batches;


• any serial numbers;


• supplier references;


• notes included in the document.

The important point is that not all suppliers need to use the same layout.

The goal is not to teach the system where specific information is located in a particular document template, but to enable it to understand the meaning of the information contained in the document.

Recognising supplier items

As we have seen with sales orders, a common problem also arises in the purchasing process: the code used by the supplier may differ from the code used internally by the company.

On the delivery note, we might find:

MAT-4587 — Transparent PET film 23 µm

while the same material might be identified by a completely different code within the ERP.

GERP can use the information already available in the ERP — items, supplier codes, descriptions, purchase orders and supply history — to identify the most likely match.

AI therefore does more than simply read the document.

It interprets the document using the ERP context.

And it is precisely this context that makes automation much more valuable than simply extracting text from a PDF.

Comparison with the purchase order

The next step is even more important.

If the received goods relate to a purchase order already recorded in GERP, the system can compare what was ordered with what the supplier declares has been delivered.

We can therefore move from:

read the delivery note → manually enter the goods receipt

to:

read the delivery note → identify the order → compare the data → prepare the goods receipt → flag discrepancies.

Suppose, for example, that the purchase order specifies 5,000 kg of material, while the supplier's document indicates 4,800 kg.

What matters to the operator is not having to enter "4,800".

It is knowing that 200 kg less than the ordered quantity has been delivered.

Automation therefore makes it possible to shift attention from data entry to exception management.

Purchase order, document and goods received

The supplier's document, however, represents only one part of the process.

The system can know:

what we ordered → what the supplier declares has been shipped → what we actually received.

These three elements can be compared.

GERP can therefore automatically highlight situations such as quantities that differ from the order, unexpected items, partial deliveries, missing materials, different units of measure or references that cannot be matched.

The operator can focus on anomalies instead of manually checking every single line.

AI prepares the inventory movement, the operator verifies it

In goods receipt processes, it is also important to distinguish between automation and the absence of control.

An effective model can follow the same logic already described for sales orders:

read → understand → compare → propose → verify → record

GERP can prepare the goods receipt document or inventory movement based on the information recognised.

The operator can review the result, correct any anomalies and confirm the operation.

For simpler and more repetitive cases, the level of automation can be greater; where uncertainties or inconsistencies exist, the system can explicitly request user intervention.

Managing exceptions instead of transcribing documents

Imagine a company that receives 30 deliveries during the day.

If 25 perfectly match the purchase orders recorded in the ERP, it is not particularly useful for a person to devote the same amount of attention to all 30.

It is much more useful if the system can report:

25 matching goods receipts. 5 require attention.

For example:

• received quantity differs from the ordered quantity;


• unrecognised item;


• missing batch number;


• purchase order not identified;


• delivery exceeds the remaining quantity on the order.

At that point, the operator's work changes radically.

They no longer need to systematically transfer information from a document into the ERP: they need to make decisions about the cases where something genuinely requires verification.

Challenges to consider

Goods receipt naturally involves situations that are more complex than simply reading a document.

A PDF may be of poor quality, a document may contain unknown codes, the same item may be expressed using different units of measure, or a delivery may relate to several orders at the same time.

Furthermore, what is written on the delivery note does not guarantee that it corresponds to what has physically arrived.

For this reason, it is essential to distinguish between information extracted from the document and physical verification of the goods.

Artificial intelligence can read that ten packages have been delivered, but this does not mean that all ten packages are actually present in front of the operator.

The process must therefore include appropriate levels of control and clearly display the automatically interpreted information, the matches identified and any anomalies.

AI should help the operator perform better checks, rather than simply replacing manual entry with an automation process that cannot be verified.

How much time can be saved?

Once again, we can make an estimate.

Let us assume that reading the supplier's document, identifying the purchase order, recording the received items and checking the data takes an average of 6 minutes per goods receipt.

With an AI-assisted process, we can assume that an average of 2 minutes is needed for verification and confirmation.

The average saving would therefore be approximately 4 minutes per goods receipt.

With 10 goods receipts per day, this means saving approximately 40 minutes every day, equivalent to around 147 hours per year.

With 20 goods receipts per day, the saving increases to approximately 293 hours per year.

With 50 goods receipts per day, the potential saving reaches approximately 733 hours per year, assuming 220 working days.

These are, of course, indicative figures: a goods receipt containing two lines is very different from a delivery note listing fifty items.

And this is precisely why the benefit can be even greater for companies receiving complex documents.

No need to ask suppliers to adapt to our ERP

One of the most interesting aspects of the artificial intelligence-based approach is that automation does not necessarily require suppliers to change the way they work.

One supplier may send a PDF.

Another may provide a CSV file.

Another may send an Excel file.

Where available, structured formats and direct integrations naturally remain extremely efficient tools. However, we cannot always impose them on our entire supply chain.

AI can become the intermediate layer capable of transforming different documents into structured information that the ERP can understand.

From data entry to process control

This is the same change we observed in sales order acquisition.

In the traditional process:

goods arrive → document arrives → a person reads it → a person enters the data → the ERP records it.

With an ERP integrated with artificial intelligence:

document arrives → GERP understands it → identifies the order and items → compares the information → prepares the goods receipt → the operator manages any exceptions.

The value of artificial intelligence applied to ERP therefore does not lie solely in doing what we already do more quickly.

It lies in changing where we invest people's time and expertise.

Less time spent reading and transcribing documents.

More time spent checking discrepancies and making decisions.

In the next article in this series, we will explore how this same principle can be applied to another particularly interesting administrative process: receiving, checking and posting purchase invoices using artificial intelligence.