AI in the finance function: where should a Belgian SME start?

What does AI actually change in a finance function?

AI in finance covers three families of tools that differ in age and in purpose. Classic machine learning learns from history to classify or predict: it suggests an account code or flags an unusual invoice. Generative AI, mainstream since late 2022, produces text, summaries or code from an instruction. Agents chain tasks under supervision, such as chasing a customer. Deloitte France (October 2025) argues for a hybrid model combining these approaches, and notes slower adoption in finance than elsewhere.

In every case, AI augments the finance team rather than replacing it. Our reading: transactional work (data entry, matching, repetitive checks) shrinks, and the time freed goes to analysis and to the conversation with the owner. It is the shift from bookkeeper to business partner described in the 2026 CFO priorities read for an SME, and an SME can start it without any platform.

The real bottleneck is not the algorithm, it is the data. In October 2025 Deloitte recalls a figure from its CFO Signals survey of the first quarter of 2023: 64% of the finance leaders surveyed felt insufficiently equipped to make their data reliable and exploit it fully, a challenge that external sources (ESG reporting, predictive models) make harder. An SME therefore has every reason to start with its foundations.

What are the six families of AI use cases described by Deloitte?

Deloitte France draws six families of use cases from its engagements, observed mostly in large groups, but each has an equivalent at SME scale. The examples quoted (agent platforms at two industrial companies, proprietary simulation tools) assume FP&A teams and an information system that few Belgian SMEs have. The principle, however, transposes.

Our reading: two families are available to an SME right away, two come through the accounting software vendor, and two remain the preserve of large organisations. The six families and their transposition:

• 1. Code generation and migration: at Deloitte, generative AI converted VBA macros to Python or R with fewer errors, and attracted digital profiles. SME equivalent: document or replace the Excel macros behind your reporting.
• 2. Automated data analysis: agents connected to the finance system read the history of spending and sales and produce a commented summary and charts. SME equivalent: the monthly dashboard with commentary, drawn from the accounting export.
• 3. Multi-scenario forecasting: a platform models many scenarios from statistical models (time series, ARIMA, ETS) and external assumptions such as inflation. SME equivalent: the three-scenario cash flow forecast.
• 4. Risk management and anomaly detection: real-time analysis of transactions spots unusual patterns, alerts before the incident and strengthens internal control. SME equivalent: checking supplier payments and duplicates.
• 5. Regulatory compliance: identifying gaps, producing the required reports and tracking changes in the rules to reduce the risk of non-compliance. SME equivalent: VAT and Peppol checks before filing, monitoring of obligations.
• 6. Financial narrative: helping investor relations prepare the accounts meant for regulators and analysts, with generated summary and commentary. SME equivalent: the commentary on the annual accounts for the bank or the family shareholder.

What are the 3 realistic AI use cases for a Belgian SME?

An SME does not need a proprietary platform: three use cases fit inside a standard ERP or accounting package. The software vendors present in Belgium already embed learning functions on bank feeds and invoices; the challenge is to switch them on with method and keep control of the exceptions.

The right first use case is the one with high volume and low risk. In most SMEs, that is bank reconciliation: hundreds of lines a month, a gain measurable within a few weeks. Cash flow forecasting comes next, once due dates are reliable, as we show in our article on steering an SME through uncertainty with simple scenarios.

• 1. Assisted bookkeeping and bank reconciliation: the tool proposes the account, the counterparty and the matching; the bookkeeper validates in one click and corrects the exceptions.
• 2. Multi-scenario cash flow forecasting: the model projects receipts and payments from customer, supplier and VAT due dates, then simulates a late payment or a wage indexation.
• 3. Anomaly detection on entries and payments: duplicate invoice, changed supplier IBAN, out-of-range amount; the tool flags, the human decides.

What are the full prerequisites before launching AI in SME finance?

A model learns from your data: if the data is wrong, it will learn to be wrong fast. The first job is therefore the quality of incoming flows. Structured e-invoicing, which Peppol is generalising between Belgian VAT-registered businesses, delivers machine-readable invoices with no re-keying: we cover this in our Peppol guide for Belgian businesses, worth reading before any project.

Deloitte lists five structuring workstreams for a large finance department; at SME scale they boil down to five questions. Its data governance (quality standards, cleansing of historical data, a dedicated committee, awareness) becomes in an SME a written posting rule and a quarterly meeting with the ITAA accountant. Deloitte also notes a virtuous circle: AI itself makes data more reliable by spotting gaps and inconsistencies.

None of these prerequisites costs a licence, and each remains useful without AI. They can be checked in one meeting with the accountant and the office manager, and fixed within a few weeks. The five workstreams, transposed:

• Data: structured invoices via Peppol, deduplicated counterparties, IBANs and VAT numbers up to date, a maintained chart of accounts, a named person who validates the output, including what is shown to the bank.
• Architecture and interoperability: Deloitte talks about hybrid cloud and integration with the ERP and planning tools; for an SME, an accounting package connected to the bank and to Peppol, with a clean export to reporting.
• Compliance: knowing where the data is hosted and who has access, and being able to trace every proposal made by the tool (EU AI Act, GDPR).
• Talent and skills: Deloitte cites critical thinking, ethics and the drafting of instructions (prompt engineering); an SME trains its bookkeeper to review, explain and challenge a proposal.
• Change management: Deloitte insists on communication, training and support; an SME explains to the team what the tool does and does not do, to avoid rejection.

What benefits can you expect from AI in finance, and for whom?

Deloitte distinguishes three beneficiaries: the finance team, the company and its customers. For the team, automating transactional processes cuts costs and frees time for analysis, in exchange for new skills (data, technology, project management, the ability to tell the story behind the numbers). For the company, Deloitte expects shorter planning cycles, more reliable financial and non-financial data and real-time steering.

The third beneficiary, the end customer, is the one an SME forgets most often. Deloitte notes that finance processes must follow new expectations: digital channels, payment methods, generations that no longer want a paper invoice. For a Belgian SME, this translates into the Peppol invoice, the payment link and the automated but reviewed reminder, good for the customer experience as well as for cash flow.

What are the obstacles and the mistakes to avoid?

The first obstacle is the lack of in-house expertise: it feeds the black-box effect and the team's distrust. Deloitte puts it at the top of the hurdles for finance departments in 2025, with recruitment and training as the answer. For an SME the answer is simpler: only use functions that can be reviewed line by line, and train the bookkeeper to do so.

The second obstacle is investment, the third is indecision in the face of so many possible use cases. Deloitte stresses that data cleansing and process redesign cost money before they pay off, and recommends targeting quick-return uses first, then prioritising according to strategy, for instance speeding up the close. In an SME, that cost is the cost of well-kept accounts, useful even without AI.

A shifting legal framework and cyber risk complete Deloitte's list. An unstable framework encourages caution and delays investment; processing large volumes of data increases exposure to attacks. An SME that entrusts its data to an external tool must know where and by whom it is hosted. Frequent mistakes to avoid:

• Trying to cover ten use cases at once, and delivering none.
• Handing the subject to the software vendor alone, with no internal owner.
• Letting the tool's proposals through without review, to save time.
• Sending customer or payroll data into a consumer tool, without a contract.

What do the EU AI Act and other regulatory frameworks mean for a Belgian SME?

The European AI regulation (EU AI Act) classifies uses by risk level and requires transparency, documentation and human oversight according to that level. Deloitte cites it first among the emerging frameworks: the American AI Bill of Rights, Chinese rules on algorithms and synthetic content, the Digital Product Passport. A Belgian SME is only affected by those if it exports or sells the products targeted.

What Deloitte retains for finance fits in three words: data processing, traceability, risk management. An SME checks that the GDPR is respected for customer and employee data, that a contract governs the tool, and that it can find out why an entry was proposed. Everyday accounting uses fall under limited risk, with no exemption from review.

How do you roll out AI in finance in 4 steps?

The method that works in SMEs fits in four short steps, from use case to standard practice. It follows Deloitte's pragmatic approach: start from real business needs, run targeted experiments with a quick return, build on experience before welcoming more mature solutions, and manage stakeholders' expectations. Scaled down to two or three people and one quarter, it keeps the same skeleton.

A pilot that measures nothing learns nothing. Set a simple indicator from day one: hours saved per month, closing lead time, anomalies caught before payment. That figure will justify, or not, the next step in front of the owner and the accountant.

• 1. Pick a quick-return use case: high volume, clear rule, measurable gain, such as bank reconciliation.
• 2. Run a limited pilot: one bank account, one entity or one month of entries, with a numeric target set at the outset.
• 3. Put human validation in place: every proposal is reviewed, every correction is logged, the error rate is tracked.
• 4. Capitalise: document the rules learned, extend to the next flow, train the team, then move to the next use case.

Why a part-time CFO to frame and validate?

Human validation is the critical point, and it needs someone who can read a balance sheet and a model. Who arbitrates when the tool flags an anomaly or proposes an optimistic forecast? That framing role belongs to the finance director, as we explain in the difference between a finance manager and a CFO, and it cannot be delegated to the software vendor.

Deloitte anticipates a finance function embedded in the business, where humans and agents work in synergy: the former decide, the latter execute more and more tasks as they learn. An SME does not need a full-time CFO for this decision-making role: a salaried finance director costs 150,000 to 200,000 euros a year in Belgium, employer charges included. A part-time CFO comes in a few days a month to choose the use case, set the validation rules and review the pilot.

Where do you start, in practice?

Start with a one-day review, not with a tender. The five prerequisites are quick to check, and the first use case is chosen by looking at where the finance team actually spends its hours. Actions to take in the coming month:

If nobody in-house can hold this framing, a time-limited outside view is an option. Argafin offers a free 30-minute QuickScan to position your finance function and identify the first useful use case, with no commitment. The Walloon business vouchers can cover part of a diagnostic.

• Switch on sending and receiving invoices via Peppol, and check the quality of the files received.
• Clean the counterparty base: duplicates, IBANs, VAT numbers, payment terms.
• Test automatic posting on one bank account for a month, with systematic review.
• Name the person who validates, set one gain indicator, and decide at the end of the month what comes next.