← mhard.nl

The 20 pressing AI questions in business

These are the questions I see on boardroom tables every day — with extra weight on data sovereignty, local hosting and European regulation. With 20+ years of engineering, working daily with Claude, Gemini, Codex and Grok (xAI), I translate them into answers that run.

I. ROI & strategy

What will AI demonstrably deliver — and when?

I start with the business case: which processes cost time or money today, what the savings are and when you break even. Measured from week one.

Which ecosystem do we choose — Microsoft 365 Copilot or open source?

Switch on Copilot or build a flexible open-source layer? I benchmark both on your use case and prevent vendor lock-in.

Build, buy or wait for our software vendor?

Your own AI layer, or wait until AFAS, SAP or your PIM ships AI? I run the numbers on both scenarios — including lock-in and exit costs.

II. From prototype to production

Why do 80% of AI projects fail — and ours won't?

Pilots fail on data, integration and ownership, rarely on the model. I build from day one with evals, monitoring and an owner per process.

How does our ChatGPT/Claude prototype become a production system?

From sandbox to hundreds of users: LLMOps, scalability, monitoring and SLAs.

How do we safely connect AI to ERP and CRM?

Reliable APIs and RAG pipelines with live access to your existing systems — with permissions and logging.

Is our data infrastructure ready for AI?

Garbage in, garbage out. I help make folders, sheets and mailboxes AI-ready before we build.

How do we test whether the AI is good enough for customers?

Eval sets with real cases and a clear bar per release: only live when the numbers say so.

III. Sovereignty, data & EU regulation

How do we guarantee our data never leaves the EU?

Sovereign cloud guarantees and EU data centers — locked down technically and contractually.

Can we run AI fully on-premise or in a private cloud?

Llama or Mistral on your own servers, even without an internet connection. This is my specialty.

Does our AI application count as 'high-risk' under the EU AI Act?

I assess your system against the AI Act and set up compliance and transparency pragmatically — before the regulator asks.

How do we keep foreign governments (CLOUD Act) out of our data?

US cloud in the EU isn't enough. I arrange legal-technical shielding.

IV. Quality, costs & risks

Who is liable when the AI makes a critical mistake?

I design processes with human-in-the-loop and clear responsibility per decision.

How do we prevent hallucinations reaching customers?

Automated guardrails, evals and observability — measuring instead of hoping.

What does this cost us monthly (FinOps)?

Token budgets, caching and the right model per task keep costs predictable — accounted per use case.

Who owns the intellectual property of AI output?

Design, code or contract: I advise on protecting and licensing output.

V. People, organisation & adoption

How do we deal with 'shadow AI' on the work floor?

Employees already use free ChatGPT. I channel that into a safe business environment.

How do we get employees to actually use it?

Training, prompt skills and taking away fear — adoption is change management.

Can our team manage it themselves afterwards?

Documentation, runbooks and handover are part of the delivery. You don't depend on me — unless you want to.

How do we prevent our solution being outdated in two years?

Model-agnostic architecture: swap LLMs tomorrow without rebuilding.