Before AI Begins: Prerequisites, Data Foundation, and the Right Questions for Mid-Size Companies

Artificial intelligence is on the agenda of virtually every mid-size company across the DACH region. Rightly so: the technology has evolved from an expert domain into a practical working tool over the last two years. But between deciding to use AI and running a productive AI process lies a gap — and it rarely has anything to do with the technology itself.

According to an analysis on AI readiness in the Mittelstand, roughly 60 percent of AI initiatives fail not because of the chosen model or tool, but due to an inadequate data foundation. Before approving your first budget or commissioning a vendor, a few foundational questions are worth answering first.

The Questions to Ask Before Anything Else

• What is the concrete problem? Not AI for everything — a clearly bounded process with measurable effort and measurable output.
• Who in the company is internally accountable — not just enthusiastic, but decision-empowered and actively involved in the process?
• What data exists for this process — in what format, in what condition, and how accessible?
• Are you prepared to address data protection and GDPR requirements for AI systems, for instance when data is processed by external cloud providers?
• What happens when the AI makes a mistake? Is there a verification step, and who bears responsibility?

Data Foundation — The Success Factor Most Often Underestimated

AI systems learn from data and operate on data. For simpler applications like an internal FAQ bot or automated invoice processing, you do not need a perfect data foundation — but you do need an accessible and sufficiently consistent one. In practice, the same five bottlenecks appear repeatedly with mid-size companies:

• Completeness: If addresses, categories, or contact histories are missing from your customer database, even the best AI solution will produce incomplete results.
• Consistency: If the same values appear differently across systems — for example München, Muenchen, or MUC — that is a genuine obstacle for AI systems.
• Accessibility: If your key business data lives in PDFs, on paper, or in systems without an interface, the first step is not an AI project — it is a data integration project.
• Timeliness: An AI system working from a 2022 product catalogue or outdated customer records will produce outdated answers, regardless of the model.
• Governance: Who is permitted to use which data for AI purposes — internally and via external providers? Without clear rules, you risk GDPR violations and a loss of trust.

Three Typical Scenarios from Practice

Three starting situations we regularly encounter in IT consulting engagements illustrate that preparation often matters more than tool selection.

• Machinery manufacturer, maintenance logs: The company wants to use AI to analyze maintenance records — but discovers they exist partly on paper, partly as unstructured emails, and partly in the ERP. First step: standardize and digitize records. AI comes in step two.
• Trading company, customer service bot: Customer data is spread across three systems with no shared customer ID. An AI assistant would respond based on fragments. First step: master data management before AI budget is committed.
• Service firm, proposal drafting: Knowledge of what makes a good proposal exists only in the minds of two experienced employees — not in documents. First step: document and structure the core process. The AI needs this framework as a learning foundation.

Quick Check: How Data-Ready Is Your Company?

• Can you export the relevant data for your planned AI use case — completely, without manual follow-up work?
• Is your core data held in at least one system with an API or export function?
• Does someone in your company own responsibility for data quality — with defined standards such as required fields in the CRM or minimum documentation requirements?
• Do you know which data you are not permitted to share with external cloud services?
• Can you describe your target process clearly — when X happens, we do Y — and is this process stable and repeatable most of the time?

If you can answer all five questions with yes, you are well-positioned for a first AI pilot. If more than two answers are no, you will invest the next phase better in data integration and process documentation — that saves months on the AI project that follows. We are happy to help you find the right starting point: in a free initial conversation or via our AI assistant as a first step. Our reference projects show what this process looks like in practice.

The fastest AI tool does not win. The company that starts with the right data foundation reaches production fastest.

AI Solutions for Mid-Size Business