Chasing Unpaid Invoices with AI: A Belgian SME Guide

Late payment: the silent drain on Belgian SMEs

If you run an SME in Belgium, you know the scene: the invoice went out six weeks ago, the work was delivered long since, and the payment still hasn't landed. Chasing unpaid invoices is one of the most disliked tasks among freelancers and SME owners — awkward with the client, time-consuming, and yet vital. Payment barometers published by credit analysis firms such as Graydon regularly indicate that a substantial share of Belgian B2B invoices, often estimated at around a third, is settled late. Every day of delay is a day your cash flow finances your client's for free. The good news: invoice chasing is precisely the kind of repetitive, structured, high-value process that artificial intelligence automates best. This guide shows you how — within the Belgian legal framework.

What Belgian law says about late payment

Before automating anything, know your rights. For business-to-business (B2B) transactions, the matter is governed by the Belgian Act of 2 August 2002 on combating late payment in commercial transactions, which transposes EU Directive 2011/7/EU. Since its 2021 reform, the payment term may no longer exceed 60 calendar days, even by contractual agreement between the parties.

In practice, when a B2B payment is late you are entitled — automatically, without formal notice — to interest at the statutory rate specific to commercial transactions, a rate published every six months by the Belgian Federal Public Service Finance and which has hovered around 12% in recent years, plus a fixed compensation of 40 euros for your recovery costs. The FPS Economy publishes the detailed rules and the rate currently in force.

Few SMEs actually claim these amounts, often for fear of damaging the commercial relationship. That choice is yours to make. But mentioning these rights in a level-2 or level-3 reminder, factually and politely, often changes payment behaviour — and that is exactly the kind of graduated escalation an automated system applies consistently, where a human forgets or doesn't dare.

Why manual chasing doesn't work

Most Belgian SMEs chase payments "when they think of it". The pattern is always the same: for the first few weeks, new projects take all the attention; then a cash squeeze forces someone to open the overdue list; a batch of reminders goes out all at once, sometimes clumsy because written under pressure.

This approach creates three problems. First, irregularity: a client who learns that your reminders arrive at random quickly concludes that paying late costs nothing. Second, hidden cost: between checking incoming payments, drafting reminders and following up on payment promises, an owner or assistant easily spends several hours a week — time you invoice to nobody. Third, the emotional load: many independents postpone chasing because they hate confrontation, which mechanically worsens collection times.

This triple pattern — a repetitive, rule-based task with structured data (invoices, due dates, payments) — ticks every box of the ideal tasks to automate with AI. Chasing doesn't require creativity: it requires regularity, tact and memory. Three qualities where the machine beats the tired human.

What AI changes compared with a basic automatic reminder

Traditional invoicing software has long offered fixed-date automatic reminders. Generative and predictive AI goes further, on four fronts.

Reminders that are written, not templated

A language model drafts each reminder with the context in mind: amount, age of the receivable, the client's history, the tone of previous exchanges. The first reminder to a loyal, reliable client should not sound like the third reminder to a structurally bad payer. This personalisation — impossible to maintain manually across dozens of invoices — protects the commercial relationship while closing the door on excuses.

Predicting payment risk

By analysing your payment history, an AI system spots clients whose behaviour is deteriorating — lengthening delays, partial payments — and alerts you before the situation becomes critical. Some signals, such as a client who used to pay at 30 days gradually drifting to 75, are invisible to the naked eye but trivial to detect statistically. This predictive logic is the same one we describe in our guide on AI applied to treasury and cash flow.

Automatic payment reconciliation

Before chasing anyone, you need to know who has paid. AI automatically matches bank statements (via the CODA exports from your Belgian bank) against open invoices, even when the payment reference is missing or approximate. No more reminders sent by mistake to a client who paid the day before — the most credibility-destroying error of all.

Handling the replies

When a client answers "I'll pay on Friday" or "could you resend the invoice?", an AI assistant classifies the reply, updates the file, schedules a check on the promised payment and prepares the appropriate response. It is the natural extension of automated email management.

Building your automated dunning workflow, step by step

Here is the typical architecture we set up for a services or trading SME.

Step 1 — Make the source data reliable. The workflow starts from your invoicing software. If your invoices still come out of Word or Excel, start there: electronic invoicing via Peppol, mandatory for B2B since January 2026, requires a structured tool anyway, and it is a healthy foundation.

Step 2 — Define the cadence and the escalation levels. Typically: a courteous reminder at 3 days past due, a second reminder at day 14 mentioning statutory interest, a formal notice at day 30, and a handover to human treatment or a collection partner at day 45. You set the rules; the AI applies them without hesitation and without forgetting.

Step 3 — Plug in AI drafting. Each level triggers the generation of a personalised email, in the client's language — French, Dutch, English or German, a real issue in Belgium. You approve the first sends manually, then switch to management by exception: the AI sends, and you only see the atypical cases.

Step 4 — Close the loop with the bank. Daily bank reconciliation automatically stops the reminders as soon as a payment arrives, and feeds an outstanding-receivables dashboard that gives you, every morning, the exact picture of who owes you what.

This project often joins a broader effort to automate the processing of incoming invoices: the two share the same connectors and justify each other.

The special case of consumer clients: regulatory caution

If your clients are consumers (B2C), the framework has been much stricter since Book XIX of the Belgian Code of Economic Law on consumer debts entered into force on 1 September 2023. In short: the first reminder must be free of charge, a waiting period of fourteen calendar days must elapse after that reminder before any cost can be applied, and the compensation and interest you may claim are capped according to the size of the debt. Mandatory statements must appear in every reminder.

This is where well-designed automation becomes a compliance asset: a workflow configured once and for all respects the fourteen-day period and the caps on every send, whereas an improvised manual reminder risks a misstep. But it does require the rule to be encoded correctly from the start — which is why we recommend having your B2C reminder templates validated by your accountant or legal adviser before going live, based on the official information from the FPS Economy. It is the kind of AI integration mistake that costs dearly when discovered after the fact.

What does it cost, and what does it return?

An automated dunning project sits at the low end of the AI project range: the building blocks already exist (invoicing software, an automation tool, a language model via API), and most of the work consists of connecting them cleanly and encoding your escalation rules. Expect a project of a few days to a few weeks depending on how clean your data is — an order of magnitude detailed in our article on the cost of AI integration for a Belgian SME.

On the return side, three measurable effects. One: a shorter average collection period (DSO) — systematic reminders from day 3 typically shave several days off payment times, which translates directly into available cash. For an SME invoicing 500,000 euros a year, gaining ten days of DSO frees up roughly 13,700 euros of cash permanently. Two: recovered administrative time, often two to four hours a week. Three: fewer receivables written off entirely, because problem files are detected and escalated earlier. Our method for calculating the ROI of an AI project applies to this use case as is.

Which tools for automating reminders in Belgium?

Three layers of tooling come into play, and you probably already own part of the stack. The first layer is your invoicing software: most solutions common on the Belgian market — Odoo, Teamleader, Billit and the like — natively offer basic automatic reminders and exports to your accountant. If yours already covers your needs with fixed-date reminders, switch them on today: it is free and solves part of the problem.

The second layer is orchestration: automation platforms such as Make or n8n connect your invoicing, your mailbox and your bank, and execute the escalation logic described above. This is where the real difference with a simple reminder is made: conditions, exceptions, escalations, automatic cut-off when a payment lands.

The third layer is the language model itself, called via API, which drafts the contextualised reminders and interprets client replies. The usage cost is marginal at SME scale: a few cents per generated reminder.

The selection criterion is not the power of the tool but the quality of integration with what you already have: a simple setup plugged into your current invoicing software beats a sophisticated platform that forces you to migrate. Also be wary of "all-in-one" collection services billed as a percentage of recovered amounts: they have their place for difficult receivables, but for routine reminders, a workflow you own is structurally cheaper. This is exactly the kind of trade-off we clarify in a scoping engagement before any purchase decision.

Where to start?

Start small and measure. Export your invoices from the last twelve months, compute your real average payment time and the amount currently overdue: that figure is your baseline, and it is usually enough on its own to justify the project. Then pick a pilot scope — for instance B2B invoices only, in French — and run the workflow in draft mode for a month: the AI prepares the reminders, you click send. Once trust is established, move to automatic sending with supervision by exception, then extend to other languages and segments.

At Aïves Consulting, we support SMEs in Wallonia and Brussels on exactly this kind of project: auditing your invoicing process, choosing the tools, building the workflow and training your team. Have a look at our services, and if unpaid invoices are costing you sleep, let's talk — the first thirty-minute conversation comes with no strings attached.