How does an SME use AI in accounting without having its own IT department?

You don’t start with an AI project, you start with a single process that already has clear rules — in accounting that is almost always incoming invoices. Without your own IT department, the realistic path is to use a standard capability already built into your accounting software rather than building something yourself. What decides the outcome isn’t the model, it’s whether the process was described cleanly beforehand.

What you’ll take away:

• The entry point is one single rule-based process — not “AI in accounting” as a programme.
• Without in-house IT: integrated standard features of your accounting software before specialist tools, specialist tools before custom builds.
• According to the Bitkom AI study 2026, 41 percent of companies with 20 or more employees actively use AI — up from 17 percent a year earlier. Companies between 20 and 500 employees lag behind the large ones.
• A process ready for automation has four traits: repetition, clear rules, digital source data, and measurable effort.
• Since the Digital Omnibus, systems that merely make workflows more efficient are no longer automatically high-risk — accounting automation is usually not a high-risk case.
• Without an IT department, the bigger risk isn’t the technology, it’s shadow AI: documents ending up in private accounts.

How does an SME use AI in accounting without having its own IT department?

By starting small and going through standard software. Take a process that repeats daily and follows fixed rules — typically incoming invoice processing: capture the document, read the line items, suggest accounts and cost centres, route it for approval. That exact step is now included as a feature in most common accounting and ERP systems.

Without in-house IT, the decisive selection criterion isn’t how capable the model is, it’s the integration effort. A feature already sitting in your existing system costs you configuration. A specialist tool with its own interface costs you a project — and projects need someone to run them. If that role is missing, choose the more boring option.

How do I know my invoice process is ready for automation?

By four traits: the task repeats frequently, it follows rules you can explain, the source data is already digital, and the current effort is measurable. If one of those is missing, you’ll automate chaos — and get faster chaos.

Concretely, for incoming invoices:

• Repetition: how many incoming invoices per month? Below roughly 100 the setup effort usually doesn’t pay off yet; above that it gets interesting quickly.
• Clear rules: can you describe who approves which invoice up to which amount? If that rule only exists “in Ms X’s head”, that’s the first work item — not the software.
• Digital source data: do invoices arrive as PDF or e-invoice, or as paper in an envelope? Paper means an extra step upfront.
• Measurable effort: how long does processing one invoice take today? Without that number you won’t be able to prove afterwards that anything improved.

The last point is the one most often skipped. Measure the baseline before you change anything. Otherwise you end up with a feeling instead of a result.

Which accounting processes are the best place to start?

The ones that already have structure. Roughly sorted by how easy they are to start with:

Process	Ease of entry	Why
Incoming invoices and account coding suggestions	high	Repeats daily, clear rules, included in standard software
Document recognition and matching	high	Pure pattern recognition, result immediately verifiable
Dunning and payment reminders	medium	Rule-based, but sensitive to customer communication
Travel and expense claims	medium	Many edge cases, but high relief for the team
Liquidity forecasting	medium	Needs clean historical data
Year-end close preparation	low	Too many case-by-case decisions, high audit relevance

Start at the top. Incoming invoices aren’t the most exciting topic, but they deliver the result you can use internally to justify the next steps.

Do I need to worry about the EU AI Act?

Usually not in the high-risk category. The Digital Omnibus, in force since 27 July 2026, narrowed the high-risk classification: systems that merely assist, make workflows more efficient or automate processes are no longer automatically high-risk — provided their failure poses no genuine risk to health or safety. Automated account coding doesn’t meet that bar.

Two points remain relevant, though. First, the AI literacy obligation under Article 4: since 2 February 2025 it requires measures to support your employees’ AI literacy — after the Omnibus without guaranteeing a specific level of knowledge, but still as an obligation to act. Second, the transparency obligation under Article 50, applicable since 2 August 2026 and subject to fines. It rarely hits accounting directly, because its output stays internal — it becomes relevant as soon as AI-generated text goes outside, for example in dunning correspondence.

If you want to use AI to support recruitment or performance evaluation, that’s a different discussion. The high-risk rules still apply there.

Without in-house IT, what is the bigger risk?

Shadow AI. If there is no usable official route, an unofficial one appears: an employee pastes an invoice into a private ChatGPT window because it’s faster. Supplier data, terms and potentially personal data then travel to servers you know nothing about and that appear in no register.

Bans don’t solve that — they just move it. What does solve it is an official route at least as convenient as the unofficial one, plus the competence to recognise the boundary. Where that knowledge is missing, the PASSION4IT Academy is the narrow entry point: role-specific learning paths, no classroom dates, with a certificate as proof.

The second half is governance, and it’s smaller than it sounds: a list of approved tools, a rule for confidential data, one contact person for doubts. That fits on a single page.

How do I measure whether it paid off?

Against the number you captured before you started. Processing time per invoice, number of manual corrections, cycle time to approval. Three metrics are enough — nobody sustains more than that.

What you shouldn’t do is estimate “time saved” after the fact. That number always comes out positive and convinces nobody who has to verify it. How to build a credible impact measurement for AI is covered in How do I measure whether AI really benefits my company.

And if you want to know where your company stands overall before the tool question: that’s exactly what the Digital Check is built for. It assesses digital maturity across five dimensions — strategy, processes and organisation, technology and IT infrastructure, employees’ digital skills, and products, services and business models — and ends with a prioritised roadmap instead of a tool list. The sequence stays the same: first the honest maturity check, then the roadmap, then supported implementation.

Frequently asked questions

How does an SME use AI in accounting without having its own IT department?

Through a single rule-based process and through standard features of the accounting software you already run. The usual entry point is incoming invoice processing with automated account coding suggestions. Without in-house IT, integration effort is the most important selection criterion — integrated feature before specialist tool, specialist tool before custom build.

From what invoice volume does automation pay off?

As a rough guide: below around 100 incoming invoices per month the setup effort usually doesn’t pay off yet; above that it gets interesting quickly. What matters isn’t volume alone, though, but whether approval rules are documented and documents arrive digitally.

How do I recognise that a process is ready for automation?

By four traits: it repeats frequently, it follows clear rules, the source data is digital, and today’s effort is measurable. If one of those is missing, clarify the process first and automate afterwards.

Is AI in accounting a high-risk system under the EU AI Act?

Usually not. The Digital Omnibus narrowed the classification: systems that merely assist or make workflows more efficient are no longer automatically high-risk if their failure poses no genuine risk to health or safety. AI in recruitment or performance evaluation is a different matter.

Which EU AI Act obligations still apply to me?

The AI literacy obligation under Article 4 has applied since 2 February 2025 as an obligation to take measures — support literacy, don’t guarantee a particular level. The transparency obligation under Article 50 has applied since 2 August 2026 and carries fines; it bites as soon as AI-generated content goes outside, for example in customer correspondence.

What is shadow AI and why is it dangerous in accounting?

Shadow AI is the use of private AI accounts without approval. In accounting it’s especially sensitive because supplier data, commercial terms and personal data travel to unknown servers. Bans only displace the problem; a usable official route plus competence solves it.

Do I need external support to get started?

For tool configuration, usually not. For the question of which process to tackle first and how to prove the benefit, it pays off — that’s exactly the difference between buying a tool and making an improvement. At PASSION4IT the principle applies: support rather than just advice, no slide decks, no tool selling.

Is that kind of consulting eligible for BAFA funding?

Yes, under the “Förderung von Unternehmensberatungen für KMU” programme. PASSION4IT is registered with BAFA (consultant number 222542); consulting costs up to a 3,500 euro assessment base per engagement are subsidised. The programme expires on 31 December 2026, and the application must be approved before consulting begins.

Further resources

• AI adoption in mid-sized companies without an IT department: what’s realistic? — the wider frame beyond accounting, including roles and governance.
• Which processes are suitable for automation? — the selection logic systematically, across all departments.

Want to know which process to tackle first? Book a conversation now.

Sources: Bitkom AI study 2026 (604 companies with 20+ employees, telephone survey) · Regulation (EU) 2026/1744 (Digital Omnibus Regulation on AI), in force since 27 July 2026 · AI Regulation Art. 4 and Art. 50 · BAFA, Förderung von Unternehmensberatungen für KMU. As of 3 August 2026.