AI and ERP: Automating Supplier Invoice Recording

From reading invoices to matching them against purchase orders and delivery notes, through to proposing accounting entries: how artificial intelligence can simplify one of the most repetitive administrative processes.

After exploring how artificial intelligence can automate sales order entry and goods receipt, we continue our series on practical applications of AI in ERP systems.

In this third article, we address a process that affects virtually every business: receiving, checking and recording supplier invoices.

Once again, we are not discussing future scenarios. The features described in this series are already available in GERP, our ERP software for business management.

The aim is not simply to read an invoice and copy its data into the ERP system.

The real value of artificial intelligence emerges when the system can understand the document, link it to the business transactions that generated it, check that the information is consistent and propose the accounting entry.

Recording an invoice involves much more than entering data

When an invoice arrives from a supplier, the accounts clerk needs to carry out a series of tasks.

They must identify the supplier, check the document, verify the amounts, determine the appropriate account allocation and, where necessary, compare the invoice against purchase orders and delivery notes.

A seemingly simple task can therefore involve several steps.

Consider, for example, an invoice for the purchase of raw materials.

The company has issued a purchase order, the supplier has delivered the goods with a delivery note and subsequently issued the invoice.

All this information is already available, at least in part, within the ERP system.

Yet, in many traditional processes, someone still has to compare it manually.

Why ask someone to recheck information the ERP system already holds?

This is precisely where artificial intelligence can help.

XML, PDF and other invoice formats: what difference does AI make?

In Italy, electronic invoicing means that a large proportion of invoices are already received in XML format.

This is a significant advantage: the data is structured and can be imported automatically into an ERP system.

However, importing an XML invoice does not mean that its accounting entry has been automated.

The XML format provides the supplier details, taxable amounts, VAT rates, amounts and line descriptions.

However, it does not always allow the system to determine automatically which purchase order a particular line relates to, which inventory movement should be linked to it or which general ledger account is most appropriate.

Artificial intelligence adds a layer of interpretation.

GERP can use the information contained in the invoice and compare it against data already held in the ERP system, regardless of whether the document was imported from XML, PDF or another supported format.

AI does not replace standard tools for importing structured documents: it complements them wherever interpretation, matching and exception handling are required.

Linking the invoice to the purchase order

Imagine that a supplier sends an invoice for 5,000 kg of material.

A purchase order for that quantity already exists in GERP.

The system can analyse the available information and look for a match between the invoice and the order.

The link can be based on several factors:

• supplier identification;


• purchase order references;


• item codes and descriptions;


• quantities;


• unit prices;


• total amounts;


• document dates and references;


• supply history.

AI can be particularly useful when references are not fully structured.

For example, the supplier might use a different description from the one on the order, or include its own item code instead of the company's internal code.

Artificial intelligence can interpret these differences and suggest the most plausible match.

Of course, a suggested match should not automatically be considered correct: where there is ambiguity, the system must flag it to the user.

Three-way matching: purchase order, delivery note and invoice

One of the most promising processes to automate is three-way matching, which involves matching three documents:

1. Purchase order: what we ordered and on what terms.

2. Delivery note or goods receipt: what was actually received and recorded.

3. Supplier invoice: what the supplier is charging us for.

This comparison helps verify consistency across the entire purchasing and accounts payable process.

Let us look at an example.

The company orders 5,000 kg of material at €2 per kilogram, for a total value of €10,000.

The supplier delivers 4,800 kg, as shown on the delivery note and in the goods receipt recorded in GERP.

An invoice then arrives for 5,000 kg, totalling €10,000.

The system can immediately identify the discrepancy:

• Quantity ordered: 5,000 kg.


• Quantity received: 4,800 kg.


• Quantity invoiced: 5,000 kg.


• Difference to investigate: 200 kg, equivalent to €400.

The user does not necessarily need to open three documents and compare each line manually.

GERP can display the discrepancy directly and request a decision.

It could be a partial delivery, goods still in transit, an invoicing error or an arrangement already agreed with the supplier.

AI does not necessarily need to decide which explanation is correct.

Its role is to identify the issue and provide the user with the information needed to resolve it.

From verification to a proposed accounting entry

Once the invoice has been identified and checked against the documents in the ERP system, the system can prepare a proposed accounting entry.

This takes the process a step beyond simply capturing data.

GERP can use the available information to determine the nature of the purchase and suggest the appropriate general ledger accounts.

For example, an invoice might relate to:

• purchases of raw materials;


• purchases of consumables;


• transport services;


• machinery maintenance;


• professional consultancy services;


• purchases of fixed assets.

Different types of expenditure require different accounting treatment.

The system can use previous accounting entries, supplier master data, item classifications and the accounting rules configured in the ERP system to prepare the proposal.

Deterministic tax and accounting rules must, of course, take precedence over AI's probabilistic interpretations.

The aim is not to let a language model invent an accounting entry, but to use artificial intelligence to prepare an entry consistent with the information and rules already held in the ERP system.

The user can then review and confirm the proposal before final posting.

Not every invoice has a purchase order

An important consideration is invoices that are not directly linked to a purchase order.

Examples include utilities, software subscriptions, consultancy, recurring services and overheads.

In these cases, matching against an order or delivery note may not be possible or necessary.

Artificial intelligence can still assist the user.

For example, it can recognise that an invoice relates to a recurring service, identify previous entries for the same supplier and suggest an account allocation consistent with those entries.

It can also flag significant changes compared with previous invoices.

If a monthly subscription normally costing €300 is suddenly invoiced at €450, the system can highlight the difference and request a check.

In this way, AI does more than speed up invoice recording: it can also help improve financial oversight.

AI prepares, the accounts team checks

As with the other processes described in this series, it is important to maintain a distinction between automation and oversight when recording supplier invoices.

The process can be represented in six steps:

Receive → Understand → Match → Verify → Propose → Post

Artificial intelligence primarily contributes to interpretation, matching and preparing the proposal.

The ERP system applies business and accounting rules.

The user retains control over exceptions and entries that require judgement.

As the process becomes more reliable, some simple, recurring cases can be handled with higher levels of automation.

However, the principle remains the same: automate repetitive tasks and leave decisions that require expertise and responsibility to people.

Key challenges to consider

Recording supplier invoices is a sensitive process because it has administrative, accounting and tax implications.

Artificial intelligence can misinterpret documents, suggest incorrect matches or misclassify the nature of an expense.

Furthermore, differences between a purchase order, delivery note and invoice do not necessarily indicate an error.

There may be agreed tolerances, partial deliveries, additional charges, authorised price changes or special commercial terms.

The process must therefore include deterministic checks on amounts, VAT rates, document references and whether accounting entries balance.

Particular attention must also be paid to duplicate invoices, credit notes, documents covering multiple orders and situations in which the invoice arrives before the goods.

Another essential factor is traceability.

Every AI-generated proposal should be traceable to the documents and information used to produce it.

The user must be able to understand why an invoice has been linked to a particular order and what information supports the proposed accounting entry.

Finally, because the process involves financial data and confidential information, appropriate security and data protection measures are necessary.

How much time can AI save when recording supplier invoices?

Let us quantify the potential benefit with a simulation.

Suppose an accounts clerk takes an average of 8 minutes per invoice to identify it, check its data, link it to any relevant purchase orders and delivery notes, and prepare the accounting entry.

With an AI-assisted process, we can assume that the average time required for supervision and confirmation falls to 2 minutes per invoice.

The potential saving is therefore approximately 6 minutes per document, or 75% of the time normally spent on the task.

Based on 220 working days per year, we obtain the following estimates:

Invoices per day	Daily manual processing time	Time with AI	Hours saved per year
10	80 minutes	20 minutes	220 hours
20	160 minutes	40 minutes	440 hours
50	400 minutes	100 minutes	1,100 hours

For a company recording 50 supplier invoices per day, the potential saving therefore reaches 1,100 hours per year, equivalent to more than 137 eight-hour working days.

This is, of course, a simulation, not a measurement of GERP's actual performance.

Actual results depend on document complexity, the availability of purchase orders and delivery notes, the quality of master data and the proportion of invoices requiring additional checks.

Savings may be lower for very simple invoices already handled through automatic imports, while they may be higher in processes requiring numerous manual checks.

However, the most significant benefit goes beyond time savings.

Systematic matching can help identify discrepancies promptly, reduce incorrect entries and improve the quality of accounting information.

From invoice entry to oversight of purchasing and accounts payable

The change introduced by artificial intelligence is not simply about entering a document faster.

In the traditional process:

The invoice arrives → the user reads it → searches for orders → compares delivery notes → identifies the accounts → records the entry.

With an ERP system that integrates artificial intelligence:

The invoice arrives → GERP interprets it → matches it against purchase orders and delivery notes → flags discrepancies → proposes the accounting entry → the user checks and confirms.

The focus shifts from transcribing information to overseeing the process.

This is one of the most important aspects of the evolution of ERP systems.

For years, we have used business software to collect, organise and store information.

Today, thanks to artificial intelligence, we can begin using that same information to prepare transactions, identify inconsistencies and provide practical support for people's daily work.

This does not mean dispensing with accounting expertise.

It means enabling people with that expertise to apply it where it creates the most value.

An ERP system that understands business processes

Supplier invoice automation brings together the journey that began in the first two articles in this series.

We have seen how AI can support sales order capture, goods receipt and now supplier invoice recording.

These are different activities, but they share a fundamental requirement: turning information received from outside the company into transactions within the ERP system.

Artificial intelligence can connect these steps and reduce the need for repetitive manual work.

With GERP, the aim is to build an ERP system that goes beyond recording what happens in the business and can understand information, connect it and help people take action.

Because the value of AI in business is measured not by the number of intelligent features in the software, but by the time it gives back to people and the quality of the decisions it enables them to make.

Less time entering data. More time managing your business.