Automating invoices and orders in a Spanish company is not about buying an "AI that reads invoices" and forgetting about it. It is about designing a concrete flow, deciding with judgement which parts run on their own and which go through a person, and leaving an auditable trail of every decision. This is the no-hype guide to how we approach it: what to automate, what to review by hand and what to have sorted before starting.
The full cycle, step by step
When we talk about "automating invoices and orders" we are really talking about a cycle with several stages. It is worth seeing it whole before touching anything, because the bottleneck is almost never where the company thinks it is.
- Intake: the document arrives by email, attached PDF, web form or supplier portal. Different formats, different senders, different quality.
- Data extraction: the relevant fields are read — supplier, invoice or order number, lines, amounts, taxes, dates.
- Validation: what was extracted is checked against what the business already knows — approved budget, purchase order, stock available, agreed rate.
- Posting: the validated data goes into the ERP or management system without anyone retyping it.
- Exceptions: what does not add up is neither invented nor forced, it goes to a person with all the context so they can decide.
Useful automation covers the five stages as one continuous flow. If you only automate reading the PDF but then someone copies the data into the ERP by hand, you have saved little and added a new review step.
What to automate and what to leave under human review
This is the decision that separates a project that saves hours from one that breeds distrust. Not everything should run on its own, and admitting that is not a lack of ambition: it is honest engineering.
| Stage | Recommendation |
|---|---|
| Receiving and classifying the document | Automate |
| Extracting structured fields | Automate with a confidence threshold |
| Validation against order, budget and stock | Automate by rules |
| Posting to the ERP for cases that add up | Automate |
| Invoices with no associated order or with an amount discrepancy | Human review |
| New supplier or incomplete tax details | Human review |
| Final payment approval | Human review |
The criterion we use is simple: automate the happy path, route the exception. A high percentage of invoices and orders are routine and add up first time; those can flow on their own. The rest — the supplier who changed IBAN, the amount that does not match the delivery note, the line with an odd discount — is exactly where human judgement adds value and where a machine deciding blind creates expensive problems.
Here a central idea of how we work comes in: no black box. Every extraction carries a confidence level, every validation makes clear which rule passed or failed, and every exception reaches the person with the original document and the reason for the rejection in view. It is not enough for the system to be right; it has to be able to explain why.
What you need sorted before starting
Most automation projects that fail do not fail because of the technology. They fail because automation started on a process nobody was clear about. Before writing a line of code, it is worth having these foundations.
Clean data and master records
If your supplier catalogue has duplicates, inconsistent references or rates that live in somebody’s head, the automation will inherit that mess. Automatic validation needs a truth to compare against. Often the first real job is not automating, but tidying up the master records.
Integration with the ERP
Automatic posting depends on being able to write into your management system in a controlled way, via API, an official connector or, at worst, a bespoke integration. It is worth knowing early what your ERP allows, because it conditions the whole design. If the system exposes no structured way in, that is the first thing to solve.
Explicit business rules
"When does an invoice get approved on its own?" must have a concrete answer before anything is coded: what amount discrepancy is tolerated, what happens if the purchase order is missing, who approves above a certain threshold. If those rules are not written down, the automation will improvise them, and that never ends well.
GDPR and electronic invoicing in Spain
Automating the invoice cycle means processing data about suppliers, customers and, sometimes, individuals. That falls under the GDPR: you have to be clear about what data gets processed, where it is stored, who accesses it and how long it is kept. A well-designed automated flow makes this easier, because it centralises the processing and leaves a record of accesses instead of having invoices scattered across mailboxes.
In parallel, Spain is moving towards a more demanding digital invoicing framework. Systems such as the tax authority’s SII for supplying VAT information, or the requirements around Verifactu and mandatory business-to-business electronic invoicing, push towards more traceable and verifiable processes. We are not going to detail deadlines or specific obligations here — your tax adviser should confirm those against the regulation in force — but there is an underlying message worth stating: the direction of the regulatory framework rewards precisely those processes that are auditable. Automating well today, with logs and traceability, leaves you better prepared for what is coming than carrying on managing invoices by hand in an inbox.
Common mistakes we see
- Automating before diagnosing. Buying the tool and then trying to fit the process to it. The right order is the reverse: understand the flow, measure where it hurts and only then automate where there is a real return.
- Trusting extraction one hundred per cent. No system reads documents with perfect accuracy across heterogeneous formats. Without a confidence threshold and an exception route, a silent error slips into the accounts.
- Forgetting the exceptions. The project is designed for the ideal case and, when the first odd case arrives, there is nobody to route it to and no context to route with it. Exceptions are not a system failure: they are part of the design.
- Leaving no trail. If, when something goes wrong, nobody can reconstruct what the system decided and why, you do not have an automation: you have a black box everyone is afraid to touch.
- Measuring success in "invoices processed" rather than hours freed. The goal is not to move paper faster, it is to give time back to people so they can deal with what does not add up and needs judgement.
The realistic return
Let us be honest with the numbers, because there is a lot of hype here. We are not going to promise you a magic percentage saving: it depends on your volume, on the quality of your starting data and on how much exception your business has.
What we can describe is the pattern that repeats. When the happy path covers most of your routine invoices and orders, the team stops spending hours typing and classifying, and starts spending a fraction of that time resolving exceptions with context. Cycle time, from the document arriving to it being posted and validated, usually goes from days to hours, simply because the retyping queues disappear. The real saving is in the administrative hours freed and the errors that stop slipping through, not in a marketing number.
The return appears when you automate with your head: the repetitive to the machine, the doubtful to the person, and all of it with an auditable log. If someone promises you total automation with no human review and no traces, be suspicious — that is not efficiency, it is risk in disguise.
The next step
Automating invoices and orders works when it is done as engineering, not as a software purchase: diagnose first, automate the happy path, leave the exceptions in people’s hands and record every decision. That is exactly how we work on process automation: only where there is a real return, with no black box and with human review of what matters.
If you want to know which part of your invoice and order cycle makes sense to automate and which does not, the first thing is to look at the process properly. Ask us for a diagnosis and we will tell you honestly where the return is and where it is not.