AI and ERP: automating the receipt and entry of sales orders

Artificial intelligence is becoming part of business processes, but its value does not simply lie in adding a chat interface to ERP software.

The real change happens when AI is able to read the information entering the company, understand it and turn it into operations within the ERP.

With this article, we begin a series dedicated to the practical use of artificial intelligence in ERP-managed processes: from receiving sales orders to goods receipt, from posting purchase invoices to managing non-compliances, through to production analysis and querying company data using natural language.

We will not be talking about futuristic scenarios.

The features we will describe in this series are already available and can be used with GERP.

Let us start with one of the most common processes: receiving an order from a customer and entering it into the ERP.

An apparently simple process

In many companies, customer orders still arrive through very different channels and formats.

They may arrive as PDFs attached to an email, as text in the body of a message, through an Excel file or as documents generated by the customer's ERP system.

The content may be understandable to a person, but it is not necessarily immediately usable by the ERP.

An operator therefore has to open the message, read the document, identify the customer, find the items ordered, interpret codes and descriptions, enter quantities, prices and delivery dates, and finally check that what has been entered actually matches the order received.

When multiplied by dozens or hundreds of orders, this process represents a significant amount of administrative work.

Above all, it is work in which a person spends much of their time transferring information from one system to another.

This is exactly the kind of task in which artificial intelligence can be particularly effective.

From the email inbox directly into GERP

With GERP, artificial intelligence can intervene from the moment an order is received.

Imagine that a customer sends a normal email containing an order in PDF format.

The system can analyse the message and the attached document, understand their content and identify the information required to enter the order.

For example:

• customer;


• customer order number and date;


• requested items;


• product codes;


• descriptions;


• quantities;


• prices;


• commercial terms;


• requested delivery date;


• any notes or instructions.

This information is not simply extracted as text.

AI can relate it to the data already available in the ERP.

This is the fundamental difference between using a document recognition system and truly integrating artificial intelligence into an ERP process.

Understanding what the customer ordered

The code used by the customer does not always match the code used internally by the company.

A description may be abbreviated, written differently or refer to a previous supply.

A customer might write, for example:

20,000 product X labels, same as the last order.

For an operator who knows the customer, the request may be perfectly understandable.

A traditional system based exclusively on rigid fields and rules, however, would have difficulty automatically turning it into an order.

Artificial intelligence can interpret the request and use the information available in GERP to identify the correct product, check the customer's history and prepare the order using the available ERP data.

The ERP therefore stops being merely the place where someone records information that has already been interpreted.

It becomes part of the interpretation process itself.

AI prepares, the operator checks

Automation does not necessarily mean eliminating human control.

On the contrary, one of the most interesting models for using AI in business processes is:

read → understand → verify → propose → approve → execute

GERP can therefore prepare a proposed order based on the information received.

The operator finds the document already completed and can focus on the information that genuinely requires their attention.

If everything is correct, the operator confirms the order.

If there are inconsistencies, missing information or unusual situations, these can be highlighted before the order is finally entered.

The goal is not to take control of the process away from the operator.

It is to remove the work that does not require the operator's judgement.

Managing exceptions instead of entering data

This also changes the role of the ERP user.

In the traditional model, the operator has to check every piece of information and enter every order.

In an AI-assisted process, the operator can instead focus on exceptions.

If 50 orders arrive and 43 are fully consistent with the customers, products and terms already stored in the ERP, the operator can focus their attention on the seven that contain something unusual.

A price that differs from the expected one.

An unrecognised product.

An unusual quantity.

A potentially problematic delivery date.

A customer code that has never been used before.

Commercial terms that differ from those stored in the system.

Artificial intelligence therefore becomes a filter between the large amount of information entering the company every day and the decisions that genuinely require human intervention.

Less manual data entry also means fewer errors

Manual order entry does not only involve a cost in terms of time.

It also creates the possibility of errors.

An incorrectly entered quantity, an item selected by mistake or an incorrectly transcribed delivery date can propagate to subsequent stages: planning, production, procurement, warehouse, shipping and invoicing.

Automating order acquisition means reducing the number of steps in which the same information is first read and then re-entered.

The data received from the customer is interpreted, compared with the information available in the ERP and used directly to prepare the ERP document.

There is no need to impose a new system on the customer

There is also another particularly important aspect.

To automate the process, it is not necessarily necessary to ask the customer to change the way they work.

There is no need to force them to use a B2B portal, fill in a structured form or use a specific data interchange format.

The customer can continue sending their order in the way they normally use.

It is the company's system that becomes better able to understand the information it receives.

This represents one of the most interesting changes introduced by artificial intelligence into ERP systems: people no longer necessarily have to adapt to the rigid structure of the software; the software can learn to interpret the way people and companies communicate.

From an ERP that records to an ERP that acts

For many years, we have used ERP systems mainly to record what happens within a company.

A customer places an order.

A person interprets it.

A person enters it into the ERP.

The ERP records it.

Artificial intelligence makes it possible to change this sequence.

The customer sends the order → GERP understands it → prepares the operation → the user checks it → GERP executes it.

This is an important transition: from an ERP used simply as a system of record to an ERP capable of actively participating in business processes.

And this is precisely the topic we will explore in the next articles.

We will see how the same principle can be applied to goods receipt, purchase invoice posting, document management, non-compliances, production and many other processes.

With one common principle: using artificial intelligence not to add complexity, but to free people from repetitive tasks and allow them to focus on decisions that genuinely require human intelligence.

Critical issues and control: where AI should stop

How much time can automatic order entry save?

Let us try to translate automation into numbers. Let us conservatively assume that receiving an order, opening the email and any attachments, interpreting its content, entering the data into the ERP and checking what has been entered takes an average of 5 minutes of work. With an AI-assisted process, we can assume that the operator spends an average of 1 minute reviewing and confirming the automatically prepared order. The saving would therefore be approximately 4 minutes per order.

With 10 orders per day, this means saving around 40 minutes per day, equivalent to approximately 147 hours per year. With 20 orders per day, the saving rises to approximately 1 hour and 20 minutes per day and 293 hours per year. With 50 orders per day, the potential saving reaches approximately 3 hours and 20 minutes per day, equivalent to around 733 hours per year. The calculation assumes 220 working days per year and is, of course, an estimate: the actual situation depends on the complexity of the orders and the level of automation that can be achieved.

At 50 orders per day, we are therefore talking about more than 90 eight-hour working days recovered every year. But the benefit is not limited to the hours saved: it also means reducing transcription errors, speeding up order processing and allowing people to devote more time to managing exceptions and to activities that genuinely require expertise and decision-making skills.