FAQ
Frequently asked questions on AI for Swiss SMEs
Over 30 questions, sorted by topic. What we get asked most. And what we answer.
First steps
Is my SME too small for AI?
From around 20 employees there are enough repetitive processes for at least two useful use cases. Under 10, a single tool is often enough.
How do I find out where AI would help us?
AI Potential Assessment: 1–2 days on-site, a written finding with 3–6 prioritised use cases by ROI × feasibility. Fixed price CHF 5,000–15,000.
What is the difference between an audit and a workshop?
Audit means: we deliver the finding. Workshop means: we facilitate, your team co-creates. Both at a fixed price.
Do I need my own data base?
For quick-win use cases, usually not. For forecasting or custom bots: yes, then data quality matters.
Where is AI most worthwhile in the operation?
Email triage, quote preparation, document workflows in office-heavy SMEs. Coding copilots in tech SMEs. Helpdesk automation in service companies.
Costs
What does AI consulting at Waldsee cost?
Day rate CHF 2,200 (half-day CHF 1,100). Audit fixed price CHF 5,000–15,000. Hardware (for on-prem) separate at the Lenovo Switzerland list price.
What does a typical SME implementation project cost in total?
Office automation for a 30-employee SME: about CHF 22,600 initial. On-prem for a 50-employee tech SME: about CHF 25,200 plus ThinkStation hardware. More in the cost guide.
Why CHF 2,200 / day?
Specialist boutique day rate. Above the generic AI consultant market (CHF 1,200–1,600), below Big 4 partner rates (CHF 4,000+).
Hidden costs?
No. VAT 8.1 % separate. Hardware separate. Tool licences separate. Nothing else.
Discount for prepayment?
No. Billed by effort (day rate) or fixed price for audit/workshop.
Tools and stack
Which tools do you use?
n8n as orchestrator, ChatGPT Enterprise / Claude / Copilot as the LLM layer, Ollama + Open-WebUI for on-prem, Power BI / Looker Studio / Metabase for dashboards.
ChatGPT Enterprise or Copilot?
M365 shops: Copilot. Google Workspace shops: ChatGPT Enterprise. For IP-sensitive data: on-prem instead of either.
What is n8n?
An open-source workflow tool, similar to Zapier or Make.com, but self-hostable. The default orchestrator at Waldsee, because the data stays controllable.
Which LLM models do you use on-prem?
Open-source models: Llama, Qwen, Mistral, the DeepSeek family. The concrete version choice happens in the architecture conversation.
What is Xinity?
A Vienna-based open-source orchestration layer (Apache 2.0) for on-prem LLMs. OpenAI-compatible. We integrate it on request for EU-sovereign setups.
On-prem and compliance
Do I need on-prem or is cloud AI enough?
Cloud is enough for many office workflows with a DPA contract. On-prem becomes relevant for sensitive data, IP-sensitive code, regulated industries or an internal rule that "data stays in-house".
What does "compliant with the revised FADP" mean?
The revised Data Protection Act (in force since 1 Sept 2023) requires a data-processing agreement with third-country providers, information for data subjects, clear purposes. Cloud AI with a US base meets this with a DPA. On-prem meets it structurally.
Which hardware do you use for on-prem?
Standard: Lenovo ThinkStation PGX with the NVIDIA GB10 Grace Blackwell superchip and 128 GB unified memory. Covers teams of 5–40 users.
What does an on-prem setup cost?
ThinkStation hardware at the Lenovo Switzerland list price plus 3–8 days of Waldsee effort at the day rate of CHF 2,200.
Can we swap models on-prem?
Yes, the stack is open-source-based. Model swaps are often hours, rarely weeks.
Process and maintenance
How long does an AI implementation take?
n8n email triage: 2–5 days. Custom GPT with RAG: 3–8 days. On-prem setup: 2–6 weeks including hardware delivery.
Who does the maintenance?
Three options: you yourself with guidance, us on demand at the day rate, or a monthly retainer.
What if a tool no longer fits later?
An open-source stack means: replaceable. Model, workflow, UI. All replaceable without lock-in.
Cancellation terms?
Audit/workshop free up to 14 days before, after that 50 %. Day rate: billed by effort.
People and culture
Will employees be replaced by AI?
Not by what we recommend. Augmentation before replacement. Employees become more productive, not redundant.
What is shadow AI?
Employees who already use AI. Privately, without governance. Real in almost every Swiss SME. More in the pillar guide.
What is the Swiss AI Paradox?
A Waldsee term for the observation that employee adoption is higher than management perception. Employees are further along than the strategy.
How do I train my team?
Modular training via the configurator. Half-day modules per department with a volume discount.
Methodology
What is the ATHENA framework?
Six phases: Awareness, Trust, Harmony, Empowerment, Nurture, Adaptation. A practical model for AI transformation. Detail under Methodology.
Is ATHENA scientifically validated?
No. A practical model from consulting practice, not a peer-reviewed research framework.
What is context engineering?
Building systems instead of prompt tricks. Data, connections, permissions, knowledge base. The context makes the difference, not the prompt.
What are the 5 kinds of prosperity?
Social, physical, financial, time, mental/spiritual. A filter for AI use cases: does the use case make the company richer along all five, or only one?
Question not covered?
Write to us. Or start with the next concrete step.