Analysis · AI for Business
Automate invoicing, follow-ups and stock without touching your ERP
Most Portuguese small and medium businesses do not need a new ERP. They need the ERP they already have to stop leaking time. The good news is that invoicing, follow-ups and stock replenishment can be automated from the outside, using the interfaces the tax authority and your software already expose, without a single migration.
Key takeaways
- Portuguese law sets a hard deadline: invoices must be issued by the fifth working day after the service or delivery, and the elements communicated to the tax authority by the fifth day of the following month, according to the Portal das Finanças.
- The tax authority already publishes a webservice specification and WSDL so any certified programme can communicate invoice data directly, which means automation can sit beside your ERP rather than inside it.
- The central warning from eesel's June 2026 guide is that billing automation fails when teams automate first and trust blindly; the fix is to hand AI only the cases it is confident about and route the rest to a human.
- Recurring invoicing is the lowest-risk starting point: a model, a client, a schedule and automatic emission, as CentralGest Cloud's September 2025 guide describes for avenças, subscriptions and monthly retainers.
- Sage's November 2025 guidance frames automation as a gradual, structural change: start with invoicing and collections, then move to treasury, expenses and payroll, rather than replacing everything at once.
The legal clock is already running
Before any automation project, the deadlines are fixed by law. The Portal das Finanças states plainly that every business or professional activity obliges the emission of an invoice for each transmission of goods or provision of services. Invoices must be issued by the fifth working day after the service is performed or the goods are made available. If payment happens before that date, the invoice must be issued on the date of the advance.
That is the first reason automation pays for itself. A missed fifth-working-day deadline is not a productivity problem; it is a compliance problem. The same page sets a second clock: the elements of invoices issued in a given month must be communicated to the tax authority by the fifth day of the following month, unless the business already uses the Portal das Finanças emission function or the ATGO app, in which case the obligation is waived. For everyone else, the monthly communication is a recurring administrative task with a hard deadline.
What matters for this piece is that the law does not prescribe which software you use. It prescribes what must be emitted, when, and what must be communicated. That leaves room to automate the emission and the communication without replacing the system that holds your clients, products and stock.
Your ERP already speaks the tax authority's language
The most underused asset in Portuguese back-office automation is the webservice the tax authority already publishes. In its frequently asked questions for software producers, the Portal das Finanças confirms that the information needed to develop a certified invoicing programme that communicates documents by webservice is available under Faturação - Regras e mecanismos de comunicação. The WSDL document describing the communication service is published in the same section.
This matters because it means the integration point is public, documented and stable. A business does not need to rip out its ERP to automate invoice communication. It can run a separate, certified layer that reads invoice data from the existing system and communicates it through the webservice. The FAQ even addresses the multi-entity case: a web-based invoicing programme serving several entities needs only one certificate request per software producer, not one per client. That is the architecture of an external automation layer, not a replacement.
The practical implication is that the decision is not binary between keeping a manual process and migrating to a new ERP. The middle path, an automation layer that sits beside the current system and uses the interfaces that already exist, is the one most Portuguese SMEs can actually afford and absorb.
The mistake most teams make is automating first and trusting blindly
The strongest caution in the current literature comes from eesel's practical guide to AI billing support automation, last edited on 23 June 2026. The guide's central warning is that billing is the queue where a wrong answer costs money and trust at the same time, which makes it exactly where naive automation goes wrong and where careful automation pays off most.
The guide distinguishes five jobs that look similar in an inbox but carry wildly different risk: invoice and receipt requests, charge explanations, payment and card updates, refund and cancellation requests, and disputes or chargebacks. Sending a copy of an invoice is read-only and reversible. Issuing a refund moves money. A chargeback is a legal process with a clock on it. The guide's recommendation is to let AI own the low-risk, reversible cases and hand disputes and chargebacks to a human, with a confidence threshold that routes uncertain cases away from automation.
Translated to Portuguese back-office work, the same logic applies to follow-ups and stock. A reminder email about an overdue invoice is low-risk and reversible. A decision to suspend a client's service or to reorder stock based on a forecast is not. The automation should be designed so that the system acts only when it is sure, and escalates everything else to a person who can see the full context.
Recurring invoicing is the lowest-risk place to start
For businesses with avenças, subscriptions, monthly retainers or regular maintenance contracts, recurring invoicing is the automation with the best ratio of time saved to risk taken. CentralGest Cloud's September 2025 guide describes the mechanics: create an invoice model with the fixed fields, products, unit prices and VAT rate; associate the model to a client or contract; schedule the start date, periodicity and end date; then let the software emit and send the document automatically.
The guide's example is a contabilista who invoices a client 150 euros per month. With automatic invoicing activated, the software emits that invoice on the 25th of each month without the process being repeated manually. The same pattern applies to marketing agencies with monthly packages, IT companies with maintenance contracts, gyms with monthly fees, and liberal professionals with recurring services.
The key point is that recurring invoicing does not require an AI layer at all. It is a deterministic, rules-based automation that most certified invoicing programmes already support. Starting here builds the operational habit of trusting a system to emit documents on schedule, which is the foundation for more advanced automation later.
Follow-ups and collections need a human-shaped escalation path
Collections automation is where the risk profile changes. The Brazilian platform IRecebi, in its guide to collection systems, describes what a modern collection platform does: it centralises the entire collection cycle, sends reminders, tracks who paid and who is late, and integrates with payment methods. The guide cites a study showing that 77% of Brazilian households carry some form of debt, with a significant portion in arrears, which it uses to argue that automated, criteria-driven collection is increasingly indispensable.
The same source's practical guide to accounts receivable automation adds a sharper figure: between 2023 and 2024, nearly 7 million Brazilian companies faced default, around 32% of active businesses. It also reports that after settling debts, almost 85% of consumers re-enter default within about 2.4 months, according to CNDL and SPC data. The lesson for a Portuguese SME is not that collections automation is optional; it is that the follow-up sequence must be designed for recurrence, not one-off reminders.
The practical design is a multi-channel escalation ladder: a polite email reminder before the due date, a firmer reminder a few days after, a phone call or personal message after a defined threshold, and only then a formal notice. Automation handles the first two or three steps. A human takes over when the relationship or the amount at stake justifies it. The eesel guide's confidence-threshold principle applies here too: the system should act automatically only on cases it can classify with certainty, and route the rest to a person.
Stock replenishment is a forecasting problem, not a software problem
Stock automation is often framed as an ERP feature, but the actual work is forecasting: knowing which items will run out, when, and how much to reorder. The Emagia guide to AI-native invoicing and billing automation, updated for finance leaders in 2026, describes the same pattern applied to invoicing: AI integrates with ERP platforms such as SAP, Oracle, NetSuite and Microsoft Dynamics to extract transaction data, validate prices, taxes and contract terms, detect anomalies before sending, and generate invoices automatically.
The transferable insight is that the AI layer learns from patterns, detects anomalies and improves accuracy without replacing the underlying ERP. For stock, the equivalent is a layer that reads sales history and current stock levels from the existing system, forecasts depletion, and generates a proposed purchase order for a human to approve. The human approval step is the control that keeps the automation safe.
What the Emagia guide does not claim is that this happens without integration work. It explicitly says the automation integrates with the ERP. The point for a Portuguese SME is that integration is not the same as replacement. The existing ERP remains the system of record. The automation layer reads from it, proposes actions, and writes back only what a human has approved.
The sequence that actually works
Sage's November 2025 guide to automating financial management is the most useful Portuguese-language source for sequencing. It defines automation as replacing manual, repetitive tasks with digital, integrated processes that run continuously with minimal human intervention, and it argues that this is a structural change in how a company manages resources, not a technology trend.
The guide's priority order is explicit: start with invoicing and collections, then move to treasury and bank reconciliation, then expense management, then payroll and legal obligations, and finally analysis and planning. That order matters because each stage builds on the data discipline of the previous one. A business that has not automated invoicing has no clean data to feed a treasury automation.
The guide also stresses a gradual, sustainable start rather than a big-bang replacement. That is the through-line of every source in this piece: the tax authority publishes interfaces, recurring invoicing is deterministic, collections need escalation paths, stock needs forecasting with human approval, and the sequence starts with the lowest-risk, highest-frequency task. None of it requires a new ERP.
Different perspectives
The infrastructure for external automation already exists and is public. The Portal das Finanças publishes a webservice specification and WSDL, certified invoicing programmes support recurring emission, and the pattern of reading from an ERP, acting with confidence thresholds, and escalating to humans is well documented. A Portuguese SME can automate invoicing, follow-ups and stock replenishment in that order, using the system it already has, and see the time savings compound without a migration project or a licence change.
The counter-argument is that an external automation layer is still a second system to run, monitor and secure. Every integration point is a failure point, and a business that automates invoicing without cleaning its client and product data first will simply emit wrong invoices faster. The eesel guide's warning cuts both ways: the confidence threshold that keeps AI safe also means a human is still in the loop for the highest-value cases, so the automation may not save as much time as the optimistic case promises. And for stock, a forecast is only as good as the sales history behind it; a business with patchy records will get patchy reorder suggestions.
Comparison
What to automate first, and what to keep human
| Task | Risk level | Automation approach |
|---|---|---|
| Recurring invoice emission | Low | Fully automate with a model, client and schedule |
| Invoice communication to AT | Low | Automate via the published webservice |
| Overdue invoice reminders | Low to medium | Automate first steps, escalate after a threshold |
| Refund and cancellation requests | High | Keep human, with AI drafting only |
| Stock reorder proposals | Medium | AI forecasts, human approves |
| Disputes and chargebacks | High | Keep fully human |
Our view
At Snip.work we have built exactly this pattern for A Batina, a 32-year-old family business that now runs one system across its point of sale, online store and stock. The invoicing is automated and the data that used to sit in separate ledgers now feeds sharper marketing decisions. What we did not do was replace the ERP. We built a layer beside it that reads what the business already records, acts on the low-risk, high-frequency tasks, and leaves the judgment calls to the people who know the clients. The result is not a headline number; it is a business where the end of the month no longer means a weekend of manual invoicing and the follow-ups happen without anyone remembering to send them. Cut the busywork, build the system, keep the growth.
What to do
- Map your five highest-frequency back-office tasks and mark each as low, medium or high risk using the table above.
- Start with recurring invoicing: create one model, associate it to one client, schedule it, and run it for a full month before expanding.
- Check whether your invoicing programme already supports webservice communication to the AT, and if not, ask your software producer for the integration path.
- Design a three-step follow-up sequence for overdue invoices and automate only the first two steps.
- For stock, pull three months of sales history and current levels, and build a simple reorder proposal that a human approves before any purchase order is issued.