AI Investment for Business Leaders: What SMEs Really Need to Plan For

In almost every initial conversation with mid-market business leaders, the same question comes up: what do I need to plan for AI? The answer that most people do not expect: more than you think — and more importantly, differently than you think. Viewing AI investment solely through an IT lens consistently leads to underestimating what is actually ahead.

According to a study by Telekom MMS and mind digital (September 2026), 41 per cent of German companies now actively use AI — an increase of 17 per cent compared to the previous year (source: silicon-saxony.de / ROI-Kompass Telekom MMS, September 2026). The study's central finding: it is not the size of the AI budget that determines economic success, but how targeted the investment is and whether the prerequisites for productive use have been established. This guide outlines the five dimensions business leaders need to plan for consistently.

Why AI Investment Is More Than Software Procurement

Classic software investments follow a familiar pattern: get a quote, buy a licence, train the team, done. AI works differently — not because the technology is more complicated, but because it intervenes in processes and decisions, not just in tools. A word processor is neutral. An AI system that suggests quotes, evaluates documents or prioritises customer enquiries changes who decides what and when. That is an organisational investment — not software procurement.

• AI requires a strategic decision from leadership — not just an IT budget. Without a clear objective, pilot projects produce results from which nobody draws conclusions.
• Employees need to want AI, not just be able to use it. Those who underestimate change management observe after three months that well-introduced tools are barely being used.
• Data and compliance are prerequisites, not afterthoughts. Those who push both to the end of a project pay for it twice.

Overview of the Five Investment Dimensions

AI investment can be divided into five dimensions — with different weights depending on the starting point, company size and planned use case. The following table provides initial orientation for an SME with 20 to 100 employees introducing AI for the first time.

Investment Dimension 1: Technology and Licences

Licences for standard AI tools — Microsoft 365 Copilot, ChatGPT Business, Claude Team — cost between €15 and €25 per user per month. Not everyone in the company needs such a licence: in practice, active subscriptions for 30 to 50 per cent of the workforce are sufficient. Those who use AI not only as a chat tool but embed it into their own applications, customer portals or processes also pay for API consumption — between a few and several euro per user per month, depending on the model chosen.

Model selection as a cost factor

The most expensive AI model is rarely the right one. For most mid-market use cases — document processing, email assistance, meeting minutes — capable but more affordable models are fully sufficient. Those who plan AI integration and automation holistically decide on model choice, licensing and infrastructure together. The right fit can significantly reduce the ongoing operating costs of an AI solution.

Investment Dimension 2: Data and Data Preparation

AI is only as good as the data it can access. This sounds obvious — but it is the most common reason why AI pilot projects fail to deliver on their promises. When product data sits across five different spreadsheets, contracts are stored as scanned PDFs on an unmaintained network drive, and customer data diverges between CRM and ERP, no AI can meaningfully take over these tasks — however capable the model.

• Data preparation before project start: Before AI can access company documents, they must be structured, cleaned and checked for data protection compliance. Depending on the starting point, this takes 5 to 20 working days — a line item that is regularly missing from project budgets.
• Data quality as an ongoing task: Introducing AI without simultaneously addressing data quality does not solve the problem — it defers it. AI makes poor data more visible, not better.
• Data-protection-compliant structure: Personal data, customer information and internal metrics must be available in clearly defined access structures before AI works with them. Retrofitting is more demanding than careful preparation.

Investment Dimension 3: Skills and Change Management

Many AI implementations fail not because of the technology but because employees either do not want to use AI or do not know how to use it meaningfully. Both are management problems — not IT problems. The most important investment in this dimension is leadership time: those who introduce AI without forming their own concrete picture of what the technology can and cannot do are delegating a strategic decision to people who do not carry overall responsibility.

• Leadership first: Managing directors and department heads should try AI tools themselves — ideally through a guided introduction — before taking responsibility for the rollout across the organisation.
• Structured training rather than one-off onboarding: Introductory training works. What does not: a single workshop with no follow-up. Companies that measure after three months find that usage rates decline significantly without ongoing support.
• Appoint internal AI leads: At least one person should act as a permanent point of contact for AI — not as a full-time role, but as a clearly assigned responsibility with reporting to management.
• Scepticism as a signal: Employees who are critical of AI often protect processes that carry real weight. Their objections deserve to be heard — they often contain a legitimate concern that ultimately makes the implementation better.

Investment Dimension 4: Process Integration

AI delivers the most value when it is embedded directly into existing workflows — not as an additional tool that needs to be operated separately. This means technical connection to ERP, CRM and document management, as well as willingness to adapt processes themselves. The latter is often the more demanding part. Those who introduce AI without questioning processes frequently automate something that was not running optimally in the first place. Well-designed custom software and process automation avoids this pitfall by treating AI as part of the workflow from the outset.

• Pilot first, rollout second: Rather than introducing AI company-wide, start with a clearly bounded process — document classification, meeting minutes, email sorting. What works in the pilot can be rolled out. What does not work costs little in the pilot.
• Honest assessment of existing systems: The older and more proprietary the target systems, the more complex the integration. An honest technical assessment before project start prevents budget surprises.
• Process documentation as a prerequisite: AI can only be integrated where processes are documented. Companies without current process descriptions frequently invest first in documentation — and only then in automation.

Investment Dimension 5: Compliance and Governance

Since August 2026, core provisions of the EU AI Act apply to all companies that use AI — not only to providers. The competence obligation under Article 4 (ensuring sufficient AI knowledge among employees) has been in force since February 2025. For high-risk AI systems, the implementation deadline was 2 August 2026. For most mid-market use cases — text assistance, routine automation, internal knowledge bases — the risk classifications fall into the limited or minimal range: no burdensome conformity review, but clear internal documentation requirements (source: Sage.com: EU AI Act 2026 for the mid-market).

• Documentation obligation: Those who use AI must be able to demonstrate which systems are deployed, which data they process, and which decisions are supported or made by AI.
• GDPR review before deployment: Before an AI tool processes customer data, a legal assessment — and for more extensive processing, a Data Protection Impact Assessment — is required. Retrospective review is more expensive than careful preparation.
• Internal AI usage policy: Companies need clear rules about what may be shared via external AI services and what may not. This is not bureaucratic excess but basic protection for business and client confidentiality.
• Closer scrutiny for sensitive areas: AI in HR processes (recruitment, employee evaluation), creditworthiness assessment or safety-critical processes falls under the high-risk category and is subject to stricter requirements.

Realistic Investment Framework for the First Phase

From our projects with mid-sized companies, a reliable picture emerges for a focused AI entry. The following table distinguishes three typical scenarios — from a lean standard tool rollout to an integrated solution with custom process components.

These figures come from NoviCogi projects and are reference values — actual effort depends heavily on the current state of data, the complexity of existing systems, and internal readiness for change. Which scenario is realistic for your company can be established through a structured IT consulting process — before the project starts, not afterwards. An overview of specific projects and results achieved is available in our reference projects.

What You as a Business Leader Need to Personally Contribute

AI is not introduced into companies by IT — it is decided by leadership and lived by the workforce.

The most important factor in any AI project is not the tool, the model or the service provider — it is leadership's willingness to invest time and attention. Not as an operational matter, but as strategic oversight: setting goals, clarifying priorities, asking uncomfortable questions when pilots stall.

• Define goals: What should AI have concretely achieved in your company in 12 months? Without measurable goals, pilot projects become experiments without consequence.
• Approve and protect the budget: AI projects frequently fail not from lack of funding but because budgets are cut at the first delay. A realistic planning horizon of 18 to 24 months is essential.
• Demand outcome measurement: Those who do not measure KPIs after go-live do not know whether the investment had an effect. Simple metrics — hours saved, shorter throughput times — are sufficient to start with.
• Communicate openly: Employees who do not know why AI is being introduced develop their own explanations — which rarely reflect well. Clear, honest communication about goals and limitations of the technology is decisive for acceptance.

Concrete guidance on choosing the right partner is available in our guide to AI consulting for the mid-market in Munich — covering selection criteria and warning signs. The patterns described there align with what we observe repeatedly in our client projects.

Plan your SME AI rollout together

Frequently asked questions

As an SME business leader, do I need to invest in AI now?

There is no universal answer — it depends on whether your competitors are already using AI, whether you have processes suited for automation, and whether the necessary data foundation is in place. What can be said: those who start with a first, clearly bounded pilot in 2026 are structurally better positioned than those who wait until AI becomes standard in their market. According to Telekom MMS / mind digital, 41 per cent of German companies now actively use AI — the entry does not need to be large, but it should be structured.

What is the minimum budget for a meaningful AI entry?

A first structured AI entry for an SME with 20 to 50 employees is feasible from around €8,000 to €15,000 in the first year — for a clearly bounded standard tool rollout with guided adoption. Below that, a sustainable effect is rarely achievable: tools that introduce themselves tend to go unused. Those who want to integrate AI into their own processes should realistically budget €25,000 to €60,000 in the first year.

Do we need to sort out our data first before we can introduce AI?

Not necessarily all at once — but you do need to realistically assess which data the AI should access, and evaluate its quality and structure. For many mid-market use cases (text assistance, meeting minutes, simple document evaluation), the starting position is often sufficient. For data-intensive applications — forecasting models, customer segmentation, process control — data preparation is not optional but a prerequisite.

What obligations does the EU AI Act create for SMEs?

Since August 2026, transparency and documentation obligations apply to all companies using AI. The competence obligation (Article 4) has been in force since February 2025. For most mid-market use cases the risk classification is low — no burdensome conformity review, but an internal usage policy and documentation of deployed systems are required. AI in HR, credit decisions or safety-critical areas falls under the high-risk category and is subject to stricter requirements.

How long does it take for an AI investment to pay back?

This depends heavily on the use case. Automation of clearly defined routine tasks — document classification, email sorting, meeting minutes — typically pays back in 12 to 24 months. More complex integrations into core processes take 18 to 36 months. Projects without measurable goals and without outcome tracking rarely pay back in practice — not because AI does not work, but because nobody measures whether it does.

This article was created with AI assistance and editorially reviewed.